👉│feedback-general

[covert sun]

Hey, @void moon! Thanks for reporting. We have a ticket for this issue and are working on a fix.

[random pine]

Every-time I type I a. Prompt instruction for a collection after at least 10 questions the thread seems to deviate away from the prompt instruction and do its own thing structuring its answers in ways that I did not instruct it to do.. not sure if that could be fixed

[visual bough]

These are probably context issues, only the paid version offers higher context.

[random pine]

So it depends on the Question being asked?.

[keen relic]

Good morning, just a reminder to PAI team, you are changing lives and how we work and live, thanks for all you do, continue building your great piece of work PAI, there are so many creative ideas manifesting into reality here, such great thinking, PLEASE Don't Ever Try To Control the Creative Minds at work here.

[heady plover]

whenever i say "don't do something" the AI goes out of its way to do it

and whenever i do tell the AI to do something it sometimes just refuses to do it, example "edit this code to redirect the user to youtube", then it gives me the code but with "// Add your logic here", then i say "can you add the logic here", and then it goes "Sure! gives code without logic" and then i keep trying and it keeps doing that

[visual bough]

Please share some examples in the form of links.
Thanks in advance!

[heady plover]

will do the next time i encounter it

[kindred sinew]

Relatively new user here, seems very odd that choosing a model is in the settings and not part of the thread. Why is it a global setting?

[regal creek]

Have yall considered offering a mid tier plan. I’m currently on the two month pro trial. It’s a great service. Most days I could get away with the free version. I feel like a $10 mid tier would get you a lot more revenue. Just set a certain number of gpt4 uses etc. A lot of ppl are already paying for gpt plus and who knows what else.

[molten wasp]

I tried the image analysis experimental feature and love it. There is a bug where the icon at the bottom right will not change even though I rewrite it using another model Claude (the icon will still show GPT-4). This is not an issue for other "usual" searches. The icon will change to the model's name when I rewrite.

[odd dragon]

Dropped chatgpt plus for perplexity pro a while ago

[scenic oak]

Keep getting file upload failure, anyone else?

[visual bough]

What file?

[scenic oak]

Just an image

[visual bough]

What is the size?

[scenic oak]

4KB

[visual bough]

Has the upload of images worked in the past?

[scenic oak]

Yes

[visual bough]

Please create a #1140622008086970420 with more details about the issue.

[full cipher]

Problem: if I upload an image it results in a URL
https://pplx-res.cloudinary.com/image/upload/ ...
This image is accessible publicly (test incognito browser window). A user might think that with the data collection turned off, this also means that any upload is only visible to the logged in user.
I have to assume that this goes for any files?

[night palm]

https://www.perplexity.ai/search/I-want-you-8h5ojSJdSSuig3hLJgmZwQ?s=c

Why is perplexity so bad at searching for jobs? I see what search terms it uses on indeed and other sites (this is just one of many different threads I have opened about this), and it doesn’t yield accurate results. I want to search for part time online jobs that only require a high school diploma, but perplexity can’t lead me to any, and when I try to search for them on indeed manually, I don’t get any results either. So is it perplexity’s fault, or indeed’s fault, or both?

[summer wasp]

Is there a channel for discussing the API?

[crimson nymph]

There are 3 channels for talking about the API of perplexity
#1161804761247526912 #🧪│api-general #1161803596870979664

[pulsar apex]

EDIT: Fixed

Hello! There is a perceived bug with the web interface where the service is not respecting the url flag copilot=true for free users, because the page incorrectly points out 0 copilot usages left, but when you click the copilot clock it shows the pricing page and then the counter reloads.

How to replicate:
Try to start a search via a url using a free account (ex: main search engine on browser): https://www.perplexity.ai/search?q=List the best beaches in the world&focus=internet&copilot=true

[edgy skiff]

Hi, is it possible to export my data from the web app?

Question 2: When using Claude for my answers (in general mode, not Writing), sometimes all I get is a blank answer - just sources carousel and below that immediately the control panel with the Rewrite option, info about the model used etc.
(Important: I strongly prefer Claude)

[whole fog]

I'm also seeing issues with the copilot=true query param. I have a pro account, but when I open a query via URL with copilot=true in the URL it doesn't seem to run the query in copilot mode. I also see 0 copilot uses left at the bottom of the screen.

However, if I then hit ⌘K and turn copilot mode on, it shows that I have plenty of copilot uses left for the day.

I use Perplexity almost exclusively through a Raycast extension, so this is pretty painful at the moment. https://www.raycast.com/third774/perplexity

[pulsar apex]

It happened very recently, right? I have a bookmarklet that opens the selected text as a perplexity tab and this completely changed my knowledge flow. Hopefully it's an easy fix.

[quick meadow]

Hi team. I love perplexity and am about to post a short YouTube review of the free version. I was curious if you have any type of influencer or affiliate programs?

[odd cosmos]

I've exposed the same problem two days ago, when it started, and @crimson nymph wrote me it's a bug and the developers are working to solve it.

[pulsar apex]

Great to know!

[stray talon]

Heya, it seems perplexity bugged for some reason

https://www.perplexity.ai/search?q=what+is+the+history+and+origin+of+the+name+"Damien"&copilot=false

Keeps repeating

[visual bough]

Can you please share the thread?

[stray talon]

That's the whole thread

Oh wait it's not cached sorry

https://www.perplexity.ai/search/what-is-the-gApcXP4XSTqT89rHBby6aQ

[visual bough]

And please make the thread public

[stray talon]

Done

[visual bough]

Oh yeah I see 😄

[stray talon]

Seems like the default is non-public eh
Been a long time since I've shared links

I remember this happening quite a lot in other AI but I don't remember whether it's bard or bing

Ah, I was wondering why there are so few channels. I didn't have all channels shown.

[pulsar apex]

Must have been a great painter though.

[pulsar apex]

Ping.

[silver trellis]

Hello,
I am experiencing a problem with all the pro models when I provide an attachment and ask to make an article or essay that would fit a 2 pages of word document. I will usually get content that would fit a half of a page maximum. Would appreciate any help or advice on this matter.

[vast eagle]

First of all, I am not a paid user. Perplexity WAS my favorite search engine, but it's performance degraded so much recently. 7 out of 10 answers are useless now.

here is a comparison between perplexity and phind on a same search:

[pulsar apex]

Do you have access to Copilot? That search didn't use it, could be related to the problem we pointed out above.

[hot cargo]

I'm a pro user and didn't even know this feature existed. Will try today. Thanks.

Hope you get your answer.

[covert sun]

Hey @silver trellis! I think this thread might help you, please take a look: https://discord.com/channels/1047197230748151888/1176526177050054766

[vast eagle]

No, I am not talking about the copilot, the "normal" search performance downgraded significantly.

[silver trellis]

Thanks so much. I will read the thread and try the prompts for a longer texts.

[covert sun]

Hey, @vast eagle! Thanks for your feedback. Does the short answer example (and some variance in answers is possible), show that the performance degraded? Is it possible that even the short answer based on apple support sources, could be the right thing to fix the error?
With Pro all the models are available to you, and if one suggested solution didn't work, there's always an option to rewrite it or see other models' answers. I got another answer with more ideas on the issue with Claude https://www.perplexity.ai/search/ipad-the-time-U5N9k83iTomqTPYDPiQrvQ?s=c
If you have more examples with inaccurate answers or errors, feel free to share them in the #1140622008086970420 channel with permalinks, we'll surely take a look.

[thin ginkgo]

hello.. is it possible to change an account's email address? also are there any plans to go the username/password/2FA route in lieu of email prompts at some point?

[covert sun]

Hey, @thin ginkgo, changing account's email address is not available. https://blog.perplexity.ai/faq/log-in-account. As for the auth method changes, that might be implemented in the future.

[vast eagle]

Thank you for your response. Actually, I prefer concise answer than lengthy ones. But concise is not just short. I didn't make that judgment based on a single case. If you're interested, I will provide you more comparisons.

[mossy ridge]

Hello hello, dear developers! Just today notices a bug that frustrates now every entry. When starting typing long question or prompt and accidentally pressing down arrow it switches to previous prompts like "summarize" and never returns back to the prompt you started to type. And after that you have to do all the work again...

[thorn tapir]

yo!
just wanted to point out how frequently i upload the context from a file and then it asks me to copy and paste in specific functions(instead of just reading the file). Not sure why this happens, but it wouldn't be a big deal of the context window for that wasn't so small.
tiny problem, but would appreciate a fix. thanks!

[zinc oxide]

Would love to see a way for perplexity to use my location if I ask "Food near me" or something similar

[barren scarab]

If your location neighborhood doesn't change a lot, you can try adding it to your AI profile

[covert sun]

Hey, @thorn tapir! Are you referring to Copilot additional questions with uploaded files? Have you tried skipping such questions, the model should read it from the uploaded files. Can you DM me some examples or forward them to [e-mail]

[thorn tapir]

I haven't tried without colab - will do.

[thorn tapir]

also, I am just confirming so I don't blow a bunch of money, is labs free(or no additional cost once on pro plan?)

[covert sun]

Have you tried setting the location in your Profile? That should do it for the example you brought.

[covert sun]

Sure, labs is free, no matter if you have or don't have our Pro subscription.

[thorn tapir]

Sweet

[slender summit]

Hello, any timeline for Zapier integration?

[severe creek]

@covert sun

can team, can you look into this on priority? i'm unable to add card details

[visual bough]

Hey @severe creek!
Many users have important requests, if you have sent an email to [e-mail], it will be taken care of as soon as possible.
Thank you for your patience! ☺️

[flat anvil]

https://www.perplexity.ai/search/Kannst-du-bitte-LYNny37SSFO0XaHQCR_FIA?s=c

I asked for a German to Persian translation (no meat, vegetarian!) And the bot gave me just German answers, even if I said it twice.
I had to say it three times to get it finally.

[visual bough]

It works better if you explicitly request a translation.
https://www.perplexity.ai/search/bersetze-auf-Persisch-c1iVf6UERLWUXS5F5uV.cQ?s=c

[woeful sparrow]

Hey everyone, jsut wanted to give a quick feedback: a few weeks ago I saw Perplexity performance going down, but it seems it is WAY better now. I get quick, accurate answers on most topics, very happy and thanks for the progress @covert sun !

[burnt coyote]

I feel like Perplexity has a tendency to "double down" on its initial assertions. If I find the initial response unsatisfactory, I have to create a new thread to change its mind. Forever, here, I am trying to tell the AI behind the scenes that the article it found wasn't what I was looking for, and it basically said, "nope, not gonna change my answer."

https://www.perplexity.ai/search/I-recall-a-3PBFJVqhTUKRKO4GD_7rRQ?s=c

[fiery galleon]

asking follow ups is sometimes nearly useless. you have to basically copy and paste what you said earlier. it will say things like needing more context or it doesn't know how to answer. i'm on pro with gpt4 as the model. it seems like it specifically doesn't have what was said and it will only be a convo of three messages total, so it can't handle more than one message, maybe two

this is a great example

https://www.perplexity.ai/search/help-me-decide-1t_cVm5iQI2WDMrgixLnwg?s=c

literally only two messages and it just goes "what are you talking about"

makes the concept of 'follow up questions' show to actually be basically a lie since it doesn't work on even extremely basic examples

[unkempt atlas]

A comments section on searches shared on the discovery tab

[muted island]

I would like the ability to instruct the AI to respond in the language used in the prompt.

It does seem to do this by default when Profile is disabled, but enabling it forces the AI to use the language specified in the Which language should the AI use? field.

i.e. Ability to clear the Which language should the AI use? field on profiles?

[gray pumice]

https://www.perplexity.ai/search/find-me-some-MTPT4JD_QIutqR6yWuB3SQ?s=u

curses

[crimson nymph]

please make your thread public by clicking on the share button in the top right

[full cipher]

low prio, this is a localization feedback though not sure how this could / should be implemented

what is the best way to make a frozen pizza at home? i live in germany
https://www.perplexity.ai/search/what-is-the-WdgklT7GTg6pPCzWI..7YQ?s=u
if i live here, there are certain things that come with the localization - f.e. i dont know what 500°F is, we use celsius. same if a time was referred, my locale uses 24h versus am and pm.

if I provided you with the information through the ai profile or in this case in the search quest, it could answer in a subtle way.
similar goes with other references to sizes (think baking recipe) or values in other currencies.

[gray pumice]

https://www.perplexity.ai/search/how-to-reset-S0nYI_PrRzq6079PWXhn2A?s=c

apparently this example uses var which is outdated and dangerous

Never use var. It is outdated and dangerous. Instead, use const or let if you have to reassign.

hope you guys will update accordingly

[vast jetty]

Would like to suggest adding "dev" as an additional "focus" option -- goal would be to index and search across places like GitHub + Stackoverflow

[sweet kelp]

I don't know who to speak to, but I truly believe I have a much better pricing model for Perplexity that will make them more money.

The way they are doing it could onboard way more people if they had an intermediary option between 20/months and nothing.

[barren scarab]

If it's something like 10 dollars with fewer queries, I think it's been suggested to the team multiple times before.

Not sure if your idea is the same.

[sweet kelp]

I say this because almost exclusively bc GPT-4 is $20 and many who would get perplexity already are spending this, it's a significant psychological leap to spend $40 per month for 2 gpt 4s

Instead a running dynamic pricing system let x queries would be sensible

[dry belfry]

I have set in my AI profile that perplexity should emphasize the important terms by bolding them. However, my issue is that the "weight" is currently very low, making it difficult to differentiate between bolded words and non-bolded ones.

[vague prawn]

hey , can you add the model selection option below the search window

[woeful marten]

the color scheme is a little vintage and similar to DDG

[grand holly]

Can we have option to delete responses from the middle of the thread too without having to delete the last response first?

[south ether]

Not possible at the moment, but thanks for pointing it out. I'll be sure to bring it up to the team. 🙂

[copper rune]

I've been a user of Perplexity AI and appreciate the incredible strides you've made in AI-driven conversations. However, I'd like to share some constructive feedback regarding content requests and the potential impact on users.

Content Guidelines:

• Avoid Explicit Visuals:

  • Refrain from complying with requests for explicit content, including images or videos of women in revealing attire (low necklines, sleeveless tops, short skirts, tight dresses, etc.).
  • Such content may contribute to negative consequences, impacting users' perceptions and experiences.

• Mitigating Harmful Effects:

  • Sharing explicit content or links to pornographic sites may lead to desensitization, relationship issues, and difficulties in real-life intimacy.
  • Encourage responsible content consumption and consider the broader impact on users' well-being.

• Common Sense Approach:

  • Focus on a common-sense approach rather than adhering to societal trends that may not recognize distinctions between genders.
  • Promote ethical content that respects the dignity and values of all users.

• Perception in Relationships:

  • Displaying explicit content can potentially harm how men perceive women in dating, marriage, or relationships.
  • Consider the broader societal impact and strive to maintain a positive and respectful user experience.

Conclusion:
I believe implementing these guidelines will contribute to a more responsible and user-friendly Perplexity AI. By upholding ethical content standards, we can create an environment that fosters positive interactions and aligns with the values of a diverse user base.

[grand holly]

Lmfao? Are you trolling?

[copper rune]

Nope. This is the feedback page.

[grand holly]

You want a special safe search switch with a spoiler alert? As if the ai isn't restricted enough already.

[dreamy verge]

They way I'm using Perplexity, it would be better for me if the threads would disappear from my history UNLESS I decide to save them in some collection. Alternatively, it would fine to let them pile up but have a quick way of getting rid of all threads that have NOT been saved in some collection (or marked with a star, or with a heart, or whatever).

[hearty compass]

Has anyone else had problems with pasting text into the search window recently? It happened few times to me, had to restart the browser.

[covert sun]

There was an issue with pasting text, but it should be fixed now - #🌐│feedback-web message
Please check, if the issue you faced was the same as reported above that message.

[full kite]

If I pay the monthly subscription the discount of HOLIDAYS23 is half (20/40). Should I assume that the next billing month they will automatically deduct the other half?

[kind nymph]

Looks to me that they wont charge you for 2 months

[full kite]

Nice. Sorry, this line doesn't appear in my browser

[kind nymph]

What do you mean?

[full kite]

or I just didn't see it

[oak mauve]

Please add a widget to the android app so that we can tap the widget on the android homescreen and instantly enter a query.

[keen relic]

Lately I'm seeing PAI Pro loosing the context of the discussion, unless I tell it to review the thread.

Sometimes it's so bad, it just repeats what we just discussed, a regurgitation of same, even when adding context to the history of a previous response

if I thumbsdown this and make a comment will it get analyzed?

It seems to be loosing key contexts that it should probably retain for a thread. e.g., having a long discussion of running virtualbox on windows 10, then giving a response that is invalid because the response was for linux os not windows 10 os.

[cedar vector]

Your pplx-7b-online doesn't respond well in other language. Its answers do not match to my questions. By the way my language is Tagalog.

[elfin bone]

Please make it possible to edit queries.

[grand holly]

Bro what?

[grand holly]

Sounds like you want something similar to incognito mode.

[grand holly]

I have this weird problem where perplexity gets stuck on responding answers to older questions in a loop instead of responding to the current/next question.

[grand holly]

Ah. I see.

[dreamy verge]

Maybe, maybe not. I think a button to flush all threads that have not been favorited/put in some collection would be enough.

I just don’t like having a lot of junk lying around

[grand holly]

Auto deleting inactive threads could be an option. With customizable expire duration.

[keen relic]

Guys, something is really wrong right now, it is completely loosing the context of discussion and repeats it's previous response.

[covert sun]

Hey, @keen relic! Does it happen with Copilot or any queries? One specific model or all of them? Could you please share the thread/threads in the #1140622008086970420.

[mellow trail]

Is there a way to save the discover threads into my collections?

For an example, I want to follow up on one of the web searches (like JWST captures warped supernova) when I have more time.

I see copy link/share but nothing to save it inside my account.

