👉│feedback-general

[mossy tulip]

yes we agree

[keen phoenix]

Will there be option to turn off copilot and just use plaint gpt4. This copilot often inaccurately represents my prompts and forgets about some things I included in it while doing his "thinking" so that I prefer plain gpt3.5, especially for programming tasks

[mossy tulip]

yes this option is available in the pro version

[crimson nymph]

Would it be possible to add more payment options like Paypal or iDeal to buy pro.

[mossy tulip]

yes, we should be able to add this feature. adding a task for it.

[weak yarrow]

When in writing mode without copilot, it would be nice to have a higher character limit. Sometimes code can be lengthy and there appears to be room for a higher limit for this mode in theory

[heady plover]

I have had good success appending "[LANGUAGE: <your language here>]" to my prompts.

[mossy tulip]

ok we will extend

[outer viper]

It would be a lot nicer if Perplexity was able to plot functions. For instance when I asked it to explain a RELU and a Sigmoid function, it would be a lot nicer if it plotted the two functions before it wrote the explanation between their similarities and differences.

Perplexity is way ahead in terms of maths when compared to a lot of the other LLMs out there and this ability to plot functions would just take it into another level.

Man I am an absolute fan of this product. I really want this to replace Google/Wolfram/Desmos + all of the other learning platforms that I use.\

Are there any open source repositories for Perplexity? I would be really happy to contribute

[heady plover]

Personally I have noticed I use less and less traditional search engines and fully rely on Perplexity and its features. It really speeds up a lot of stuff for me that I do on a daily basis!

[heady plover]

Would you prefer to allow both or restrict both

[heady plover]

As Perplexity uses GPT-3.5 and GPT-4 models in the background, I think this is mostly an issue with those respectively, not necessarily Perplexity.

However, if Perplexity is able to do something against this, then that would still be great.

[wooden minnow]

I would prefer that it just do what is asked.

[outer garden]

Feedback about the writing function: Co-Pilot seems a bit out of place there. When you want it to do something creative it will constantly ask questions what it should do 😄 Maybe replace co-pilot there with 4 GPT-4 (non co-pilot) responses for the free version

I want it to write random fairy-tale creatures into non fitting settings and it keeps asking me what fairy-tale creature I want and in which setting.

[outer garden]

In the non co-pilot version A formula like ```Here's a formula for a Stable Diffusion image prompt:
An image of [adjective] [subject] [doing action], [creative lighting style], detailed, realistic, 8k, trending on artstation, in the style of [famous artist 1], [famous artist 2], [famous artist 3]

Only use artists that specialize in realistic art.``` does not work at all. (it explains the formula to me instead of using it)

[heady plover]

Can you please change the behavior of the Share button? When I am in full-screen mode, I can see the drop-down options for sharing. However, once I shrink the viewport of Perplexity, the functionality of the Share button changes. There are no longer options to select as it redirects straight to Twitter, which is not good because I do not use Twitter. Can the Share button remain as it is in full-screen mode regardless of the viewport? // Forgot to add, the behavior happens when using Firefox. In Brave and Vivaldi, some options are still available, but "Copy Link" is no longer available, which is the most useful one for me.

[mossy tulip]

this is availabe in the pro version

[trail iris]

#bugreport (or may be unexpected behavior report)
Link to the thread: #👉│feedback-general message

[covert sun]

Thanks for reporting, @trail iris! Though it looks like the link attached is not the link to the thread.

[trail iris]

Sorry, here the link https://www.perplexity.ai/search/9bb2f60c-2835-41e0-b077-d9d136b45865?s=c

[azure vigil]

I'm seeing similar behavior too :
https://www.perplexity.ai/search/1015f458-6ef0-4046-876d-b256586afebd?s=c
https://www.perplexity.ai/search/c176218a-9acf-4c67-b240-ca116118a82a?s=c

[pearl wraith]

I got one too: Perplexity AI: Are these functions invertible?: F(x) = x + 3 and G(x) = x -3 https://www.perplexity.ai/search/aec4992b-e6e2-4582-b6b2-44ac38706302?s=mn

[mossy tulip]

all should be fixed

[covert sun]

Would be great if you could check and confirm if it's working fine for you, @pearl wraith, @azure vigil, @trail iris. Thanks again for reporting this!

[pearl wraith]

@covert sun it's all working for me! Thank you

[azure vigil]

Yep all good now

[slate granite]

hi guys I am interested in an api acces to build saas products . do you have on your roadmap an api?

[glacial granite]

Hey @slate granite! Perplexity AI does not currently provide an API.

[trail iris]

It is working now. Thanks!

[gleaming current]

The related follow up questions went off track @covert sun

[junior wedge]

This is very disturbing if true. It reflects that the AI models are biased.

[covert sun]

Thanks for reporting @gleaming current! I'll add this case to the ticket, too, and we'll check.

[gleaming current]

https://www.perplexity.ai/search/0f0939e3-0ad3-4a49-9deb-9e2f56616adb?s=c

[junior wedge]

I verified that other issue. The AI model does not write speeches against any country or religion or even any heads of state

[brisk cedar]

http://zip.yi5.dev/u/Nxzx5d.png
on the beta website in the payment menu, there is a "saving bla bla"-div with no innerText

[covert sun]

Thanks for spotting that, @brisk cedar! We'll fix.