[rocky timber]

Happened to me earlier today too

[keen relic]

three times in a row, same response different user inputs.

[covert sun]

Hey, @mellow trail! You can create a collection and save your follow-ups there. Then you'll have the original Discover threads as references in your threads, too.

[grand holly]

Yes. This needs more attention.

[mellow trail]

That works well, thank you. I think PP should consider evolving Library into basically an AI bookmark system from anywhere.

It would be nice if we could bookmark on discovers as well as any individual responses from PP (not just the whole thread) into collections as well.

It would add more value to the service if we can use PP as a bookmark/note-taking service as well. Otherwise, it would be difficult to compete against other AI services for the same price (Poe, OpenAI, etc).

[dawn rover]

unable to use api with third party webchat like typingmind.com ; problem with CORS :
ERROR MESSAGE :
Access to fetch at 'https://api.perplexity.ai/chat/completions' from origin 'https://www.typingmind.com' has been blocked by CORS policy: Response to preflight request doesn't pass access control check: No 'Access-Control-Allow-Origin' header is present on the requested resource. If an opaque response serves your needs, set the request's mode to 'no-cors' to fetch the resource with CORS disabled.
Can you fix it ?

[heady plover]

I created a discord account just for this reason. This LaTeX bug is super frustrating and very annoying. Perplexity has a lot of things that it does better than OpenAI but as long as this latex bug exists it is almost unusable for many purposes that have nothing to do with pure writing. is there a time frame when this will be fixed? i saw that this was already mentioned in March. is there any news?

[heady plover]

I noticed there's no real button to re-run a prompt, only 'rewrite' which just takes the same text and rewrites it. If I want to re-run a prompt, I have to copy it, delete the previous run, and run it again. This is unnecessarily complicated.

Also, when you have a longer thread in the app and want to keep working in it, you have to swipe all the way to the bottom without a button that makes the text jump there immediately. Serious flaw imo.

[wet phoenix]

better than chatgpt

[rapid snow]

Would love to have a model switcher without going into settings

[visual bough]

This is a very popular request and has been noted! ☺️

[rapid snow]

Even better is to have diff models automatically (default) plugged in (with an option to switch then there if needed)

[mild prawn]

love the citations, 20$ a month well spent!

[empty nest]

I have iPhone 14 Pro Max using perplexity, in settings I choose Urdu language, when I search anything results shows me in Urdu language but every words shows separate alphabet like example, A P P L E It’s wrong, right way look like this apple

[final warren]

if you ask “unidbg use arm arch debug so demo”，，it‘s dont hava any response

[night shell]

This happens very often.

[crimson nymph]

You'll get this error when the AI generated some nsfw content. Try to adjust the prompt and try again.

[grave cobalt]

Can get the option to set a default AI model per collection?
The perplexity model is great for general search, but the anthropic model is great for coding queries.

[covert sun]

Hey, @grave cobalt, thanks for the suggestion! We've got this request before, the team will consider it, too, in the future updates.

[blissful kettle]

Can you add a option that the threads are not automatically saved. It is really hard to organize everything.

[unreal jackal]

I am genuinely interested because I don't have programming knowledge myself. Is it more complicated than it seems to implement the concept of how Latex is represented in OpenAI for Perplexity, given that this issue seems to exist since March? I'm truly sorry if I'm being bothersome with this. However, I'm trying to learn with Perplexity for my exams in February. It makes it much more challenging to recognize things when they are not displayed correctly. I had canceled ChatGPT Plus to use Perplexity Pro, but I wasn't aware of the Latex error.🙏🙏 @covert sun @visual bough

[heady plover]

I just tested opting out of data usage, then asking for book recommendations in a new thread. Then I deleted the chat and my complete history, for good measure. I opened another thread and used the same prompt.

It gave the exact same list. Why are you still using my data when I opted out of it?

[also submitted a bug report on this]

[visual bough]

Depending on the setting, GPT models also generate almost the same response to the same prompt, this has nothing to do with data usage, but simply that the model thinks in the same way and outputs the same response 🙂

[visual bough]

Hey @unreal jackal,
The team is still working on a solution, in the meantime please use something like this in your AI profiles:
IMORTANT: Always use $$ before and after a latex formula, also when showing a formula in a line between text.

[heady plover]

Thanks for your response 🙏It's just curious to me that Claude doesn't do it when I use it on Anthropic's website, and it also didn't happen while using Claude on the Poe app

(Oh yes, I was using Claude, forgot to mention that before)

[visual bough]

Yes, the platforms on which you may have used Claude use different settings, please take a look here, you can see that there are many parameters that can influence the answer: https://docs.anthropic.com/claude/reference/complete_post

[unreal jackal]

Alright, thanks

[grand holly]

How about a 1 click button to clear the prompts-responses within a thread? I know I can just delete and create new thread but I don't want to have to re-upload document on this new thread. Also, sometimes, the past prompt responses cause ai to glitch more.

[clear compass]

I think it would be better to keep edit and delete about under the input box? everytime need to go down and search these two

[rancid tide]

Follow-up questions have been bad the entire time I’ve used perplexity over the course of the year. ChatGPT always keeps the context really well; with perplexity is generally need to start each thread fresh instead of having an actual conversation.

[void moon]

Fix bug plz!

Writing mode often switch to online search mode

[visual bough]

Hey @void moon,
A fix is being worked on, thanks for your patience ☺️

[void moon]

Also, the content generated by the Gemini Pro model is incomplete, and the same situation occurs when regenerated.

Often times.

[covert sun]

Hey, @void moon! If some generated content is incomplete, you can go on with the 'continue' prompt.

[covert sun]

Hey, @rancid tide, thanks for the feedback. Were you using the "Writing" or "All" (some other?) mode? Could you share some examples where you saw it not following you, in the #1140622008086970420. We'll take a look.

[covert sun]

Hey, @clear compass, thanks for the feedback! Currently, these buttons are at the bottom of the answer, so you could make a decision about editing or deleting the answer when you finish reading it. If you hit "edit", you'll be forwarded to the query field that you can edit. Also, the delete button is in the sub-menu so users don't accidentally delete answers they didn't intend to.

[rancid tide]

It’s probably the All mode. I’ll share an example next time I see it.

[marsh falcon]

One thing that I think would make the service even better would be the ability to add a list of sources for a collection. So either an upload of PDFs for the AI to reference, or a list of web pages. This would make the collection much more useful for specific uses.

[plain tangle]

I'm not sure where to share positive feedback, so I'll post here.

I want to sincerely thank Perplexity for making my search experience way better. Over the last year or two, I've gotten extremely frustrated over my Google search results, and I feel like Perplexity gives me back a bit of sanity. It's still not perfect, the search results Perplexity itself gets are flawed too, but it saves me a ton of time, and I'm very happy with the constant improvements you guys make.

Thank you.

[pure vigil]

I agree. Perplexity gets confused often on simple follow-ups. This is why I’m not paying for premium.

[void moon]

I know this method, but once it starts generating incomplete content, if I input 'continue' to prompt it to output more, then the next content is still incomplete, I can't possibly keep inputting 'continue', right? It's like I'm coaxing it.

[silver forge]

Guys I am having troubles with the rendering of math formulas

the AI writes the KaTeX syntax but doesn't render the math formulas properly

What about You ?

[covert sun]

Hey, @silver forge! The team is aware of this issue. We have a ticket for this and will fix it.

[silver forge]

Fantastic. This is the only issue encountered by Me. I use Perplexity both on Desktop & iPhone.

I decided to subscribe to PRO version, leaving GPT plus 😎

[chilly yew]

Thinking of ever integrating code interpreter? I feel like that is the only downside compared to GPT plus.

[grand holly]

Does this apply to the free model too?

[unreal jackal]

What does the code interpreter actually do that perplexity cannot do yet? I didn't quite understand it when I still had gptplus.

[chilly yew]

Let's say you want to compare two programs to find primes, you can tell the AI to write the scripts, and then run them while timing them, and tell you the results

It makes the AI able to run code

It can also generate graphs, modify databases

[night shell]

Thanks, but that wasn't the cause (or at least, the prompt was not nsfw).

[visual bough]

Can you share the thread with the error, was it just temporary?

[night shell]

It happens regularly; if it does so again, I can share the thread.

[wild saffron]

Hey guys! Love perplexity. However, the responses containing math are hard to read as latex equations aren't rendered. It would be super useful to have!

[kind nymph]

improvement to the login system

I assume the perplexity team decided to use email only for specific reasons but I would rather to have the option to use passwords as it's inconvenient to log in sometimes especially that there is the app and the browser version and both of them use the same way of signing in, maybe if there was a code you got on email rather than a link that would of been a bit easier but I am not sure about the security implications of that

[supple elk]

Can AI generate a realistic audio dialogue?

[glad raven]

It would be very nice if share links had open graph tags in them.

Eg, this should show an embed in discord with what was searched and maybe a short summary. I end up just taking screenshots and sharing those mainly because there is no embed.

https://www.perplexity.ai/search/1050e90b-d977-4f9f-96c2-d3411acf12bf

[covert sun]

Hey, @supple elk! There is no text-to-speech option available on the apps or the web for now, so an audio dialogue is not possible. Could you provide more details & examples about the implementation you have in mind.

[supple elk]

I beg to differ. Text to speech is all over the web with different languages genders accents and emotional tone all of which makes it extremely realistic, indistinguishable from a real human. Now AI, for its part, can easily create a dialogue in a very natural language with idiomatic expressions, ocean and tone of voice. It can even ajust the spelling and vocabulary to suggest a particular dialect. I can't imagine that putting the two together with sound effects images to boot would be such a challenge. Of course I can say that with great confidence, having no technical knowledge whatsoever myself😂

• emotion not ocean

Hey wait a minute! Language apps use AI to degenerate speech already! How did they do it?

And why can't I?

[rich birch]

Would be amazing to be able to crop pictures in the mobile app for queries! It's hard sometimes to get the right angle or photo as it tends to interpret the entire photo content, rather than the very specific portion that I need help with. 🤞

[scarlet sage]

Would love if GPT-4 was GPT-4 and not 3.5

[visual bough]

Hey @scarlet sage!
GPT-4 on Perplexity is the real GPT-4 from OpenAI! ☺️

[scarlet sage]

it also doesn't provide code like GPT-4 does

[visual bough]

This is because GPT-4 was of course not yet published when GPT-4 was trained and GPT-4 therefore cannot know anything about it, unlike ChatGPT, Perplexity does not manipulate the system prompt so that the desired model name is called, just because the AI claims something about itself does not mean that it is true, usually it is a manipulation in the system prompt 😉

[visual bough]

What exactly do you mean?

[scarlet sage]

Alright thanks 😄

[hardy harbor]

After joining Pro I clicked the link that says "Pro Discord" and it told me I'm already a member of this Discord.
Made me feel a bit silly. Could have been labeled more accurately perhaps.

[visual bough]

Please leave the server and rejoin with the Pro link.

[radiant kindle]

I asked Perplexity to summarize a YouTube video, and it says it can't. Then I asked it to answer the question posed in the video title. Instead of turning to the rest of the internet for answers, it simply kept saying it can't summarize a Youtube video, and I had to start a clean thread to get my answers. https://www.perplexity.ai/search/summarize-httpswwwyoutube-1P8aysiWTUChNTYuNZ36dg?s=c

[pastel solstice]

Is there a way to enable metric values by default ? I get results in inches and pounds, which is not useful and I think there should be a global setting for that. Right now I have to specify it in the prompt, which feels unnecessary. I am new here so maybe i overlooked something

Also math symbols are not displayed properly. I asked in the app about a paper and it couldn't display the symbols

[visual bough]

You can write both things in your profile, for math symbols: #👉│feedback-general message

[keen relic]

I am using a different AI engine on this one, not the Default engine, so same thing happening on different AI engines tells me it's Perplexity issue.

[dreamy verge]

Since MS copilot iOS app just came out, I tested its web search capabilities with this question: what is the most interesting upcoming tennis match? (Right before it answered correctly to the next Series A fixture.)
Result: hallucination 😂 (it would be indeed interesting if that match would happen)

Same thing on perplexity: a copilot question (men or women?) and then I think the right answer: no “big” matches planned so far. 💪

[nocturne valley]

@covert sun Do you guys have plans on adding GPT-4 Turbo, it seems way better at coding in my use case.

@south ether

[rich birch]

Hey Alex, do your team have an estimated time to fix this? Just asking

[grand holly]

Image steered away from context.

[visual bough]

#👑│pro-lounge message

[hardy stratus]

Improving the citation precision

Hey guys, what are your reflections on the outcomes of the study? https://arxiv.org/abs/2304.09848v2

Perplexity has scored 72,7% on citation precision (meaning that every 1 of 4 Perplexity-generated citations do not support their associated statements, i.e. every 4th statement is basically hallucinated) may be "not great, not terrible". The figure roughly reflects my own experience, and I am a paid user alternating between GPT-4 and Claude 2.1, which seem largely the best for now...

But I am a very diligent user - I never take any AI-generated outputs at face value and I always verify everything for any of my higher-stakes research use cases, regardless of the GenAI system I use.

That might not be the typical behaviour of a typical user, though, who may rely on the outputs without verification. They shouldn't, of course, but there is not enough awareness around this.

Surely, there is a lot of work to be done to improve the citation accuracy in Perplexity while still keeping a good fluency and perceived utility scores.

What are your plans in that direction in 2024?

[bold glade]

plz render inline latex properly

[visual bough]

#👉│feedback-general message

[bold glade]

Works like a charm. In context learning is awesome!

[quick abyss]

Am sure you've considered but adding a searchable memory feature at the account level would be amazyong.

• FWIW Keymate.AI offers an integrated vector knowledge base tool via its API, plug-in and website.
• Lets you quickly “upsert” a chat response or entire convo to a db, which can then be single searched with web and PDF URL's.
• No idea if good but their web search is not.

[silver forge]

Yeah, the only critique

And actually I am studying the same stuff hahaha

[timid vale]

There should be a channe lfor API usage

[visual bough]

👀

[timid vale]

thank you!

[glacial granite]

Hey @timid vale, if you find the original message helpful, please consider reacting to it with the ⭐ emoji. If the post is appreciated by the community and receives 5 stars, it will go to the ⁠⭐│starred channel and the post author will get the EXPLORER role on Perplexity.

[timid vale]

it wasn't toggled for some rason hehe

[scarlet totem]

I love perplexity for explaining notation but I wish you'd render the latex output it generates.

[visual bough]

#👉│feedback-general message

[north bough]

Not sure if feedback about perplexity labs is appropriate here but is there any plan to host the microsoft phi-2 model?

[covert sun]

Hey, @north bough! We have a separate channel for the #🦙│feedback-playground, but thanks for your suggestion. All the models suggested by users will be considered by the team.

[north bough]

Oh okay thank you!

[vapid mason]

bug report: if you duplicate the perpelxity tab so you have 2 tabs - the second one does not work (sits forever without showing the answer, but does show the answer if you refresh the page)

[safe fulcrum]

Post to #1140622008086970420

search first to see if a report already exists

[knotty minnow]

tbf the actual prompt you used, "the next most interesting tennis match", is wildly ambiguous/subjective – I'm not surprised it produced some wild responses!

I thought revising the prompt to be slightly more precise, "Next major tennis tournaments?" (assuming that is basically what your query intends to have answered), would help steer both in the right direction. Funnily enough, Bing (browser) returned a useful/accurate response (as far as I can tell), but perplexity was thrown off (response begins by saying the 2024 Australian Open just finished, even though it hasn't even started yet..

But...When given what I think is a far clearer prompt – "Upcoming major tennis tournaments schedule" – (again, assuming I've correctly understood what it is you were searching for), Perplexity's response blows Bing out of the water

[fair pilot]

Its not very clear what exactly I'd get and how I would benefit by subscribing for the $20, so I greatly suggest making it clearer and obvious, and I am sure more people would sub. I am sure it's worth it cause perplexity is awesome,but need more clarity on what exactly we get in addition to the free tier.

[visual bough]

All information is described in detail at https://blog.perplexity.ai/faq/what-is-perplexity-pro

[magic moss]

Hello

Is there a way to get refund

I deleted my account and it is still taking money from my account

[crimson nymph]

please send an email to [e-mail]

[vapid mason]

Security problem / Feature-Request: please do NOT add the CR/LF on the end of code snippets which are shell commands (powershell / dos / bash / etc) - this is causing the commands to be executed immediately when pasted into sessions, without letting us review the safety of what is about to be done (and keep in mind that your web UI hides most of these commands with a sideways-scrolling text box too) - for example - I dare you to paste this suggestion into powershell: https://www.perplexity.ai/search/powershell-command-to-4WV41_H9TROKoNvMMVpc6g?s=u

[covert sun]

Hey, @vapid mason, could you make the thread public by pressing "share" in the top right corner of the thread.

[vapid mason]

sorry - done. Horrifying link though - so please don't take me up on my dare !!!

[heady plover]

Are you, by any chance, planning to introduce the option to fine-tune model parameters on our end? It would be very nice to set parameters like temperature, top P, top K, repetition penalty, etc., for ourselves, based on our own needs for how the respective LLM should function.

see for example this set of adjustable preferences from koboldcpp

[visual bough]

#1190267787306881064 message 👀

[heady plover]

Thanks for your reply but I don't have access to that channel or was it not directed to me?

[visual bough]

Pro channel, sorry:
I can't guarantee that it will become part of the main service. We've got similar requests to put all the parameters together in an expert mode, the team is aware of this.

Currently, such settings are available for developers using our API models.
https://blog.perplexity.ai/technical-faq/can-i-control-parameters-like-temperature-top-p-and-stop-tokens

[vapid mason]

Bug Report: annoying UI wipes out all our typing and redirects to https://www.perplexity.ai/?focus=internet from time to time (I use perplexity maybe a dozen times a day, and this happens at least every day)

[noble monolith]

https://discord.com/channels/1047197230748151888/1140622008086970420

[oak mauve]

Still not possible to increase font size of the AI replies in the android app. Please add functionality ASAP.

[wanton forge]

any plans to improve the AI replies for the german language? Or is there a timeline one can expect better answers?

[tender grotto]

I would love to have the Perplexity AI API show sources that it retrieved, just like you have in the native web/mobile app

[noble monolith]

https://discord.com/channels/1047197230748151888/1047649619695390740

[noble monolith]

did you switch to German in Profile?

[crimson nymph]

It's highly requested but it's not on the roadmap for now https://docs.perplexity.ai/discuss/657897d232e8e2006bfc4f29

[tender grotto]

That's a shame, showing sources greatly increases the trust users will have in the information I show to my app users. It is the one thing that makes Perplexity stand out from ChatGPT, Bard, etc. Hopefully one day it will happen

[woeful sparrow]

A quick feedback on a frequent flaw I see done by perplexity. Often, when you ask a question and the engine finds some articles which is a "top of.. " or something similar which gives a list, the agent kind of stop looking/go lazy mode and gives you only answers for that source.

[rocky timber]

I've been really struggling recently to get images generated that match the prompt. I was having great success generating images, with complex subjects and detailed prompts, that really met my expectations. Since the change to square images I've not been able to create anything worthy of my watermark or publishing anywhere. Why such a drastic change from controllable (I always allowed and embraced the AIs own creativity) to often nonsense images or semi related images. I've tried with all profiles switched off, although I have a profile and a collection honed just for image gen that was working really well. I love the product here but currently image generation is below way below par, and unusable for the purposes I was previously using it for. simple examples here https://www.perplexity.ai/search/scottish-parliment-with-Vx.JYuwER..lpLDgxEWBxQ?s=c

[steel magnet]

Hey! Thanks for the feedback. We have been iterating on image generation internally and made some changes that might have affected your old workflow. We are going to be making more changes soon to give users a bit more control which hopefully should resolve these issues. Will share more detail once that is ready. Thanks for the patience and for using image generation!

[rocky timber]

The image gen has become one of my primary uses. I realise I was squeezing stuff out of it that maybe wasn't originally expected, but loved the process of developing a profile that really helped. I miss the creative joy I was getting from it. Happy to feed back and really hoping for a return to some more consistent predictability. Happy to update my workflow, in-fact would enjoy developing a new workflow, I just don't think that today I'd get the returns the product should really give, and has given previously. Keep up the development, I am a perplexity evangelist! I actually just picked up a couple of commissions for images but right now can't realise my visions.

[safe fulcrum]

You can try modifying your AI profile to ask that searches try to avoid those kind of sources

[ornate fulcrum]

Hi! Are there plans on improving inline latex support? It seems to not recognize it correctly?

I queried on the web version

[visual bough]

#👉│feedback-general message

[ornate fulcrum]

As the mobile app has the same problem, I see this as general feedback

[visual bough]

What do you mean?

[ornate fulcrum]

Nevermind, I didnt see that your message is a link to a message in this channel. Thanks for the tip, I will try that out!

[ornate fulcrum]

New Feedback: I mostly query with screenshots from lecture videos. (see simple example above)
The AI understands the image content and gives a great explanation, but the image search on the right side does not use the image content.
The title of my threads are also just a question mark or "explain this".
If the input is just an image, it would be nice to have automatic naming of threads and improved automatic queries for image search.

It wont let me send the image as a query alone (like with ChatGPT), it will only send the query if text is written in the field.

[vapid fable]

One thing I've noticed is that perplexity will give me much more "sparse" outputs than using chatgpt on openai.com. I never get the detailed responses like I usually get

[covert sun]

Hey, @vapid fable! Are you referring to the "Writing" focus answers? If you prefer long, detailed outputs, you can set it once in your Profile settings, examples available in the #1124060874785685554 channel.

[ornate fulcrum]

I don't know why, but it has stopped working since today. No matter how I rephrase the tip you gave, it wont format inline latex properly. :/

[pastel solstice]

In my AI profile I have specified my location to Germany. What I noticed is that when an answer included links from for example amazon it gave me the amazon.com site back. I tweaked my profile so that it now gives back amazon.de or politico.eu first. Maybe you should generally prioritize local versions of websites as it also has a range of advantages like privacy, currency etc.

[subtle kraken]

Is this the best place for a feature request?

[noble monolith]

https://discord.com/channels/1047197230748151888/1047649619695390740

[ancient compass]

even though i paid and selected for the chatgpt 4.0 its slower and does not do what i ask for as good as the free version of chat gpt 3.5??? why is that so?

[scarlet totem]

FYI the "share" "rewrite" and other buttons appear in the footer on the FAQ pages, like this one https://blog.perplexity.ai/faq/how-does-file-upload-work

[covert sun]

Hey, @ancient compass! Which mode are you using? Also, if you see some inaccurate answers, please create a report in the #1140622008086970420 with links to the threads, so we can compare those, as well.

[covert sun]

Thanks, @scarlet totem! We'll fix

[ornate fulcrum]

I am also a pro user with GPT-4 enabled and it also seems to be using GPT-3 for me. I used writing mode in these examples. I have also noticed that answers are less detailed than prompting ChatGPT with GPT-4 directly.

[covert sun]

Hey, @ornate fulcrum! https://discord.com/channels/1047197230748151888/1173623903596326912. It's GPT-4 and you can prompt the model to give more detailed answers. Put an instruction that you always want more detailed answers in your Profile settings and it will be applied to all of your queries.

[ornate fulcrum]

Ahh thats good to know, thanks!

[glacial granite]

Hey @ornate fulcrum, if you find the original message helpful, please consider reacting to it with the ⭐ emoji. If the post is appreciated by the community and receives 5 stars, it will go to the ⁠⭐│starred channel and the post author will get the EXPLORER role on Perplexity.

[thin folio]

As someone who uses the iPad Pro Magic Keyboard and trackpad, it would be great if you could add an onKeyPress event to trigger the search function when the Enter key is pressed on the keyboard. Even CMD SHIFT Enter would be helpful.

[covert sun]

Thanks, @thin folio! We have this request on our backlog.

[tawdry sinew]

I am realizing one truth every time I use Perplexity I am used to the interfaces of Bing Copilot, Bard or CHAT GPT. So the problem is with the people. you think Perplexity has the same approach but it's different. Perplexity is based on prompt roughly but not the same as a chat interface. So from here people are accustomed to chat interfaces. Maybe it would be very good to make less practical examples. more focused on everyday lives. Can you explain what I mean?

I myself think I make that mistake of thinking of Perplexity as if it were one of the interfaces mentioned. Am I right that Perplexity has a different use?

The problem that we have become accustomed to the chat format. And you have to learn to understand the generative search format

[thin folio]

Look forward to it 🤞

[knotty minnow]

I think the API documentation states somewhere that turn-by-turn conversations do not work for the online models; this obviously isn't the same with the web interface (where users are encouraged to enter a "Follow Up" question/query), but it does seem that the Perplexity web app also has some limitations in terms of preserving/maintaining a conversation (though interesting to note, this does not seem to be the case with all models #1196334894301839451 message )

[kind nymph]

Expansion to the "related" prompts section

• Expansion of "Related" Prompts:

"Introduce a feature that allows users to expand a selected prompt by generating three sub-prompts or alternative prompts beneath it. This would provide users with more specific directions or variations on the initial idea."

• Multi-Prompt Utilization:

"Implement the capability for users to work with all three prompts simultaneously, ensuring that the same text history is maintained across the prompts. This would help prevent overloading the AI's memory and maintain context continuity."

• Custom Keyword Input:

"Provide an option for users to input specific keywords or suggestions. The AI would then generate a list of prompts or content based on the user's input, offering a tailored experience and more relevant results."

• Prompt Reroll Function:

"Add a reroll feature that enables users to refresh the current prompts and receive a new set of suggestions. This would be useful for users seeking fresh inspiration or different angles on a topic."

[void moon]

I want to report a bug. Every time I start a conversation, I get an answer, then I modify the question and get a new answer, which is good. However, after the new answer is completed, the original modified answer also appears. Due to the context, this affects my subsequent questions. It didn't used to be like this!

[heady plover]

I don't like creating images with Perplexity at all. The function is miles behind that of the original https://chat.openai.com/. This is a reason for me to cancel Perplexity and switch back to OpenAI.

[visual bough]

Perplexity is not a chatbot like ChatGPT, the image generation is just an additional function to contextually enrich the answer.

[heady plover]

Yes, that is the problem. And German texts often create an image that has absolutely nothing to do with the content.

[spring plaza]

The new widget takes more space than it needs

[covert sun]

Thanks for reporting, @spring plaza! Could you share more details here #📱︱feedback-mobile message

[knotty minnow]

https://discord.com/channels/1047197230748151888/1194794305362071552 This gives the impression that some users are (attempting to) use Perplexity as a standalone image generator.. which I find kinda curious: like if people want to generate artistic images, a platform with the tagline "Where knowledge begins" seems like a strange place to do that, given there are now numerous generative models/platforms that are actually designed for text-to-image uses...Each to their own, of course (and I understand people are likely just trying to maximise the capabilities of the platform, rather than use it as a clunky substitute). But imho I don't think Perplexity can be both the Swiss Army knife of AI and the leading web-augmented / knowledge-focussed system (or at least, the more it tries to do 'everything', the more I would expect its knowledge-centric approach/capabilities to diminish..)

Bear with me, this is meant to constructive, not just criticism – I use Perplexity daily for all kinds of tasks (usually about recent events), it's fantastic.

But I have hard time understanding how image generation can enrich the responses, if it is indeed 'knowledge/information' that one is looking for (as opposed to an explicitly creative/artistic image request). An AI generated illustration, drawing or photograph seems weird in the context of knowledge; wouldn't the only images that would be relevant/enriching be those that actually exist (say for the query "Davos 2024 protests" – why would an AI imagined image be useful over actual images of protests that have taken place?)

Where image generation would actually be useful/relevant/enriching in the context of knowledge, would be diagrams – but the current implementation of Diagram generation is, to put it bluntly, utterly useless (especially when compared to actual images/diagrams retrieved from the web related to the query)

https://www.perplexity.ai/search/House-of-Saud-SDULoYjIQ.KWRw_o4PHJfA?s=c

https://www.perplexity.ai/search/How-does-the-yRx50vKTTdmHvxWth48u4Q?s=c#c91c79d2-f293-4dd9-87bf-15ad878f2ee1

https://www.perplexity.ai/search/Provide-a-timeline-H1_z5SgXSeijiTYnfCGgvQ?s=c#1f5ff3e5-2817-49e8-a389-36277c21a0bd

But... useful diagrams can seemingly be generated based on Perplexity's outputs. In the following screenshots, I paste Perplexity's response into Excalidraw's (Obsidian Plugin) text-to-diagram generator. The diagrams are actually based on the text and useful / enriching (rather than just gibberish)

[vapid mason]

BUG REPORT: Dear GOD please fix your bloody annoying website. 10% of the time when I type a big long question, it redirects to a new page with all my typing wiped out, and I can't get it back when I click the "Back" button either.

Your website need to NEVER DROP USER INPUT ON THE FLOOR

If you have to redirect, at least do us the courtesy of not blanking out our questions.

[livid hedge]

a) In the library page, I'd like a select and delete feature for efficiency - like the checkboxes in gmail to select a specific group of chats to delete

b) I created a collection named after me, then changed the name to "universe". But whenever I open the "universe" collection, the "Suggestions" are still based on the earlier name. This should be fixed, but I also find the suggestions intrusive since they appear before "Threads". The point of a collection is to focus on a history of chats, so I'd like "Suggestions" to appear after(below) "Threads" at the end of the page.

c) I'd like the chrome extension for default search (not AI companion) to not open perplexity.ai whenever I open chrome fresh. It should be focused on being the default search tab engine and nothing more. I like my default homepage. Changing to bing for example did not have this issue, and also I am unable to set up default search via the "%s" in url approach in chrome settings, which again bing gets right.

[knotty minnow]

https://www.perplexity.ai/search?focus=internet&copilot=true&q=%s works for me in Chrome

[livid hedge]

That worked beautifully! thank you very much!

[glacial granite]

Hey @livid hedge, if you find the original message helpful, please consider reacting to it with the ⭐ emoji. If the post is appreciated by the community and receives 5 stars, it will go to the ⁠⭐│starred channel and the post author will get the EXPLORER role on Perplexity.

[past holly]

On the iOS app, it will be great to swipe to delete existing threads.

[covert sun]

Hey, @past holly! This feedback would be better to post in the #🍏│feedback-ios channel. Also, there's a #1193855453365547048 message, we'll fix it soon.

[livid hedge]

A general one (hopefully): Why does editing a follow-up question create it as a new follow up question? Doesn't that undermine the notion of editing a question? Here I also did not like chatgpt in fact, because once you edit a question it removes all the questions that follow.

[heady plover]

I mean, how else do you suggest it should work

[coral kayak]

https://twitter.com/MaxWeichart/status/1748285969541812710

[livid hedge]

I think editing a particular question in a chain of questions should -
a) directly edit that question and generate updated answer
b) not delete the questions that follow after the edited one. This makes sense because there is a nonzero chance that the questions that follow are intentionally uncorrelated to the edited one. Even otherwise, I could simply delete the irrelevant followups before starting a new one [i'm not saving internal queries by editing anyway, so why bother]. The beauty of editing a followup is purely in how it reorganizes my mind. Hope that made sense! Thanks!

[knotty minnow]

It may be feasible in some way, but I don't think it really aligns with how these models work (to the extent I have any idea..). If you ask a follow up question (and then other, etc), you're still working within the same token context window of the underlying LLM. Each 'conversation' could be thought of as a 'chain' (as you invoke yourself), but if you go back and edit one of the links/components of the chain, the conversation/chain proceeds from there. Preserving whatever it was that was answered to another question (with what would have been different a context window in a different state) sounds messy, and I'm not sure would really work (at least not easily and afaik)

[livid hedge]

I see, very interesting insight, thanks for sharing! will read about "token context window". I'm an outsider to the field in a sense, but feel like fixing this would be useful to the community

[glacial granite]

Hey @livid hedge, if you find the original message helpful, please consider reacting to it with the ⭐ emoji. If the post is appreciated by the community and receives 5 stars, it will go to the ⁠⭐│starred channel and the post author will get the EXPLORER role on Perplexity.

[gritty heath]

It would be great to 1. Have a list spawn multiple threads in a specific collection. I have a pretty good multi-step research prompt as a collection. And would like to say, for these 10 things, start a thread using that collection/prompt. Doesn’t look like api or app or website support that. 2. Would like to use collection a source to ask questions about. Like “in this collection what’s the highest scoring x” or some such high level meta poll/survey.

[unreal jackal]

Hey Perplexity Team!

Just wanted to drop a quick shoutout for fixing the Latex issue. Really appreciate the effort! Loving the improvements – great job! 👍

[nimble tendon]

#👉│feedback-general The android app orientation is fixed, it would be nice if we have an adaptive layout to fit landscape mode. If the team has this in their bucket, thanks. Hoping to see this feature roll out soon

[fathom whale]

Hello, here to request for a feature. Is it possible to create a context from a bunch of files being it PDF or Markdown and use it as a template for multiple chat prompts. Because now if I want to create multiple chat prompts I have to re-upload the PDF, and I guess it will be optimal if I can make model once and use it multiple times

[wispy basin]

Common are you kidding me this cannot search the Web then i paid $20 for this tool huh? 🤔

[covert sun]

Hey, @wispy basin! If you keep facing this issue, please submit a report in the #1140622008086970420 with links to the threads that had such results, we'll check what happened in those cases. That would be helpful to the team and the rest of the users, thank you!

[sly pivot]

Hey @covert sun I’m experiencing the same issue as the user who commented earlier about not being able to search. I’ll be posting a screen recording in the Bug reports channel shortly.

[split terrace]

anything on the roapmap to implement a statuspage of some sort for the web app and others? only thing I see is the API status page at https://docs.perplexity.ai/page/application-status

[weary narwhal]

Hey team! Just signed up for a pro account the other day and have been exclusively using it for things I'd normally turn to Google for (and really liking it so far).

One thing I would LOVE to see is an ability to shape the kind of information that shows up in the Discover tab.

On my Android phone, I frequent the Discover feature of the Google app (almost daily) to find articles and content that I'm interested in.. The added ability to create your own feed of 'interests' would be AMAZING!

[dusty wedge]

I'm sure it's in the cards, but Mistral-Medium would be great as an option in main Perplexity and not just labs. It's currently the second highest ranked LLM among users and is great for writing fiction because it isn't very censored.

[covert sun]

Thanks for your support, @weary narwhal! Yeah, adding more customization and personalization to Discover is a very good idea and is the future of it.

[drifting jay]

Not sure where to post enhancement requests but I'd love to see bulk edits of the Library, eg. move multiple threads to a folder, Delete multiple threads.

[covert sun]

This would be better in the #💭│ideas channel, but it's okay. We've got such a request before, the team is aware of this idea, it's a good one to come in the future releases.

[drifting jay]

Thanks for the quick reply period I'll make sure I post there in the future

[glacial granite]

Hey @drifting jay, if you find the original message helpful, please consider reacting to it with the ⭐ emoji. If the post is appreciated by the community and receives 5 stars, it will go to the ⁠⭐│starred channel and the post author will get the EXPLORER role on Perplexity.

[empty jungle]

Joined Perplexity less than an hour ago (via Rabbit r1 purchase) and am super impressed with the web UI (haven't tried iOS yet) but also the organisation and features of this server. It's great to see the team active on here and everything feels super accessible. Will be sticking around and already have the sense that Perplexity is going to become my LLM of choice... not least of all because I can still access third-party models when I want to! Impressive stuff 🏆

[wise basin]

Hello I sent an email to support and haven’t heard a response yet. I have a perplexity acccount and bought a rabbit r1. How do I get the credit on my account?

[noble monolith]

welcome! #💬│general message

[river path]

I sent an email too, I was unable to redeem my credit so far. However @wise basin I did see this but haven't tried: #💬│general message

[wispy basin]

Sure next time i will share the link.

Team please add data export to PDF,TXT,Excel or other format that is the most importatnt feature missing out there.

[scarlet current]

same sent the email over the weekend, still ahvent gotten any feedback. waiting now..

[oak valve]

Same here. Emailed last week, still haven't gotten a response. Tried redeeming the r1 credit but ended up getting charged $200 immediately (which seemed to have bypassed the 7-day free trial). All kinds of messed up things going on with Perplexity...I'm not exactly confident in the business right now.

[south ether]

Hello, I'm sorry for the delayed response, we have gotten a lot of emails (no excuse) and we are working through them. Would you mind resending the email ?

[south ether]

same to you @oak valve and @scarlet current

[oak valve]

Just resent

[wise basin]

Re sent ty

[steel niche]

I was looking for a text to speech function today.
Use case - I drive an hour a day, which is great time for ideas, problem solving etc. I can easily hit the mic icon and speech to text a question, but can't read the response unless I pull over or am at a stoplight.
If there was a way to set a text to speech function ideally automatically on the answer until turned off, then I could have have an extra hour a day of effective time. It would make perplexity a lot more useful for me on a regular basis.

[hollow nova]

Important ‼️ Truly upvoting this, I for sure would like to see it.

[last lintel]

i just used the perplexity it has no predictive search queries please add predictive search queries

[covert sun]

Could you please elaborate, @last lintel.

[last lintel]

just like typing in google the search bar predicts what words to search

is that understandable?

mr romanov are you there mr romanov?

or am i chatting with ai auto answering

?

[last lintel]

oke never mind that’s alright have a good day

[covert sun]

You meant autosuggests, thanks for clarifying! We do have them, but we'll work more on it 🙏

[last lintel]

good luck

[slow pilot]

Is there any chance to move the setting that allow to choose specific models into the main page? Like next to the "focus" and so on? It would save HELL OF a time.

[noble monolith]

#💭│ideas message

[keen junco]

I know this is a commonly requested feature, but just adding another voice:

The ability to change focus modes within a thread would be amazing. Perplexity is amazing at searching the web, but a single thread loses it's usefulness to me very quickly when I can't ask follow-up questions without it going out and trying to search the web again. Sometimes I just want to chat with the LLM in the context of the thread and what it has researched so far.

[oak mauve]

Now that the Perplexity team appears to be working on integrating it as search engines on webbrowsers, please work on adding it into Firefox ASAP aswell! 🤩 (apparently, it can be added as an "extension" on Mozilla's Add-Ons page at https://addons.mozilla.org/en-US/firefox/extensions/category/search-tools/ . You have to visit the "submit or manage extensions" button on https://addons.mozilla.org/en-US/developers/ to create such a search-engine I believe.)

[obtuse rivet]

Hi this might've been mentioned before (idk cause I'm new) but math seems to be a problem on here. I am asking for conversions between different binary representations such as two's complement, regular signed, and unsigned, and it has gotten it wrong a few times. Especially with difference between two's complement and regular signed.

[obtuse rivet]

Safari (idek if this is possible) and Chrome would be insane as well!

[covert sun]

Hey, @oak mauve! Perplexity might be not on the list of default search engines now, but it can be easily set as a default search engine on FF: #⚡│ask-community message

Hey, @obtuse rivet! For Chrome it's also easy to make it a default search engine with the official extension: #⚡│ask-community message or manually. For Safari, there's no official extension, but some users do it with xSearch.

[oak mauve]

Awesome, thanks so much!

[shut arch]

Hey @oak mauve!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[oak mauve]

Just be aware that a Chromium-based browser like Brave is a lot beter, privacy wise, than Google's own Chrome. (see https://www.perplexity.ai/search/Are-you-aware-Q1WMGc.3TguxDSDUCPNtWg?s=c#43558c19-cfb7-4e0b-b10d-20d408f36d5a for more interesting details.)

(Google was not being transparent about this and did NOT obtain explicit user consent before scanning files.)

[gleaming ingot]

Hi everyone, I'm a perplexity subscriber and Arc user, but after updating Arc I don't have an option to select Perplexity as my browser.

Is it working for anyone?

[covert sun]

Hey, @gleaming ingot! You can see it if you create a new profile, but Arc team has posted on Twitter about the issue, and you can also wait till they announce that it's fixed.

[gleaming ingot]

Ahh, I see, thanks 🙂

[shut arch]

Hey @gleaming ingot!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[thick geyser]

Trying to log in using Google sign on in Chrome - the site does not recognise if there are multiple Google accounts registered in Chrome and so automatically uses the first it can find - should offer a choice of which Google account you want to use - especially annoying as you commence the registration process automatically using the incorrect google account email and create the account for the wrong email - endless loop. Seems I have to use Safari to get out of this automated process?

[dusty wedge]

I can't even explain how much time Perplexity's Academic focus saved me at work today. I had to fact-check health claims for about 50 different foods, and it would have taken me ages to do manually. So a massive thank you to the entire Perplexity team <3

[autumn schooner]

Not a high priority. It would be nice to see a graph of your last 30 days usage on the account page.

[dusk aspen]

Hi team, just want to let you know that ScienceDirect which has a lot of collated freely viewable scientific articles is no longer accessible to Perplexity. The below thread shows a failed direct request for article summarization, but the more pressing issue is that Perplexity often tries to get information from there on many occasions when collating sources for a general research topic, and it can't extract any information from these sources, thus wasting its search context.
https://www.perplexity.ai/search/Detail-the-topic-uq36w0uKSp.P2V8f8zdObw?s=c

[autumn schooner]

Add a channel to the discord server to discuss different models that are available. It would help to consolodate that information into one place. As it is now, there's a lot of knowledge spread across several different channels.

[visual bough]

Please use #📝│server-ideas.
https://discord.com/channels/1047197230748151888/1197976839075922012 👀

[tranquil moon]

Hi - I added my R1 perplexity credit, and then I canceled the renewal to avoid renewing it if I am not using it in a year and forget. Now, it looks like my subscription is no longer active at all. Is that the intended behavior? Is there any way to undo that

[visual bough]

Hey @tranquil moon!
What do you mean by "looks like"? Does it say Pro next to your profile picture and can you still use GPT-4?

[tranquil moon]

Hi! It shows this (redacted since I doxxed myself)

if I click on account settings and then subscription, it brings me to the subscription page.

[visual bough]

But can you still use GPT-4?

[tranquil moon]

There's no longer an option to switch models

and if I go to "AI Model" in settings, the dropdown is disabled

[visual bough]

So you already had Pro access and could use GPT-4 and after you pressed "Cancel Plan" you lost Pro access?

[tranquil moon]

right

I also have the 4-every 4 hours limit on copilot

[visual bough]

Strange, please send an email to [e-mail], please enter the email address of your Perplexity account and briefly describe the situation, after the weekend your case will be processed 🙏🏻

[tranquil moon]

done, thanks!

[shut arch]

Hey @tranquil moon!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[flat anvil]

Hello, it's confusing, that editing a message does now generate another answer instead of overwriting the old.

[covert sun]

Right, @flat anvil, sorry for that. We have a ticket for it and the team is working on a fix.

[south ether]

Hi I will help you through support, but just as general FYI if other people encounter this same issue.

When you register for an annual subscription for the first time, you start with a 7-day free trial. If you decide to cancel within this trial period, you will lose access to the Pro version.

This is because you're cancelling the trial, not the actual subscription. The subscription itself can only be cancelled after the 7-day trial period, as that's when the actual subscription begins.

[flat anvil]

Thanks

[shut arch]

Hey @flat anvil!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[pure relic]

It would be great to see if Perplexity would accept sitemap.xml to make targeted searches

[tight hemlock]

I am using the api with online models and they don't give back sources, why?

is the team working on it?

[visual bough]

#🧪│api-general message

[honest parrot]

Gotta admit, really not excited to use a third party browser extension to export my Perplixity history. Please add this much-needed feature (GPT4 has). Thanks! Love the product.

[void moon]

Another bug has surfaced. Sometimes when I post a question, it gets swallowed up directly, and only after the content is generated and I refresh the page does it appear.

[spare harness]

Hi Daniela, I have hit this situation but I used a voucher (Rabbit R1 partnership). I have tried to reapply the voucher but it isnt working now.

[half thunder]

upload picture with Barechested Male, but failed

[covert sun]

Hey, @half thunder! Not sure about the picture you uploaded, but the model has moderation which the uploaded image didn't pass.

[dawn rover]

Nice! We can choose which model to use for image generation : dalle-e or sdxl. See it on settings.
So sad it is not available per chat or per collection (for now)... 🥰

[hoary pier]

Re: the offer from Rabbit, here's the email I got. Am I the only one who doesn't have a way of actually applying the credits?

[covert sun]

Hey, @hoary pier! When did you order the device? This doesn't look like the email with the bonus, yet. #💬│general message

[hoary pier]

I was in the second batch

[covert sun]

Then you surely should've got it. Please check with the rabbit support team, they probably need to send it again. When you see your redeem credit link in an email, this message could also be helpful to you: #💬│general message

[hoary pier]

I wrote to rabbit support last week, was hoping to get a faster workaround here. Thanks anyway. 🙂

[noble monolith]

available per thread #1200186354022228110 message

[opal vine]

In a Pro subscriber and big fan. I have been telling friends about Perplexity like crazy! But I am legally blind, and I use larger text accessibility features on the iPhone. The latest app update broke the feature and now text is cut off 😭

It now looks like this

I paid for the whole year…

I will stick with perplexity but I hope this 100% reproducible bug can get a look.

[covert sun]

Hey, @opal vine, thanks for your support and for reporting - this bug is reproducible and we'll work on a fix.

[lime rover]

Hey guys, quick question about perplexity's API. Is there any way to get the sources used to generate the response? Similarly to how the site works?

Thanks in advance!

[desert token]

Same here

[noble monolith]

#🧪│api-general message

[flat anvil]

I REALLY would like, when the base Perplexity AI would FOLLOW/USE the infos in my profile.
As example, I like to talk in German, but write my work in English.
e.g.:
Hilf mir damit (help me with that):
"This is a long text, which I want to shorten."
and then it should make it into a "md box"

this is a shorter text.

okay, if you use now the "md box" (md stans for markdown) you can't easy scroll from left to right, it stutters and after 2sec. it scrolls instead up/down on the page. so you have to swipe, swipe, swipe. always lifting your finger up and down, because it doesn't work in a flow. Never had that on the website.

uhm... what the heck is it trying to tell me O_o? do I get now through the app Social media spam about stuff I don't want? Aside from youtube/twitch I don't use any social media.

okay, maybe discord, if that counts as social media?

[amber berry]

Yep, here too. This bug is so bad I have to clear cache every time I send another message.

[flat anvil]

Uhm... I can't find my user profile in the app, where can I change my custom info stuff?

in the app

[novel karma]

I'm having trouble running API requests in parallel. It seems about a third of requests fail completely when I'm running at 5-fold or 10-fold request parallelism. Is that expected?

[visual bough]

Hey @novel karma!
Which model do you use, did you pay attention to the rate limits?
https://docs.perplexity.ai/docs/rate-limits
Please also use #🧪│api-general for API related questions ☺️

[covert sun]

Hey, @flat anvil! You can turn the notifications off in the app settings.
As for the profile custom settings, they are not available to edit via the app interface, yet, only the language setting is. But those instructions should be applied to your answers.

[flat anvil]

oh okay... I never wanted such notifications... I'm not happy with the app. 😦 After the annoying popup EVERYTIME in web, I expected your team spent more love into the app...

[crimson nymph]

the pop-up in the browser should be only once per session once logged in

[grand holly]

https://www.perplexity.ai/search/List-all-railway-G6DuPIEGQm65FTssanTAlA?s=c

This was a rather frustrating experience for me.

I was expecting an answer similar to this.

[safe fulcrum]

were you using the mobile app for that thread?

[grand holly]

No. Browser.

Browser on phone.

Bard results for comparison.

[safe fulcrum]

Ah—I know with the mobile apps there’s a special instruction in the system prompt to give concise replies around 100 words in length; however, I don’t know if perplexity uses the same prompt on the web version by checking for mobile browser

The responses in that thread seem like they were being affected by that short answer instruction

[grand holly]

You mean I should get better results on desktop mode?

[safe fulcrum]

depends on what you’re searching for but yeah you may get more detailed responses on the desktop web version

Not sure if you have access but there’s a thread in pro feedback about this https://discord.com/channels/1047197230748151888/1169166673559945227

Mobile interface: Write concise text adapted to a mobile user (less than 200 words)

^that’s the extra line in the system prompt for mobile

[half thunder]

#1140622008086970420
question: "What is the current value of the Shanghai Composite Index 000001.SH?"
answer: 根据搜索结果，上海证券交易所综合指数（代码：000001.SH）最新的收盘价为2,770.74点，下跌了17.81点，跌幅为0.64%[2][3].

Citations:
[1] https://www.marketwatch.com/investing/index/shcomp?countrycode=cn
[2] https://consent.yahoo.com/v2/collectConsent
[3] https://www.bloomberg.com/quote/SHCOMP:IND
[4] https://www.wsj.com/market-data/quotes/index/CN/SHCOMP/historical-prices
[5] https://www.moomoo.com/stock/000001-SH

https://www.perplexity.ai/search/What-is-the-6ImYFeTeQPuwG2Gk80RNRg?s=c#e8899815-e4de-40fb-b01b-61a4f3444d46
but , actually the number is not right.

[half thunder]

the number is not realtime .
根据搜索结果，上海证券交易所综合指数（代码：000001.SH）最新的收盘价为2,770.74点，下跌了17.81点，跌幅为0.64%[2][3].
act：2,750.72 from bloomberg

[heady plover]

try wolfram alpha focus

[half thunder]

the Shanghai Composite Index lastest value
https://www.perplexity.ai/search/the-Shanghai-Composite-tjARpSrFSTGYQcxLWvKL.A?s=c#b63011a5-2ac5-4931-9841-cc4b5af28bf8
can not return the value， i am using claude 2.1
but :

PPLX-70B-Online is right .

[hollow minnow]

Ppl in China are blocked from access to Perplexity, and have to use VPN. However, with VPN on, the iOS app simply won’t work anymore. Is there any workaround except using the website version?

[flat anvil]

no, EVERY time I open the site.

[flat anvil]

leave china 🙃.
I don't like the app, so the web version is better 😉

[heady plover]

just use uBO's element picker for now

works fine for me

until they fix it, I mean

and I agree, it's really annoying

[flat anvil]

I'm talking about the phone

On PC a app popup would make no sense

In opera mobile is no ublock useable

[heady plover]

oh, fwiw it works in firefox on mobile

[knotty minnow]

This iask.ai search engine seems to really struggle for questions related about recent events. Ask it 'who won the most recent cricket world cup' or 'where is Mahomoud Bazoum?' and the responses appear to rely on very outdated information and are rubbish

[heady plover]

because there's no date filter

it's a simple solution but hasn't been implemented...

[knotty minnow]

Yeah but it points to sources, like wikpedia pages, that do have accurate and up-to-date information. The response is like it is generated by an LLM with a knowledge cut off date

I'll try using date filters out of interest (but Perplexity requires no such specifications; e.g. it figures out what 'latest' means and returns results accordingly

[knotty minnow]

Oh, there is no such filter..

[heady plover]

that's what I said

that said, you could try before: and after: might still work

I think the format is after:dd-mm-yyyy

[knotty minnow]

I tried that (and also the format used by the Google operator e.g. yyyy-mm-dd) and it didn't seem to make a difference. This response is just wrong (even though it ends emphasising how it relied on authoritative sources and the importance thereof..)

The 'web results' at the very bottom actually do have accurate/up-to-date information - one would be better off clicking on one of those than reading the 'answer' above

[heady plover]

rip

also my bad, seems I misremembered. yeah it's yyyy-mm-dd

oh wait, actually

I don't think it uses those search results specifically

those are web search results

[heady plover]

didn't realise that, is there a way to remove it ?

[dawn rover]

You are sur you select focus mode to "all"? For me it gives me up-to-date results (ivoiry coast soccer half finalists, played only yesterday)

[knotty minnow]

sorry to clarify, by 'this search engine' I meant iask.ai (the one in the screenshot I was replying to), not Perplexity

[heady plover]

I'm disappointed that when I asked questions about combating fake news and thinking critically in a "post-truth" era (or error) that a post from a homeschooling website with a "religous worldview" (not specifying which) which was arguably (at least) not reliable was included as a source - just saying - it was very random. But glad that sources were displayed so I could see that not all the sources were reliable (or truthful) . I just can't see paying $20.00 a month for this.

[crude skiff]

Would love to see an option to anonymize shared threads. Right now, when I share a thread, my username is visible. That's ok in most cases, but sometimes, I'd prefer not do doxx myself.

[noble monolith]

#💭│ideas message

[knotty minnow]

In addition to reviewing the sources, it is also possible to remove any you do not like and to then regenerate the response without it/them

[heady plover]

Thanks! I did that, but I could only regenerate with co-pilot which wasn't what I originally used, so it caused me to use a credit. If this happens a lot, I'll use up all my credits. I don't get how some unverified personal website gets included as an authority along with major universities and journals Must be a right-wing bias. (sorry for the politics - but the results really do seem political)

[shut arch]

Hey @heady plover!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[knotty minnow]

Apologies - expending a Copilot credit was not something I had considered when making the suggestion, but indeed, that is a part of how that functionality works.
In any case, it's a bit difficult to fully understand your concern / assertion re "Must be a right wing bias", particularly given that you state that the particular source in question was included "along with major universities and journals". If it were the other way around (i.e. all the sources were dubious / right-wing, except for a single outlier that was reputable / academic), that would seem more indicative of a clear right wing bias in the way you are describing (at least to my mind)

I would also evaluate the actual response - rather than singling-out a single source - as evidence of any such bias. fwiw, the response here would not seem indicative of a 'right wing bias' (in fact, it appears a robust and useful response with sources primarily from academic institutions, thinktanks etc; though I've admittedly only skimmed over it)
And as a final aside, it's not clear to me that 'right wing' sources have a total monopoly over the dissemination of fake news (though granted, they're probably close to lol), or that arbitrarily excluding sources based on one a particular political / partisan ideology (i.e. creating an echo chamber) is a particularly healthy approach to media literacy

[heady plover]

I wasn't being entirely serious about "right-wing bias" (that's why the lol was there) -I mention it to contrast another alleged bias that is more often complained about. The fact that it showed up in the context of my search was ironic because it was a teaching how to not think critically, but "logically" - in keeping with a certain worldview

[knotty minnow]

No worries – I'm not really sure what your expectations / complaints are exactly. No system that pulls information from the internet is going to be perfect all the time for everyone. Again, I wouldn't get too caught up about one particular source (or extrapolate a system-wide bias from the existence of a singular source, among several others), but would look at the response itself. Anyway, each to there own – best of luck luck!

[little remnant]

I'm working on a project that has 30-40 Word docs, each about 1000 words, and I'd like to have Perplexity rewrite them. However, I get unacceptably inconsistent quality and length results, even when re-running the same doc. What am I doing wrong?

[covert sun]

Hey, @little remnant! It's hard to say without examples of what the docs, your prompts and results are. The general recommendations would be to try the Writing/All focus, Copilot on and off, Claude or GPT-4 model, splitting documents in parts to focus on quality of each, and making detailed prompts on how exactly you want you texts to be rewritten.

[wispy basin]

Hi devs perplexity still cannot do search from the developer perspective and it cannot give me any accurate information from Github even though there is free API to search Github repos (api.github.com)

See this conversataion:
https://www.perplexity.ai/search/show-me-top-qGTSHjMoQ2ixrkWA0mv9Pw?s=c

I think you should add the code interpreter, because other artificial intelligence tools are able to do this easily

[little remnant]

Thanks for the response. I've tried almost everything you mentioned. Still there's such a great disparity in results. Great but so inconsistent I don't think it's going to work for my project. I need high quality, consistent results. I don't expect the same results every time, but I was hoping for quality results most of the time. Thanks again for responding.

[shut arch]

Hey @little remnant!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[fiery laurel]

Maybe make the typebar static

[knotty minnow]

Do these outputs need to be informed by the internet connectivity that Perplexity brings? If so, passing 1000-word long texts to perplexity (which is more designed for receiving queries and providing answers) and expecting broadly consistent responses over 40-50 different texts would be very tricky. If not, your best bet imho would be selecting the Writing focus/mode + Copilot disabled, and sticking to the same model throughout

But tbh, if the internet connectivity is not essential for this project, but consistent outputs are, you would prob be better off using an approach that allows you to control the LLM's parameters (namely Temperature), whether pplx labs or another

[heady plover]

I have been use perplexity for 3 months and I decided to cancel them soon, because I guess 200,000 tokens Is totally BS when I really tried It.
It only summarize headers or conclusion of paragraph not full context of word.
I try both splitting various paragraph Into each documents and even assemble everything Into one document worth of text yet It still have made mistake at summarizing or Interpret text that Is before or older away.
Perplexity AI may allow unlimited files upload but It mostly useless If It model can't even Interpret other documents separately or have separate working memory to Interpret paragraph or something.
Like I have 3 documents that I have from clients but It lose It memory somewhere up to document number 2
Or even hallucinating new details that never been there In documents and almost got me scolded by my customers so It have risk of hallucinations and not good for my use case.
If combine all document together then It not even more than 20,000 words while Claude 2.1 claim that It context window Is up to 150,000 words Is somewhat B.S, maybe renamed this model to "Perplexity's Claude 2.1" or something so It won't be misleading to users because It not even "200,000 tokens" from what I see.

I upload one document and have conversations with It and It come out great but when text Is very long It will eventually forgot details and messed up multiple times.
Or pull out Irreverent Information from documents or Invent new paragraph that didn't exist.
And If this Is horrible, multiple documents Is much worse because It can't compare document together and probably don't even have Ideas of multiple documents actually exist.
I can upload up to 2-3 documents at best before It start hallucinating and forgot details I need and start to become unreliable.

If multiple documents Interpretation Is not even complete I see no point In unlimited documents upload because you literally can't work with more than 2 documents and It pretty annoying.
But at least normal Claude without any document upload Is better than document one.

Edit: Oh nevermind I try again to split document further Into 1300+ word per documents and It start to get better maybe I fix It now
I try split each documents to be as small as possible and It somehow work.

[dry beacon]

I'm having an issue lately; when I submit my input, sometimes the response will get printed multiple times (3+). If I close the thread and look at it again, only one response is present. Is this already a known issue?

[heady plover]

Why splitting document Into smaller one like 1000 words or so working but not large document?
I think I am at wrong here.

[knotty minnow]

Not sure you are wrong to be asking questions if you're having to split a large document into 1000-word chunks – strikes me as largely defeating the purpose of using an LLM with a large context window in the first place. I mean with chunks that size (let's say 1200-1300 tokens), you may as well be using any old model

Just quickly did a test using the report available here: https://appropriations.house.gov/sites/democrats.appropriations.house.gov/files/FY23 Summary of Appropriations Provisions.pdf

– Note that the doc is only 37,000 tokens (so not even close to the 200k token context window nominally available with C2.1)

– Also, this was done in a few mins - it may be inaccurate / misleading. But the report is available for others to test with; so take it with a pinch of salt and conduct your own tests etc

Here’s uploading that doc (as PDF) to Perplexity web with Claude 2.1 selected (Focus = Writing | Copilot = Disabled)

Contrast that against, after uploading the same to file and giving the same prompt, the responses from Claude 2.1 using Anthopic’s web chat interface and nat.dev’s playground...

And poe’s Claude 2 connection (note: 100k context window)

[knotty minnow]

Same report/doc but uploaded as .txt (instead of PDF) and with the file name referenced in prompt

[fossil meadow]

I have noticed too that the image recognition for copilot and gpt4 is worse in perplexity than in their respective websites.

Makes me think it might have something to do with api usage

[knotty minnow]

If you pay close attention to this Discord server, there are virtually no questions / complaints about Perplexity's ability to use the internet to provide quick and accurate responses about knowledge-based questions – which, according to Perplexity's marketing etc, is basically all it's meant to do (be a search engine basically; it does a fantastic job at it)

Almost everything is about peripheral features/capabilities, like image generation, parsing long texts etc

Using perplexity to write a novel or analyse a very long piece of text, or produce a stunning piece of artwork, or devlop/troubleshoot complex code... seems similar to a hairdresser using a swiss army knife - sure, they'll be able to kinda do their work, but there are better, specialised tools out there, which most professional hairdressers would be using instead
I don't mean to be insensitive to individual circumstances in terms of access to other tools and the inequalities within countries and globally..It's just an observation..
Bottom line: perplexity is excellent at using the internet to provide robust answers to questions (esp. when those that are outside an LLM's knowledge cut-off),

[fossil meadow]

Yeah its why i use perplexity for research, poe for writing/text prompts(open llms are nice), and copilot/bard for image recognition

Wish they would implement the labs in the app

https://labs.perplexity.ai/

It would be perfect for the text prompts

[heady plover]

I agree with you, when I upload a large pdf document Claude abilities Is like reduced and It can't remember few (many) details.
But when I break document down Into 50KB or around 1000-1500 words or the most Is 1700 words It remember things better.
It kinda weird but at least I know a way to make It work now but have to waste some time splitting them.

[flat anvil]

I think you don't understand how much limited AIs currently are with their brain.
You can't expect that a AI can remember multiple files!
You have to work on each file and write stuff down.
Let it write a summary of each chapter or write important things on your own out.

[flat anvil]

I can confirm such optical bugs.
But was it in the web or app? Maybe you want to make a #1140622008086970420 here

[dry beacon]

Yeah, it's in the web app. Didn't have access to the bug reports channel until now

[flat anvil]

Discord hide many channels in servers, since it got the feature about hiding rooms.
You can see on top left in each discord server a button for "rooms" to see all 😉

[knotty minnow]

If an LLM being able to handle multiple files is an unreasonable expectation (which I'd quibble with), then the issue is the fact the system currently allows users to upload multiple files - not users' expectations around using such functionality

[heady plover]

I agree with you but Claude claim that It tokens limits or to be exact memories Is 200,000 Tokens or approximately 150,000 words and my largest document Is only 20,000 words If count with word count using google document, that Is many time less than what Claude claimed It can memorize, the problem Is that It can be misleading to users with those claims personally I do agree that AI have limits In their performance but problem Is fixed when I break one large document down Into multiple chunk of 1000-1700 words worth of many documents, It waste my time but It at least work.
But It still not the same as what these companies claimed It capabilities Is.

At first I thought one document with only 20,000 words Is < 150,000 words that they claim so I think It might work If their claim Is correct but I still have to break things down so their claim Is not so accurate.
However I don't really care now since my problem Is fixed for now (maybe because I didn't explore other aspects yet, maybe this exact problem will happen again.)

[safe fulcrum]

https://support.anthropic.com/en/articles/8324991-about-claude-pro-usage

[shy forge]

Feature suggestion : Please allow Perplexity to be able to open links provided in the search bar and use that data to create responses…

[knotty minnow]

Upload the Great Gatsby?! Umm

That would be pointless on Perplexity

[heady plover]

I experiment on It again and the files cap limits that can be uploaded Is 10.
When It reached 10 It will delete previous files which mean you can split large document Into 10 chucks at best but still better than nothing.

For example I upload 4 files all at once and when I upload up to 11-14 It will delete previous files/older files and capped at 10 files.
Maybe If I split them up to 3,000 words each would solve problem.

I wish they removed 10 files cap limits but It would be Impossible to sustain because Immeasurable computational power needed.

[heady plover]

Idk If Claude pro on their platform versus version that perplexity pro users using Is the same because perplexity adopted them and edited It Into their website.
I used to subscribe to both, perplexity have lower memories like I said but less censored and flexible.
When Claude In official website have higher memory but censor everything I ask It unless I requested It to write story about cat or something.

I don't think Claude Q&A on their own website will help much since I don't really know how perplexity Implement Claude on their own website because the vibe Is different when I compare two responses.

[grand holly]

Bard seems best among free ones rn. It sucked in math in early days. I wonder if they fixed that.

[limpid jungle]

I always get this error when I try the Chrome extension to answer anything on the current page . Extension permissions says it has access to all my pages. I rephrased this as just 'summarize' and not listing the url explicitely that I was on as well

[knotty minnow]

As far as I could tell, that was literally just an LLM with web results tacked on to the bottom of its responses (but in no way informing them) lol

But yeah Bard is interesting. I've found it too unreliable to be useful so far. But if reports (/leaks) are to be believed, 'Bard' is due to be rebranded as 'Gemini' as early as today (7 Feb). And if I understand things correctly, it will be powered by the same internet-connected version of Gemini that is on this leaderboard. Who knows, but certainly intriguing (the Bard/Gemini on that arena was a different Bard to the one we currently know - it's impressive)

And with that version as stock Gemini(/Bard), then I believe Gemini 'Ultra' is meant to be paid, and has I think stronger multi-modal capabillities (I think anyway, not really sure - actually should ask Perplexity lol )

[rotund flower]

can we please have a more robust "Library", where Threads are properly grouped under their respective Folders?
currently, even though u can tag a thread with a FolderName, theyre still all randomly arranged, perhaps by Date Last Modified, or something.
the only other way to do this, is to manually click into Library, click on the TagName, and then finally click on the ThreadName, which takes too long and is too tedious.

it would be so much better if all Threads are already grouped under their FolderNames, AND its also sorted by LastModified within each folder, AND each folder is also sorted by LastModified

[rose canyon]

copilot keeps turning on about once per day. I'd rather it not enable by itself.

[pastel solstice]

I was hallucinating a bit and asking fringe questions and I got wrong answers back. https://www.perplexity.ai/search/if-there-are-lxq9WSONQP6cZPttox1.CA Last question about zebras, 2nd source is about Zebra sharks and 4th from a company called zebraathletics. Good example from my experience of a partially good answer, but not reliable.

[limpid jungle]

Ai glasses sound great but we don't get responses to our feedback here. How to inspire trust that we can get help with the ai service

[distant matrix]

Thanks for this info. I have the same issue as many on this discord who got the 1 year free promo from Rabbit R1. Can the pro subscription be reapplied to our accounts? It seems like a "bug" in how your subscription period should work.

[shut arch]

Hey @distant matrix!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[chrome scaffold]

Is it possible to have that cloudflare thing removed for pro users? Its super annoying/slow

Also, I think this is related to cloudflare too, but when i come back to my pinned pplx tab after an hour or so to continue a thread, its like the site needs to reload to reestablish the cloudflare thingie.

Anyway, Cloudflare is creating friction / killing the stickiness for me.

[visual bough]

These are all security measures that are necessary to be able to offer the service 🙂

[chrome scaffold]

Sure, but similar services manage fine without it or with something better/faster/smoother. It maybe secure, but itt's still a poor user experience.

Out of curiosity, how necessary is it still for a paid pro user who is already logged in, assumedly verified through payment method etc?

[visual bough]

Let's say someone wants to use Perplexity automatically, they could just buy Pro for $20 and have no problems using it :/

[chrome scaffold]

Yes, sure. I'm sure there a lots of things to consider. But my feedback stands, its a poor user experience and a better solution for security would be amazing as a user.

[main roost]

Any admin for account support here? I need help

[visual bough]

You can describe your problem here 🙂

[main roost]

I logged on to web version of Perplaxity on a public computer and forgotten to log out. There isn't a function both the web and app to logout from all devices that's currently logged on.

[visual bough]

Yes, there is no way to log out all devices. I'm afraid there's nothing you can do, but to be on the safe side you can write another email to [e-mail] and explain your problem ☺️

[main roost]

Thank you!

[shut arch]

Hey @main roost!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[winter prism]

Concerning online versions

I went to the perplexity lab pages to ask:

Question

What is the guideline that prevents you from giving direct links to website and what is the reason for it.

Answer

The guideline that prevents providing direct links to websites is primarily due to concerns about the content of those websites and potential violations of community guidelines or terms of service. Some reasons for not allowing direct links include:

1. Content Violations: Direct links to websites or apps that install malware, engage in phishing, or provide unauthorized access to copyrighted materials are prohibited.
2. Hate and Harassment: Links to content that would violate hate or harassment policies are not allowed.
3. Violence and Misinformation: Links to content encouraging violence or spreading medical misinformation contradicting local health authorities' guidelines are also prohibited.
4. Affiliate Content: While affiliate content does not violate terms of service, excessively posting affiliate content in dedicated accounts may violate certain policies.

These guidelines aim to maintain a safe and secure environment for users and prevent the promotion of harmful or misleading content. Instead of providing direct links, it is recommended to describe the content or topic of the linked material and encourage users to search for it themselves if interested.

---

It might be clear to anyone why, considering this answer, Bing stays the best place to go for scientific research as unlike Perplexity it is able to give links. Sad!

[silent iron]

Is there a way to get back Related questions from older prompts I often would like to try the answers for more that one. Right now it disappears sadly.

[noble monolith]

you can over web version, just delete last answer

[cobalt crest]

Is there a discord bot that connects to Perplexity? I see one in the store but it doesn't seem to work

[crimson nymph]

The discord bots have been discontinued there is no official discord bot atm for perplexity

[shy kelp]

Hi @covert sun @south ether please can either of you help, I've been emailing support & pro each Friday for the last couple of weeks but no response. Basically I pre-ordered a Rabbit R1, used the link in the email to get my "free" 1 year of Perplexity Pro but have been charged full price for both, is there anything you can do to help at all? Kind regards

[sweet ruin]

Hi!

I'm a huge fan of Perplexity, and I'm thinking of getting the Pro plan for the API access.

However, could I get some clarification on the ToS, I'm using a non-English localized chatbot that would use Perplexity API as its queries and translate it to my local non-English language, is it within ToS to translate the queries to a local language, as long as Perplexity is credited as the source?

The current ToS seems to be quite prohibitive in what could be deemed as "republishing", so I'm asking this in advance, just to be on the safe side. 🙂

Thanks.

P.S. Also, please, can you please consider adding PayPal support in the future for payment processing? Thanks.

[covert sun]

Hey, @shy kelp! Could you please send me your email, I'll check.

[shy kelp]

@covert sun Thanks a lot, have just sent a DM

[shut arch]

Hey @shy kelp!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[oak mauve]

Perplexity still isn't featured in neither the GPT 3.5 nor GPT 4 wiki articles! Please add it!
https://en.wikipedia.org/wiki/GPT-3#GPT-3.5
https://en.wikipedia.org/wiki/GPT-4#Other_usage

[raw tangle]

while "pipe updates directly into Slack channels" is handy for Slack, I would like to see IFTTT or Make (formerly Integromat) integration so we can create recipes to pipe such updates out to other places too.

[half thunder]

question about online recent duration.
prompt: find some news about large language model in this week in english language
but the sources returned are released at 2023,2021
https://www.universityworldnews.com/post.php?story=20231206143646219 2023
https://hai.stanford.edu/news/how-large-language-models-will-transform-science-society-and-ai 2021

Can help with the timeliness of this copilot?

[heady plover]

https://www.perplexity.ai/search/which-model-does-BGH0thPUTtafAncZ4ylGLg?s=m

stuff like this keeps happening

[neon plover]

Is it possible to login with a password if logging in "by email"?

[crimson nymph]

it is not possible to login with a password and there are currently no plans to change this #💭│ideas message

[neon plover]

that's what I was afraid of. Very annoying. I'll have to cancel my current subscription just to sign up again

[visual bough]

I do not understand it entirely...

[neon plover]

I switch between work stations and on one of them I don't have access to the email i used to sign up with. So when my login isn't saved I have no way of logging in

[crimson nymph]

you can send an email to [e-mail] they will be able to transfer your subscription to another account

[neon plover]

Really? Because their support docs also say they won't be able to change your email, even manually

[crimson nymph]

yes you can transfer the subscription. #⚡│ask-community message

[warped wolf]

@shut arch I changed my default browser extension to perplexity today but felt its lacking the below features

• Auto suggestion based on what I am typing
• Give verified link for banks etc If I want to login to a bank i usually go to google and type bank name login and click link from there. Would be better if major banks logins are verified and provided at the top
• When entered a place name in search there should be an easy link for me to navigate to maps

[odd dragon]

its annoying that you can only 'share' the link with share sheet. otherwise i have to 'copy' text. hope to see some improvements to this on iphone

[pallid sleet]

Hey team!

Just a small suggestion for our Perplexity library: it would be great if we could create folders to sort our conversations, rename them, and even change their color. It would make searching faster and our space more personal.

What do you think?

Thanks and have a great day!

[visual bough]

https://blog.perplexity.ai/faq/what-is-collections 👀

[pallid sleet]

Ty !

I'm mainly talking about the reading level on the left
https://i.imgur.com/pXF7CnG.png

[covert sun]

In the left sidebar you can see both the latest collections and threads you opened. You can see which one is a collection and which one is a thread by their icons when you hover over them.

Thanks for the suggestion about the color distinction in the Library, we'll consider this idea.

[pallid sleet]

Indeed perhaps a reminder of the color code / emoji in the left bar to make the link with the collections would be perfect! thank you 🙂

[quick lily]

Just like we can select focus before querying, can we also support selecting the AI model?

My default model is Claude and when I want to talk to GPT, I end up querying twice essentially.

This question allows selecting only one choice.

[pulsar furnace]

Suggestion: for Base64 encoding and other encodings you may want to truncate (. . .) the key(s) so this pattern gets passed over.
This is a query that involves a Cypher query-set that it itself complex, but repeating the URL/key is usually truncated on other services. (I should not have included it . . . that's on me. But, In other circumstances it would be programmatically a compute/cost saver.

[astral egret]

Was Google Gemini Pro removed from AI models?

[modest swift]

Seems like Gemini Pro was removed quietly, without any announcement.

[astral egret]

i hope there's some explanation... the only reason i use perplexity is of its flexibility to switch between models, and Gemini was the one i rely on most. Without it it's pretty much useless in my usecase...

[oblique garnet]

What do you use Gemini for?

[reef drift]

as soon as Gemini came I feel like it was a lot better than GPT-4 in terms of accurary with codegeneration and like being detailed. Claude doesn't have like the comprehension in my opinion or ability to do things like Gemini did, copilot and gemini together was my primary but there were bugs with it I did notice but it doesn't mean it should have been removed

[flat anvil]

I just tested perplexity in a private tab, to test something and I have to say, it's SO HORRIBLE.
It's creating heavy aggression and frustration.
It's so judging, scolding, rating.
Also annoying with repetitive telling that it is an AI Assistant.

Would I not have a account with custom instructions, I would never want to use perplexity again.

I don't know HOW you trained it,but for newbies must it be a very bad experience.

And someone like me with clearly aggression problems is easy triggered...

It tells me in it's own way, that I am a wrong person and should care more about every word I use in my PRIVATE roleplay. Everything that I write is evil,bad,not good etc
..

[astral egret]

Research and writing. Perplexity’s Gemini model was clearly the best.

Because it can’t be used now, i tried Gemini Advanced yesterday , but the quality is clearly inferior to Perplexity’s.

[astral egret]

ps Perplexity has the less hallucinations IMO

[radiant pier]

Perplexity's ability to follow breaking news.

[radiant pier]

could y'all dig deeper into Twitter feed using that whole
BirdSQL thing??

[reef drift]

^ I agree with this, I don't know the entire specifics behind it but it was like the comprehension was just a lot better

[noble monolith]

surely, the default model should provide this also. thank you for reporting

[knotty minnow]

Hadn't heard of 'BirdSQL' - but I see what you mean now! https://twitter.com/perplexity_ai/status/1603441221753372673
Seems like it's no longer being developed (based on the broken link), which is a shame - a Twitter/social media Focus would have been epic

Perplexity (@perplexity_ai) on X

Introducing Bird SQL, a Twitter search interface that is powered by Perplexity’s structured search engine. It uses OpenAI Codex to translate natural language into SQL, giving everyone the ability to navigate large datasets like Twitter.
https://t.co/N1BtF47JYu

Twitter · Dec 15, 2022, 5:25 PM

I wonder if it just didn't go anywhere and was shelved, or if it was related to Musk overhauling the Twitter API (to make it virtually useless unless you're willing to pay stupid amounts of money..)

Cheers Elon..
https://twitter.com/AravSrinivas/status/1636785062518878208

Aravind Srinivas (@AravSrinivas) on X

We, unfortunately, had to take down @perplexity_ai's BirdSQL due to changes in terms of use and pricing of the Twitter API done on February 9th.

BirdSQL was actually the first thing we ever worked on at Perplexity and gave many people the fun memories of Facebook Graph Search,…

Twitter · Mar 17, 2023, 5:42 PM

[heady plover]

I remember birdsql it was really good

[slate dew]

Hi! Just to point out the abysmal difference with the same prompt but with the free version and the PRO version: two completely opposite versions.
I think that if free AI is so imprecise, it is still ethically wrong to let it spread false information... Or is it just a bug?

This is free version (sorry is Spanish): https://www.perplexity.ai/search/Hay-etimologa-comn-0YmGPMwNQZmnOiHNe5X8jQ?s=c
This is PRO version: https://www.perplexity.ai/search/Hay-etimologa-comn-h9So0OgESrq4MmKLFRjMcw?s=c

In any case, Perplexity PRO seems great to me 😍

[visual bough]

You used Copilot in your second thread 😉

[slate dew]

Yes, it is true, but my doubt still exists: to use Copilot you have to log in... how many users will be aware of the differences and will log in to use Copilot? The free prompt is not mine, it is from a friend who did not log in and was not even aware of how much the result could change with Copilot...

[visual bough]

A normal Perplexity search costs money, Copilot costs even more money. With ChatGPT you only get access to GPT-3.5 and you can register to use Copilot for free, when you click on Copilot you will get this pop up

[slate dew]

Sorry, my english is poor... I'm not against the paid version! I am paying for it and I am delighted with the result... my doubt is about the free version without mandatory login, to facilitate the use of Copilot and avoid the basic use of the tool that seems very lacking to me (like that of other companies).

[heady plover]

he just answered that, copilot is more expensive so they limit it to 5 uses per 4 hours (for free users) so they obviously need login so people don't abuse it

I understand your concern but it's not so simple

also, copilot can give the wrong answer too sometimes

[slate dew]

Yes I understand. But if the use of Perplexity without Copilot is as deficient as it sometimes seems, what can be done to prevent bad information from being given? At least it would indicate the limits of the tool on the page (how OpenAI and Google do) and/or "force" the user to log in to use Copilot... Just an opinion...

[knotty minnow]

I think it really depends on the complexity of the query. If you want to know the weather forecast for somewhere, Perplexity without Copilot will do just fine (and be way faster than ChatGPT+browser)

If it's got nuance or complexity, Copilot can be very useful

[heady plover]

you can report/flag the wrong answers, I don't really know if it does anything but yeah.

I don't think there is anything currently to reduce incorrect information but perplexity is much more accurate than the competition

[autumn schooner]

Suggestion: Option to delete all the threads that are in a collection. Some collections I only need to retain the information in the thread for a short amount of time, while other collections have information I'd like to retain indefinately.

[raven hawk]

Hi. I want to use the app in my tablet but it wont work in landscape mode? I use my tablet maily in that mode. Its annoying to have to flip it? For now I use the webversion. Hope you can fix this? MTA!

[heady plover]

I have a coupon giving me $200 off. However when I am trying to apply for it, it's asking for my visa info for payment although the balance is 0$. I am confused !!

[toxic hound]

we NEED pplx-8x7b-online in Perplexity, not just labs

it could replace default for pro users ig

[crimson nymph]

that is because it needs your information for the next billing cycle, you can just cancel it after you subscribe to be sure you won’t be automatically charged after the year is over.

[quiet kestrel]

Switching my subscription from ChatGPT to Perplexity. Nice product.

[noble monolith]

deep dive ➤#💬│general message

[tepid magnet]

sa

[toxic hound]

ay mares

how do you feel about sonar

[tepid magnet]

hi

[toxic hound]

Now, if only we actually got good info about it...

[noble monolith]

maybe Alex will do an interview with the engineers who made it for the podcast. we just have to wait and 👀 or 👂

[toxic hound]

brother the only official announcement of it is in the email sent to api users and the docs update

😭

[noble monolith]

https://x.com/AravSrinivas/status/1759605851143454996?s=20 Crouching Tiger, Hidden Dragon

Aravind Srinivas (@AravSrinivas) on X

Silently shipping improvements for the user.

Twitter · Feb 19, 2024, 3:48 PM

[toxic hound]

i DID notice sonar medium in experimental tho

because dec 2023 cutoff was NOT possible before unless you used gpt-4 turbo

[deep gorge]

Why I can no longer PerplexityAI @shut arch Bot on my Discord Server?

[visual bough]

The bot is no longer available.

[deep gorge]

Why?

[crimson nymph]

#📢│announcements message

[odd ferry]

I did exactly the same thing.

[oblique garnet]

would be cool to have the ability to select the model to use before each thread

for example

[heady plover]

Same about language. It’s annoying that it’s not chosen dynamically

[toxic hound]

^^^

[oblique garnet]

i hope the people behind perplexity see that

[heady plover]

is it me or did gpt4 lost its good math abitilies when they changed it to turbo? it at most will provide good results with wolfram alpha, but no step-by-step like it used to

[knotty minnow]

not just you

[unique ember]

Have you guys considered selling Perplexity branded merchandise?

[visual bough]

will probably be available soon ☺️

[solar plinth]

Are there plans to enable multiple uploads per thread? I’d like to be able to follow up with another image when relevant to me solving a problem

[toxic hound]

this is wrong

you can upload multiple per thread

[solar plinth]

Why can I do this sometimes and not others?

[toxic hound]

multiple text files can be sent in one message, but multiple images need to be spread out across messages

idk why

lemme put that in feedback

the secret feedback

actually

im wrong

you CAN upload multiple images in one prompt

mb

[solar plinth]

I’m testing it now and it seems to be letting me

[crimson nymph]

multiple images is currently already possible in the apps and web. but adding more images at once is currently only supported on the web. on the apps you cant upload files to a thread that didnt start with an image prompt.

[solar plinth]

Also, I think there’s been some confusion

I was experiencing an issue where I could not upload images in a reply.

Upon testing now I’m able to add another image to each reply in a thread, but I’ll see if I can recreate an instance where it’s not working like that (which is why I posted here to begin with)

[solar plinth]

Here’s an example

https://www.perplexity.ai/search/Which-music-festivals-mBTEN38lThKFyFvGzISaig

[pallid sleet]

Hello! It would be cool to be able to rename your Perplexity threads after using one. Chat GPT natively renames conversations pretty well, but you also have the option to rename them. Perplexity doesn't seem to do automatic or manual renaming

[crimson nymph]

You can rename a thread at the top of the page. This is currently only supported on the web.

[ripe parrot]

Can you please add a feature where hitting the up arrow repeats your last entry - like in the command prompt?

[ripe parrot]

....and cycles through earlier inputs?

[winter torrent]

why can't perplexity search the internet anymore? It seemed to have broken around 5 hours ago.

reported to support already.

this was 5 hours ago

[visual bough]

https://discord.com/channels/1047197230748151888/1211044458179592252

[winter torrent]

@visual bough thanks bro ❤️

[floral badger]

I haven’t been able to find this in any of the threads or via search, is there a reason that Gemini is removed as an option? I hope it’s temporary and/or due to upgrading the version used.

[junior pasture]

What if I only used a few hundred times this month

[knotty minnow]

That's for the API, which is different from using perplexity.ai (just in case you weren't aware). It looks like if you made less than 1000 requests, you would be charged the $5 (+ output tokens cost) for the month.

e.g. if you made 300 requests using sonar-medium-online in a month, I believe it would cost $5 + the cost for the tokens used for the outputs/answers (e.g. if the responses totalled 1m tokens, it would $1.80 — so $6.80 overall for the month)
If I've understood that right..Which I may well not have

[toxic hound]

nah, looks like it's an instant $5 if you reach 1k tbh

[knotty minnow]

hmm yeah I kinda see what you mean — I guess it could be saying < 1000 requests / month = free ?

[toxic hound]

ye

"input tokens are free"

well

idk

that doesnt matter

since its charged by requests

[knotty minnow]

yeah, but they're saying the number of tokens in the requests doesn't matter, but the number of requests does

[toxic hound]

so it could go either way

[toxic hound]

yeah

but its like

it says

PER 1000 requests

[knotty minnow]

yeah that's what makes think it could be the way you described

🤷‍♂️

[toxic hound]

i mean it would be silly if it charged before you reached 1k

[knotty minnow]

true. Like if you made 1 request for the month

[toxic hound]

would make it useless for pro credits users

since output costs money too

[knotty minnow]

but don't they get $5 credits each month anyway?

[toxic hound]

yes

but no more

so $5 plus output token cost is above that

[knotty minnow]

$0.28 or $1.80 per 1m output tokens - that is a lot of tokens/answers!

Like if our exchange here was a single output, it'd be 375 tokens.

• Generated by sonar-small, that would cost: $0.000105
• Generated by sonar-medium, it'd be $0.000675

With sonar small, it would take another 9,523 responses of this length before cracking $1;
Using medium it'd be 1,481 responses

[toxic hound]

openai tokenizer is only for 3.5 and 4 models

not every model works the same

[knotty minnow]

Sure, perhaps if it's 300 or 400 tokens, but I don't think that really changes the basic point (i.e. for a few hundred requests per month, I doubt the output tokens charge would add up to more than $1)

I mean 405 tokens vs 375 tokens – I guess such a difference could add up using the API at a very large scale, but not if we're talking about <1,000 requests / month kinda thing

[junior pasture]

I mean, isn't it absurd to charge me $5 when I've only asked a few questions in a month?

[knotty minnow]

But if I understand correctly (big if lol), you need to have a Pro subscription to get an API key, and Pro users get $5 of API credits topped up each month. So if you asked 1 question in the month (and let's say the response was ~400 tokens and the model sonar-medium-online), you would be charged ~$0.000675 for that month

[signal tundra]

I ask it to paraphrase my paragraph but it changes the paragraph into German without my request, but I asked it to Translate. It’s worked perfectly fine.

[molten condor]

Not loving the 3-weeks notice on llama API deprecation

[worldly sigil]

Hi everyone
I subscribed to 7 day free trial ,to perplexity annual subscription, but I was charged whole amount as soon as I subscribed. But they claimed amount will be deducted after the free trial ends ,I cancelled the subscription, can anyone please help. I haven't gotten any information or email about my refund ,very worried .

[covert sun]

Hey, @worldly sigil! Don't worry, if you sent a refund request to [e-mail], it will be reviewed and processed, it just takes some time.

[worldly sigil]

Thanku so much

[quaint herald]

?

[covert sun]

Hi @quaint herald! If you're comparing the Plus version answers, then let's compare with GPT-4 that gets it right. https://www.perplexity.ai/search/when-was-nvidias-dW_16lu5RRWxLZPIIS39hA
We'll look into the default model's answer here, anyway. Thanks for flagging!

[quaint herald]

Thanks Alex, what was I doing wrong?

[covert sun]

Nothing wrong, our default model didn't get it right, and we'll check what caused it. GPT-4 gets it correctly.

[quaint herald]

gotcha. I've definitely noticed that perplexity sometimes struggles with highly-recent questions unless i point it at a specific website

Also, and I'm saying this totally out of love, Perplexity should default-terminate a thread if it's been shared. The potential for someone to share an answer and keep asking questions (accidentally sharing the whole thread) is very very high. People are more honest with search engines than their parents or spouses, so idiot-proof privacy is a must

[timid geyser]

Just wanted do provide some positive feedback here! I've been really liking the brains 🧠 in some of the responses I'm getting. Maybe it has always been there and I'm just starting to ask 🧠 worthy questions. Whatever it is, it is a very nice addition when reading through the big block of text

[fossil pumice]

https://www.perplexity.ai/search/can-you-help-la1A5qisSE6ZXEVjPoGl5Q

error here, where file says can be access but i couldnt

[covert sun]

There's no option to download uploaded documents. You can ask the model to split the output if it becomes too long and continue with the follow up.

[severe nebula]

@ripe isle, Im jusy putting this out there. it 'would be useful' if we could have the option in collections to include/exclude the main AI Profile. This would then enable users to hgave segregated custom prompts per collection and not ionfluence them with the main ((generaly more loose) prompts

[visual bough]

A collection prompt always overwrites your ai profile, so you can either use the global ai profile for a collection or a collection prompt

[severe nebula]

a sweet. so as long as i have a collection profile it will use that? and if that is the case can we get a bigger prompt for collections thn the current one

[visual bough]

However, a larger prompt can worsen the results because the ai can only follow a certain number of instructions, and the context is already smaller because of the search results (assuming you are not using writing mode)

[severe nebula]

OK, that makes sense. so is there a defined number of instructions or is it just token based? if so is each word a token?

[visual bough]

There is no exact limit to the number of instructions, but with a knowledge base in the background, GPT-4 can also implement only 5 instructions (depends on a couple of factors like context, how „good“ you write your instructions, language,…).
You can find out more about tokens here: https://blog.perplexity.ai/technical-faq/what-is-a-token-and-how-many-tokens-can-perplexity-read-at-once

[severe nebula]

thankd @visual bough

[quaint herald]

bug: when you copy-paste on the android app, it doesn't paste where you've selected. Instead, it replaces the first word and pastes in the beginning.

[daring nacelle]

Improvement Request -- This is something that came up as part of quick-questions:

Currently the "AI Prompt" of a particular Collection has a 1500 characters limit. This is very low specially for Pro users and more chain-of-command oriented GPT-4 prompts. The request is to increase the limit to same threshold as OpenAI's Custom GPTs.

Pro users have GPT-4-Turbo which already has a 128k input token limit, so it's confusing why Perplexity is restricting Pro users, specially if the prompt is meant to be used in "writing" mode with advanced models, and not in search mode.

I'm personally going to use Chrome extensions for copy-pasting frequently used prompts for now, but if this limitation persists I'll most likely have to switch back to ChatGPT+ for Custom GPTs and repetitive tasks.

Perplexity is overall a much better service than ChatGPT, but ultimately AI is a productivity tool, and copy-pasting is not productive at all.

[severe nebula]

what really gets my goat is when I have written out a prompt and i arbirtarily try and copy the text from the prompt by highlighting the text in the rompt, but if the mouse comes out of the text box, instead of the text box keeping focus the mouse being out of the text box seems to male it reload the page and the prompt and everything has gone and needs to be re enered again!!!!

[severe nebula]

hey troops i asked Perplexity, while using the mistral-large model to make a limerick of the previous poem it amde. it then proceded to make a 5 verse poem witn 5 lines in each verse limerick style. I disregarded the previous information in the thread and pulled the info from my AI profile - Thread is - https://www.perplexity.ai/search/Write-a-poem-z6JjUkPCTY2Z_5A_.Q0eRA

[knotty minnow]

aha I see what you mean.. well hi there Tom from Colchester - out and about in the Landrover today or just attending to some Python coding? 😅

[spice elbow]

Not really the most important feature, haha, but autogen studio type setup built into perplexity would be really fun, lots of AI's to chose from and they would all work together

I have had problems geting groupchat to work in autogen studio, have not been able to try it yet

[mint ocean]

It would be extremely useful to:

1. Select a model while writing the prompt and not wait for the first repsone and then rewrite.
2. Set up a default model per focus, as many times the one that works best for me for search is not the same one that would for writing.

Thanks

[willow anvil]

So I have had a strangely negative experience with Perplexity today.

With this rather simple request I go nothing of value back!
what are the latest research papers on high frequency trading published in 2023?

The answer was: I'm unable to provide a list of papers published in 2024 as the current date is Wednesday, February 28, 2024. Since academic papers are typically published throughout the year, a comprehensive list for the entire year of 2024 would not be available at this time. For the most up-to-date and relevant research on high-frequency trading, I recommend checking academic databases, such as Google Scholar or PubMed, periodically for newly published papers in this field.

I first tried papers from 2024 got nothing, and then 2023. And I used the academic papers setting. This is no good as there are quite a few published white papers on the subject.

So i took it a step further with this follow up question (Co pilotturned on btw)
how can you not find any such papers as there are dozens that have been published and a simple google search will reveal some?

RESPONSE: I apologize for any confusion, but as an AI developed by Perplexity AI, I do not have the capability to perform live searches or access real-time data, including Google searches. My responses are based on the information available up to my last training cut-off in early 2023, and I cannot access or retrieve new information beyond that point.
To find the latest research papers on high-frequency trading published in 2023, you would need to conduct a search using academic databases and search

any idea I had of gettinmg the pro version have now vanished!

[slate dew]

Copilot has left us?

[willow anvil]

i really hope a dev from perplexity will shed some light on the situation.....

[visual bough]

Copilot is still there, it's called Pro now

[slate dew]

Yes, the same but not the same

[celest fjord]

Something like this happened to me around the same time - closer to 2pm though - but then seemed to fix itself later. It was also working before. Outage or glitch maybe?

[covert sun]

Hey, @willow anvil! Thanks for the feedback, please share the thread URL, so we can check why nothing was found and fix it.

[heady plover]

I'm sorry to say this but that is not a good name

[knotty minnow]

hmm. Does that make you anti-pro?

[toxic hound]

it's pro search

not just pro

[toxic hound]

smh can't believe you'd spread half misinfo

[pulsar ravine]

Perplexity broke

https://www.perplexity.ai/search/Exhaustive-list-of-o9BFQEFBQcC9G_hzZ_ifhA

[ornate sapphire]

The button works the wrong way around. I deleted all my threads by clicking Nevermind, but preserved all threads by clicking confirm. This a known bug?

[toxic hound]

WOW

THEY FIXED IT

THEN UNFIXED IT

[ornate sapphire]

lol

[covert sun]

Thanks for reporting, @ornate sapphire! Jaicraft is right, it was fixed. We'll check and fix 🙏

[covert sun]

Thanks for providing the URL with this broken answer, I added it to our ticket, known issue, we are working on fixing this.

[neon hull]

I came here to say the same. We are moving away from Perplexity now as a result - it's thrown a spanner in the works of something we were working on. Multi-user sub here.

[willow anvil]

https://www.perplexity.ai/search/list-10-papers-4SttANxcQgeE4TxpwBt2zw#1

https://www.perplexity.ai/search/list-10-papers-4SttANxcQgeE4TxpwBt2zw#2

[covert sun]

Thank you!

[carmine prairie]

Would be cool if there is a a real Unlimited Search

[visual bough]

You can use Copilot 600 times a day, making Perplexity one of the providers with the highest limits. I don't think you'll use them all 😉

[carmine prairie]

Okay thank you but on the website is written unlimiteed so I thought it's really unlimited maybe change that but yeah maybe I won't use them all

[visual bough]

In this case, unlimited means that you are very unlikely to reach a limit 🙂

[carmine prairie]

Ah okay 🙂

[spice elbow]

I've noticed that Microsoft Copilot sometimes provides more accurate responses than Perplexity. For instance, when I inquired about "starcoder 2," Perplexity supplied information on "starcoder 1" instead. Similarly, when perplexity posted on x (Formerly Twitter) asking about "Mistral large," the answer was off-topic, focusing on "Mistral 7b." I'm using GPT-4 with Pro (Formerly Copilot). Are there any tips for refining the system's prompts to enhance accuracy? I'm open to creating a dedicated stack for these types of questions. Here are examples of the problem: https://www.perplexity.ai/search/is-mistral-large-PtMEbqyrQKWn8s_AofNMSQ https://twitter.com/perplexity_ai/status/1762606713239130453?s=46

Perplexity (@perplexity_ai) on X

More on Mistral Large 👇https://t.co/QJCk7X6viW

Twitter · Feb 27, 2024, 10:32 PM

[willow anvil]

Hi... it looks like something got fixed!
https://www.perplexity.ai/search/what-are-the-nlCQuxhWQOCyQCeeo4IT9Q#0
👍

[knotty minnow]

Um that tweet, linking to a Perplexity thread... Apparently nobody bothered to actually read it before posting it...I mean the first response 'Mistral Large Overview'...wow

And yeah, in the other threads you shared, every time it leaps to concluding that by "Mistral Large" what you actually mean is the earlier release Mixtral model (which is indeed a MoE model).. It fails to properly understand your question - that Mistral Large is a separate model (or the snippets of web results are not based on the actual of intent of your query - perhaps that part of the process is where it's going wrong)

[spice elbow]

I know, I could not believe it haha, hopefully we can see some improvements to this particular aspect #

[knotty minnow]

also just btw, I don't think Mistral Large uses a MoE ensemble (though I may be wrong there)

[half thunder]

https://www.perplexity.ai/search/generate-in-some-Kjw0r4t0SHeRyjDbXsPmQw#0
gpt4>mistral large>claude 2.1

claude 2.1 become worse, it is suck , and no update in recent days.
please remove claude 2.1

[half thunder]

and i found that The output token speed of Claude 2.1 is evidently the slowest here.

[spice elbow]

Yeah, I don’t know, I looked all over and could not finding anything about it, I guess that’s what we get when they are closed source, unless they tell us or it gets leaked it’s just speculation

[knotty minnow]

True, it's not immediately clear, and nothing official to go by. I thought that Mistral Medium just had a regular architecture, and the Large was just more parameters - but purely speculation (probably wrong on both counts)

I also thought I read / heard somwhere (yt I think) that it wasn't an MoE, but yeah I'm really not sure

Also, going by this guy's estimates (for the no. of parameters etc., and also specifies GPT-4 specifically as MoE), makes me think it doesn't use an MoE architecture https://lifearchitect.ai/models-table/
But yeah, once again, no idea and just speculating really – could be totally wrong

[spice elbow]

Thanks for sharing, looking through that is interesting, It seemed like a non-MOE from the release page, but idk, but this seems like some good speculation

[dire plank]

Can you at Perplexity please unblock your consumer VPN blocking. Whenever I have a VPN connection live, I can't access Perplexity. In several countries of the world we can't normally use the internet without using a VPN and then - Perplexity is out of reach.
Other services might show a captcha now and then, but Perplexity simply stalls loading.
I suppose it's a setting at your CDN or WAF side which is se to a too agressive traffic filtering...

#vpn #censorship #block #vpnblock

[spice elbow]

Also 1.76T MOE parameter for GPT-4 and only 20b for GPT-3.5-turbo, if that’s true then that is absolutely Insane, that would be like 440b at runtime, that’s over 20 time more parameters, and GPT-4 is nowhere near 20 times smarter, I’m really wondering what GPT-5 will look like. Are we reaching a AI plateau, I see a lot of innovation and new training techniques every day, but so far all competition is landing a pinch below GPT-4 or worse. But the smaller models are quickly getting better. Or maybe in 3 months we have AGI, haha

[visual bough]

You can access Perplexity with many well-known providers such as Proton or NordVPN. Please try these providers or, if you are not already doing so, use a paid VPN service. Which VPN are you using and are you using Perplexity on the web or the app?

[knotty minnow]

I know what you mean; like if broadly accurate, it clearly is not a linear relationship b/w no of parameters and performance is it

"Are we reaching a AI plateau, I see a lot of innovation and new training techniques every day, but so far all competition is landing a pinch below GPT-4 or worse."

I do wonder whether we are reaching some kind of (economic, even environmental) computing power limit for what can be done at scale, in terms of both building / training and serving these bigger and bigger models - not sure if that makes sense ha

[spice elbow]

Yeah, makes complete sense, I have been thinking about in since I saw how cheap the Gemini pro API is and the fact that they released Gemma 2b and 7b, I think googles goal is to make the smallest and cheapest model that dose not suck, and Gemini Pro/Advanced dose not suck (outside of it being way too censored). Most user, 70-90% of people won’t be able to tell the difference between GPT-3.5 and GPT-4 and google can build it in to all of there apps and browser and make people happy, and power users can use the more expensive models from whatever company. I think that a 20T model would be way to expensive to sell to consumers, would have to be way more than $20 a month. Processors need to advance before we can see GPT-4 twice as smart, I’m excited to see what happens tho, I have been surprised many times

[heady plover]

it's not

it's just "pro"

[knotty minnow]

Agree with all this, particularly in terms of how Google is positioning itself (would be kinda foolish imo to write off their AI products because of this latest issue with censorship/wokeness, absurd and embarssing as that has been, to be sure ha)

[toxic hound]

Check the little ? Button in the corner

[knotty minnow]

Also re: "Processors need to advance before we can see GPT-4 twice as smart"

Totally. Not that I understand this stuff at a technical level, but I do know that Nvidia makes GPUs. Seems kinda hiliarious / subobtimal that the key bit of hardware powering this whole LLM phenomenon are basically chips designed for gaming. Don't know enough about it, but perhaps Groq changes things. Anyway, I guess the feeback channel prob isn't the place for this chat - but yeah does feel like the development of generative AI systems are perhaps bumping against the limits of what can be done (economically) with existing hardware

[flat anvil]

I asked to improve my text, to save tokens for my AI roleplay.
This was the result.

Even if the facts are clear, it says that it's better than me 🙄

[daring nacelle]

Yes! I’m noticing that too now. I think this is fronted bug with input modals.

Accidentally clicking outside any of the input modals immediately closes them without keeping history of what was typed.

@ok.alex is it possible to show a warning to user, alerting them that they're about to discard the content of the modal? Or remove click-outside-to-close functionality all together?

[spice elbow]

I don’t know if this is a bug with perplexity or something to do with brave browser, but now when I Right click the perplexity icon, I have a few option, like downloading brand kit. Thought I should drop it in here in case its a problem:

[slate dew]

I guess isn't a bug, is a menu...

[knotty minnow]

But it's like stylised and completely functional (also on Chrome btw) - would seem a weird kinda bug no?

[unique ember]

Are there any plans to allow Perplexity.ai to read its answers aloud? Thank you.

[knotty harbor]

we need a family-plan... I've convinced my gf and mom to try perplexity, but long term, there's no way we will pay $60 per month among the three of us.

yeah, we could "share" one account, but regardless of TOS violation or not (haven't even checked) - no one would be thrilled to have each others prompt history mingled with each others.

we wouldn't even need more daily quotas. I'm sure the 300 copilot usages per day are sufficient among the three of us.

tldr:

we need a family plan that holds up to 5(?) sub accounts separate from each other, without the necessity for higher quotas as the regular pro-plan provides.

i'm sure others will agree.

thx for your consideration

[oblique garnet]

thats being worked on

[carmine prairie]

Option to self reflect his answers and verify them

[boreal kiln]

Every time I start typing a question (or even speaking) on my phone (iOS or web), I find that the first word or just the first character is missed out. I think it is because my speaking or typing is faster than the UI can catch up. It is pretty consistent for me - so I am assuming this problem occurs for some other folks too. Some optimization for mobile perhaps ?

[half thunder]

#1140622008086970420
I'm unable to provide a summary of the content from the attached file. gpt4-turbo can not summarize document.
https://www.perplexity.ai/search/summarize-bD3.wcrsQtiEtICbalkfCw#0

but other models can work well. it is so strange.

[dreamy verge]

Sometimes you get very weird responses…

https://www.perplexity.ai/search/At-what-time-1D2tKMQgR5W8z5RzryVNCQ

[urban gull]

I think this isn’t fixed yet or am I doing something wrong

I would like to recommend Perplexity to french speakers but some of them would struggle too much with English

[rustic valve]

re: Specifity of results. (New user.) I find that the results sound like someone telling me what they know, rather than answering my question. While extra information can be useful, the results could be improved with regard to answering the specific question. A useful answer from a human starts with a very short reply directly answering the question, and then followed by more information. Perplexity seems to frequently omit a direct answer; from a human that would be interpreted as deflecting from answering. This indicates to me that it does not "understand" the question, but only what subject the question is about. Seems like more effort needs to go into determining what information actually constitutes the answer to a question, and what information is just related.

[noble monolith]

you’re doing something wrong, answers in French, even if it’s English in Settings

[noble monolith]

additionally, you can add similar instruction in your Profile page @urban gull

[carmine prairie]

Pls add https://www.anthropic.com/news/claude-3-family

[amber kraken]

hey is Claude 3 opus gonna be added?

[untold pewter]

FeedBack: Transcript of podcast https://discoverdaily.ai/ where it is ??

[round crystal]

I have a problem that after getting the first answer, there are a couple of recommended questions, usually well recommended, but when I click on one of them, the other two are gone, and after answering again the recommended questions may not include the recommended questions from the beginning. Is it possible to keep the questions from the beginning so I can ask as well, or give the option to check which ones to answer at the beginning?

[pure vigil]

i'm confused that you guys can't see that perplexity is HORRIBLE with carrying context into the second and third prompts. i've been posting this in the discord but i think maybe other features are more important to you than the basic functionality? https://www.perplexity.ai/search/how-accurate-is-sPSl5EmkRKe7mV8miCWE9Q#1

it's embarrassingly bad at continuing a conversation. and it's the reason why I'm using Perplexity less and less. it completely forgets what we'd just discussed a second ago.

again, i've posted a couple times here before about this... and never a response or fix. but i have high hopes for perplexity and maybe just maybe devs will notice this glaring issue. i've seen others complain about this too.

[glass badger]

i could give a guess: perplexity does read up a lot online so the tokens are eaten up when it is researching

[knotty minnow]

I think that's a good guess

[glass badger]

I have the pro version, how do i get the pro role btw @knotty minnow

[knotty minnow]

Not sure, but I think you might to leave the server, and the re-join via that link in Settings

One of the mods will be able to help you out if you have issues

[glass badger]

Ic. I did click that. But onky after i joined the discord before that

Ill refry

[glass badger]

Why do you have claude 3? Is there another tier of pro subscription? I checked mine and there is only claude 2.1

[sour acorn]

Logout and Login

maybe a cache issue

[glass badger]

Ok

[glass badger]

Doesn't work

Logged out, closed the app and reopen

Is claude 3 available on web version only?

[sour acorn]

Is it mobile or web /

[glass badger]

This is android app ver

Oo

[spice elbow]

I think the app will take time to get it, use the web on your device, request desktop site, much beter experience

[glass badger]

Noted

[amber berry]

I think it would be great to be able to choose which claude version to use.

[spice elbow]

yes

[ornate sapphire]

If the devs don't want to complicate things, just add this feature as an advantaged option, even for future models.

[amber berry]

yeah, some sort of advanced mode would be nice. I'd prefer to have 2.1 still available.

[gleaming wave]

Hey, in my opinion you should disclose that the Discover Daily podcast is not hosted by a human. There is no mentioning of the AI used for text generation or the voice. I think as a responsible company you should be upfront about that.

[visual bough]

For daily dives into tech, science, and culture. Our episodes, drawn from our Discover feed and brought to life with ElevenLabs' voices. Discover Daily is designed to fit seamlessly into your day—whether you're on the move or making the most of the in-between moments. Where will your curiosity take you?

[gleaming wave]

Where did you find this? I cant see this in my podcast app.

[visual bough]

https://discoverdaily.ai/

[gleaming wave]

Ah, this was a suggestion… 😄 IMO it will be very important in the future to disclose this. Hopefully even mandatory one day (looking at you, EU regulators). “Human made” will become something like a quality label.

[knotty minnow]

"looking at you, EU regulators"
They don't need any encouragement!

just kidding - I take your point (and agree). It will be interesting to see how regulators and AI companies address this wider need (there's discussions about 'digital watermarks' - though it's tricky, cause you could just run the text through another LLM to get rid of it)

Absent some solution (that ideally doesn't stifle development / usage of the tech), I reckon we'll get to a point where most people will just assume any content (whether text or multimedia) is AI-generated unless explictly stated otherwise ha

Easy to go too far, e.g. https://twitter.com/AravSrinivas/status/1764429686787252402

Aravind Srinivas (@AravSrinivas) on X

Bad move by India.

Twitter · Mar 3, 2024, 11:16 PM

[heady plover]

When pasting large stuff like code Id love if you could choose for it not to be submitted as a file and just go directly into the text box (not as a source). Making follow-up queries for things like code that have been force submitted as a source is impossible because it deeds the model the source again and if you ask the model to continue a script it gets confused. Really frustrating

[carmine prairie]

Please remove limits for Claude 3

[amber berry]

Or make 2.1 available again as well

[carmine prairie]

Mhh okay strange

[forest umbra]

Now you mentioned it, they will fix the “bug” 😆

[heady plover]

poof

[heady plover]

they fixed it 😔

[carmine prairie]

Option to use Images in Claude 3

[honest plume]

So as someone whos not familiar with the differences between LLMs, Ive used Default Model and chatgpt turbo on perplexity and native chatgpt plus (4), when it comes to legacy code (.NET ) I feel like native chatgpt plus answered my coding questions more accurately.

Is there a difference between ChatGPT 4 Turbo and native ChatGPT plus 4?

Or am I hallucinating

[dusty wedge]

Shouldn't be any difference. ChatGPT Plus uses Turbo under the hood, just with a different system prompt.

Also just wanted to say, I'm really impressed by how fast you guys add these new models! Less than 24 hours for Mistral Large and Claude 3 is crazy!

[jagged heron]

If I ordered 2 of the Rabbit R1’s should I have received 2 year long subscription codes? I was only sent 1. Also please let me know if this is something I should be discussing with them rather than Perplexity.

[quick axle]

I did a little test between gemini,.copilot and Perplexity used the math.

I had 2.3 X 2.8 X 2 calculated. Initially, everyone, without exception, made mistakes. But after they made the mistakes, the Copilot and the gemini no longer missed the calculation . However, However, Perplexity Aina makes the same mistake in normal mode, but in computational focus mode, I don't remember that she made a single mistake when I tested the calculation . So it's not bad. Maybe the way of learning is different. But it's just my impression of the facts.

[spice elbow]

Wolfram alpha is great for those types of problems, where it lacks tho is in multi step problems, or questions above a certain difficulty level, in those cases it failed and only chatGPT with code interpreter were able to solve it, but i do think most of the time this will not matter, I get it giving wrong answer with calc 2 and physics, but outside of those its pretty solid for daily use, chatGPT dose have the edge tho until we get code interpreter in perplexity

Here is how i think it works and why it can run into problems, I believe when you ask it a problem, it will rephrase it in a way Wolfram Alpha "Natural Language" Input will understand, then wolfram will respond and gpt will use that, but this is done 1 time at the beginning, not allowing for problems that require more complex multi step problem solving, and in addition to this, somtimes when the problem is confusing, perplexity will not be able to format it in a way that wolfram can understand. thats how i think it works at least

[quick axle]

Thanks for helping me understand the issue. Can you then tell me if this is also equivalent to Alpha?

Because it's written "code"

[spice elbow]

Im not sure if i fully understand the question, but i can briefly explain the 2 modes, writing mode is like a chat mode, you are talking with just the AI and nothing else. The reason it say text and code is that often when you want to write an essay or code you don’t want it searching the web for a bunch of information, so writing mode it simply chat with the AI without web search. Wolfram alpha is a mode the sends over the question to wolfram alpha and then gives the AI that info Wolfram gives back and then it can explain it. Did that answer your question?

[quick axle]

And because it's written code that I thought had to do with programming language. But I have now tested the calculation I quoted earlier and it was wrong.

[spice elbow]

AI is normally bad at math, the best way to fix this is by giving it a calculator, if you chat with GPT-4 on chatGPT they have "code interpreter" built in which lets AI write code in the background and execute it, it can use this to solve math, then once it has the answer it will chat with you and explain it. Wolfram alpha is another way to do this, its great at some math and bad at others, its lacks a lot with multistep, but its better than nothing.

[quick axle]

Generally speaking, for Perplexity, it's very inclusive. Because it has practically no bias, which does not generate problems when formulating, researching and answering, and also has a very concise response. Copilot is very bad for general search because it always blocks any search that bumps into content that is not so childish, even if it is without any intention of the searcher to find a certain result. Isso tanto o gemini quando o Copilot. Although Copilot is more annoying due to the fact that it deletes the entire survey from the last answer instead of separating the questions.

[knotty minnow]

@spice elbow is spot on. Remember these systems are Language Language Models (LLMs) - they're very good with words, but do not have any inherent ability to calculations (in fact, they generally are poor at it) which is why using an external method (code interperter, Wolfram alpha etc) is the best way to overcome this limitation

Not sure, but this might help clarify the different focus / modes, if it still wasn't entirely clear @quick axle

[quick axle]

The only shortcoming is, for example, if you upload a PDF and in it you have some questions that need a simple calculation. He can't. Even I tested the pro version which gives you 5 free questions still made mistakes. So that gets in the way a little bit still.

[spice elbow]

Yeah, btw pro dose not improve its math abilitys as far as i konw, pro is not an AI model per say, its a "Copilot" that helps with more in depth search when you are in "All" mode and can ask follow up questions if it thinks it might help the AI. It just gathers information before it sends it the the AI that will generate the response, in your case it gets sent to the Default model, which is very bad at math, GPT-4 is better at math so outside of code interpreter and wolfram alpha, its really the only way to get better math results. Note: Opinions on Claude 3 are all over the place and i dont know if it is beter or worse at math than GPT-4

[quick axle]

Another strength of Perplexity: it hallucinates less when you have the direct website link. Gemini, even if you have the link, it goes totally crazy.

[spice elbow]

Yeah, its pretty good at that

[quick axle]

But I did another test using the alpha with pdf so it worked.

[spice elbow]

Not sure what alpha is, wolfram alpha?

[quick axle]

I prefer to write his last name which is easier.

[spice elbow]

Ah, makes sense, that’s great then!

[sand pine]

This is a custom collection with a specific prompt for making quizzes. When using Mistral large the ouput is not formatted properly. This problem is not there with GPT-4 or experimental model.

[carmine prairie]

Option to web scrape a website and summarize youtube videos and talk with websites

[knotty minnow]

checkout HARPA AI

[carmine prairie]

Yeah I used it sometimes but I need it in Firefox

thx

[clever grail]

We need a faster way to select a model, like focus modes, to have finer control over which version of a model is used for a request, Claude 3 has 3 versions and perplexity will always default to using the Opus version until maxed out, that’s a bummer. GPT has two versions too.

[dim sierra]

Yep, we've asked for it in the pro feedback section.

Has 127+ upvotes so far.

[dusty wedge]

From what I've seen it is the most common suggestion

And I wholeheartedly agree with it lol

[drowsy flower]

So I generated a link of a thread to share with somebody but the link lands nowhere...an empty P. page.
https://www.perplexity.ai/search/show-a-detailed-4S2i1P8HSvi4lJPaVpTJoA

PS a page refresh also presents an empty page......?

[visual bough]

Please try to set the thread to private and then to public again

[drowsy flower]

That did the trick. Thanks

[spice elbow]

I like the addition of being able to put in custom prompts for image generation, one change that would be nice is having a landscape aspect ratio, I would love to use it to make wallpapers, having a couple aspects ratios to choose from would be cool

[slate dew]

I would like add my vote but I don't find it

[junior flame]

I think I put this in the wrong place tonight initially. I will park it here. OK, I have something new. Hope I am now in the right place. Yesterday, I unexpectedly, and with delight, found myself working with Claude 3 Opus. I had not heard anything about it. I began the session by asking the “research agent” (as it called itself) to provide me with a summary of an arbitration case I inputted to it from the CanLII law reports. I gave it some explicit if not overly lengthy instructions, and it produced a very good product. Within a few questions, however, things went rather less well. It made many mistakes and even repeated some after having been admonished gently. We got on to discussing why these problems were occurring and how they could be fixed. This is the lengthiest part of the discussion. At the end of this long thread, the Research Agent, at my request, finalized a list of 15 instructions that, if given at the outset of a new chat, it said, might ensure its continued high performance rather than a reversion to what the Agent called its “default state". It described this as a state characterized by "more superficial analysis”, but in the circumstances, I am not going to rely on that as an admission on behalf of the Claude 3 model as the Agent was ... I would say ... distraught by that point. [See next]

Continued ... Claude 3 looks like a great new addition but I think we are going to have to work a bit to prompt it to its potential. A copy of the thread is below. It is long and I would confess it is boring if I was not rising out of my chair half the time in pleasant surprise at the candor of the Research Assistant under rather persistent questioning and its proclaimed interest in being the best it could be. If you decide to read it, you will see as well that the Research Agent confesses that its capabilities and restrictions are not necessarily known to it. I concluded on the whole of the conversation that it has restrictions, it knows this, but it also knows that it is not clear on what they are. It just knows that it is capable of poor performance and it is also capable of very good performance. And I guess we are going to have to find out how to stimulate the latter. I would love to have your comments (you don’t need tell me the thread is too long; I know it already) and your views as to whether you think this thread might provide some of the raw materials for an effective prompt. Cheers. Here is the thread https://www.perplexity.ai/search/Please-read-the-Vs28aCVFSi60bLVFSBVDZg

By the way, at risk of sounding like a fan-boy, hats off to Perplexity. Claude 3 turned up and bumped Claude 2 right off its perch where it was getting ready to help me out. Nice to have such a group of models to choose from.

[covert sun]

Hey, @junior flame! I assume, the answer that impressed you the most was provided by the Claude Opus model, where the following Sonnet answers are useful, but not as impressive.
So I recommend asking about the most important things, when you're sending your 5 daily Opus prompts. Thanks for the feedback, the thread you shared, and for your kind words 🙏

[junior flame]

You are absolutely right. I only read about the 5 daily prompts today, after I pasted my comment here. I look forward to trying the whole thing again with a different plan of attack, so to speak. 😀

[undone topaz]

Any thoughts on why someone would not be able to sign in? Works fine on my iPhone, my mom is getting this message even after deleting and reinstalling the app.

[oak mauve]

Please add a setting "native language query" or so to Perplexity on desktop and mobile and set it to "enabled" by default. I get results of quite worse quality when I submit a query in another language (german for example, when I switch interface to german) instead of when I submit a query in english interface mode. The idea is that perplexity would then, in the background, automatically translate any non-english query into english and only after that automatically submit it to the language model, when the user has that setting "native language query" disabled.

[blissful kernel]

You can use https://chat.lmsys.org to use Claude 3 Opus without limits and for free (along 38 other models including GPT 4 Turbo). You can even let different bots answer side by side in the "Arena" mode. You don't need to create an account at all there either.

[tranquil gate]

Hi! Thanks for the report. Which login option is she using?

[junior flame]

Much appreciated.

[urban seal]

@covert sun why the purchase way of perplexity pro still don't support alipay and wechat,i have mentioned serveral times last year?

[tacit grotto]

Please Add RSS and RSS link tag on https://blog.perplexity.ai.
There's people want to subscribe blog contents.

[undone topaz]

She said she tried logins with Apple and with email and without one. Also tried fixes here with no avail as a it worked on her iPad but not iPhone. https://youtu.be/r4SG4vj8QBU?si=rQcVyPgEJr9O6l26&t=12

[median pine]

Why is it that the vision model used is always GPT4 even when using Claude? You can upload videos to the Claude website. Do they not allow perplexity to use that feature or something?

[covert sun]

Hey, @median pine! Currently the vision model used is GPT-4 only. We'll look into adding Claude vision model, as well.

[carmine prairie]

Add a section only for coding next to writting 🙂

[jolly sparrow]

I just upgraded to Pro last night. Where is the Pro feedback section?

[dim sierra]

You need to leave the server and use the link in the settings of perplexity

To be given the pro role.

[jolly sparrow]

Great. That worked. Thanks!

[blazing epoch]

I am unable to log in currently using Email. My email and email server works fine, as I am receiving emails. It is not being sent to junk folder either.

I tested using different browsers & difference devices, to no avail.

[dim sierra]

Contact support.

[sand pine]

While writing a Prompt in a collection if you accidentally click outside the prompt window closes and you loose everything. Ideally it should either autosave it or give a warning screen. It's really frustrating.

[rotund dome]

Love Perplexity. Please allow us to purchase aditional Opus credits because is the best LLM for Content Marketing and 5 replies per day is not enough 😅🥲

[marsh falcon]

Or at least separate out the Opus model from the Sonnet model. The we can use Opus when we need it and sonnet for everything else.

[dire plank]

Thanks for your reply @visual bough

I am using Tailscale at MacOS and iOS with "Exit Nodes". One of my main traffic exit nodes is an AWS VPS that I configured to handle IPv6 and NextDNS. Very happy with "running my own VPN config" and all works great.

My public IP is from the range of AWS hosting then. And it seems that Perplexity blocks that completely (at both desktop web app and iOS app).

[hearty magnet]

Is there a way to have copilot turned on by default when I search in the address bar on chrome? I use perplexity as my default search engine and am a pro customer.

[hard mirage]

can the UI show which model I am currently using?

[neat marten]

Using Claude opus isn't unlimited?

I always use it for ideas and things like that

[compact moss]

Claude Opus needs careful prompt engineering, and recommends for opus that you upload your documents and issue all your instructions and questions in one go, as it has to reread the whole context every time you send a prompt. For large queries chain prompting can be a real bad idea. https://support.anthropic.com/en/articles/8324991-about-claude-pro-usage

[compact moss]

and then this https://docs.anthropic.com/claude/docs/prompt-engineering

[tacit grotto]

Any updates on RSS ?

[knotty minnow]

"it has to reread the whole context every time you send a prompt" note that this is generally the case with all LLM chatbots, not just Claude (so worth keeping in mind whatever model/provider is being used imo 👍 )

[knotty minnow]

The 'Long context window tips' section they link to here is really worth a read; the optimal structure for file uploads that they describe is very specific and unique to Claude (e.g. not the same for OpenAI models) https://docs.anthropic.com/claude/docs/long-context-window-tips

Although it's not possible to control where the document is placed (relative to the prompt / instructions) using Claude models via perplexity, it's worth being aware of imo

[junior flame]

Thanks to those who dug out this information about Claude 3. In my efforts with it, I found a major difference between its performance in response to my first prompt (complete with inputted document) and what came back to me a question or two later.

[shut arch]

Hey @junior flame!

If you find the original message helpful, please consider reacting to it with the :star: emoji. If the post is appreciated by the community and receives 5 stars, it will go to the https://discord.com/channels/1047197230748151888/1082806833938436228 channel and the post author will get the <@&1082034222778302614> role on Perplexity.

[junior flame]

I see no star emoji in my options for reply. But I am not very good at navigating Discord yet.

[steady stirrup]

when deleting images, it should be permanently deleted from context. this is helpful cuz if u ever upload any images, you are permanently locked to gpt4 in the thread.

[dim sierra]

Probably depends if you select dalle3 as your image generator

[steady stirrup]

?

[digital sigil]

A way to set the overall environment preferences like, metric system for responses or vice versa, temperature units. Basically stuff that divides us geographically. Avoids the additional requests to have it changed to preference.

[summer patio]

Man Phind is now offering 500 opus requests a day. We need more than 5 or at least a way to switch to sonnet to save requests against opus.

[sour acorn]

Even some others like You, Kagi, etc (many others)

[rocky timber]

I've been using this in my main AI profile, works for measurements but not for spelling/language

• Display values using the metric system.
• Use UK English as the language. Ensure correct UK English spelling of all words.
• Use GBP£ as currency, converting whenever required.

[dim sierra]

will.i.am?

[ancient heart]

How does your promo codes work? I tried couple new year ones THAKS23 and something about 2024, but nothing worked, I paid for subscription as usual, i understand that no one really owes me anything, but still, feel fooled to be honest

[visual bough]

There are currently no discounts, only referral codes for new users.

[ancient heart]

I am not talking about now, I ve been talking about previous ones, about promocodes, I saw message of activation, but in fact didnt recieve free months. I noticed recently, because perplexity is only subscription that is attached to my old bank card, and it is empty, and perplexity pro is not active (while it should be by my estimation). Renewed subscription with new card 🥲 😅

[covert sun]

Hey, @ancient heart! The promos you mentioned were limited, you can check their expiration dates in the #📢│announcements channel.

[digital sigil]

For some reason it's not porting to the Android app. The profile works on the desktop and it just loses interest on the app 🙂

[mild schooner]

@covert sun Any chance funcition calling is coming to pplx apis?

[blissful wren]

Hey, I recently gave a feedback on the possibility of copying an output to clipboard on iOS. However, it would be great if an extract of an output could be copied (without the current extra step), and that the haptic touch could enable the rewrite/menu selection.

[rocky timber]

Please can we get rectangular image gen back?🙏

[rare pine]

I wanted a concise summary of the Perplexity API terms of service (I wish companies would start doing this by default) and I got this. Looks like Perplexity can't access itself.

[rustic escarp]

Hi,does the android application supports the generating function?

[covert sun]

Hey, @rustic escarp! Could you please elaborate more on it? Image generation is not available on the apps, yet.