[heady plover]

I just have a question about

The reason why is because I try them out and I ask questions on Academic and wolframe alpha and the answers don't seem to relaible or detail in any way as compared to internet(all) and I really don't get the point of those displays

[lavish holly]

I usually ask it to show all steps on wolfram

i find that its pretty good with using the right concepts but the basic calculations can be sketchy

[heady plover]

im confused with Academic and why it only gives a bunch of Somantic sources

[quiet lodge]

I keep getting these weird searching-related suggestions (Which are not related to my query in any way)

My query about genes expression in cell line or about a principle in High Performance Liquid Chromotography, and I got related search about what is the capital of France!!!

[heady plover]

what is High performance liquid Chromeotography???.

[covert sun]

Thanks for reporting, @quiet lodge! Sorry for this. We have a ticket on the issue and are working on a fix.

[heady plover]

i need an explanation to the functions and capabilities to Academic

[vapid kelp]

Looks like academic can only read the resumens on semantic scholar and writes base on that.

[stuck basin]

Hey, does anybody have the same problem? When I search something there are nothing more than links, but when reload app answer appear.

[heady plover]

yes, it happens sometimes

[vapid mason]

Backwards logic in this reply ( https://www.perplexity.ai/search/a9535c61-6257-4525-90a5-a0037452fcbf?s=u ): "If the name server is recursive, the response will not have the "recursion available" RA bit set in the response headers. If the name server is not recursive, the response will have the "recursion available" RA bit set in the response headers"

It got it right in the first paragraph, and the sources are correct - but it's weirdly inverting the logic in the middle.

[vapid mason]

Do we report wolfram-alpha issues to you or to them? It's lying about some stuff https://www.perplexity.ai/search/a5f3eb41-d04a-4e16-bd27-d9e9362569dd?s=u it says "This means that the area takes into account the curvature of the airfoil." - not true - and it takes 5 more questions before it suggests is knows this is the case "The total area of the wing is the three-dimensional area that takes into account the distance between the leading and trailing edges along the wing surface. This total area is not relevant to the lift and drag forces acting on the wing in flight"...

[still wolf]

^^

[elder barn]

neither gpt 4 nor gpt 3 have started to provide multimodal features to the users

gpt 3 is not even multimodal

[austere helm]

Hello. I'm having the same problem!

[covert sun]

Hi @austere helm! Please check and let me know whether you can access it.

[austere helm]

yeah i can! thank you.

[mossy tulip]

looks like a bug, we will investigate

[stuck basin]

@mossy tulip I wish u success, u are doing great job

[mossy tulip]

this is our / AI bug, seems like you pushed it a bit too far 😄 try to see if copilot can answer this question correctly

[still wolf]

what does that even mean

[elder barn]

the models perplexity uses under the hood (gpt 3.5 and gpt 4) are not multimodal... meaning they don't support images/videos as input

gpt 3.5 doesn't support images/videos input at all but gpt 4 does

[still wolf]

I mean images with text

[elder barn]

but that isn't available for users to use

[elder barn]

as an input, right?

like you giving perplexity images with text, right?

[still wolf]

something like this

[elder barn]

yeah

that's not possible with gpt 3.5 because it is not capable of understanding images.. it is possible with gpt 4 but it is not available for users to use as openai is currently heavy gpu limited

[still wolf]

put this photo into perplexity and it would copy the text

[elder barn]

^^

and since perplexity runs on these models, it doesn't have this feature

does this make sense?

[still wolf]

ig

[heady plover]

oh those would be nice! Wolframe/alpha actually generated images similar to diagrams and statistic. but then it stopped doing that in Perplexity.

im not sure if other Language processing systems like Youchat, or chatSonic run off of the ChatGpt engine though.

[vagrant badge]

There should be a method to cancel the generation similar to ChatGPT. Sometimes I notice that the generated text is not what I intended with the prompt but need to wait until the model is finished.

[rapid steppe]

how about a feature of turning off popular threads because it annoying and not very useful at all ... looks blaoted

[hollow echo]

Getting the same problem - just subscribed for annual pro.

[crimson nymph]

Hi for now this process is still manual. Just let Alex Romanov know you’ve bought it or rejoin the discord using the invite link provided in the pro section.

[cyan yarrow]

I was hoping that when you decided to charge monthly for your service, you would give some kind of prize or recognition to those of us who have been with you for months when you were just starting out. But I see I was quite an idiot and deluded thinking that I would earn some kind of privilege when you started charging, I see that none of that happened. Sure, you will say that the free plan is still there for gpt-3.5, but that's not what I wanted to happen, for that I also use Poe, Bing or chatgpt itself. When you asked us to create a user, I only did it because I thought it would bring me some advantage when you decided to charge.

[mossy tulip]

there is DISCORD10 discount code

[austere helm]

Nobody owes you anything.
It was just your misconception that you expected something will in return. Meanwhile they have a discount code.

[timber escarp]

is it possible to make the ai do magic to read thru GitHub source code and get information from it so you can ask questions about it? if possible please add it <3<3 [not sure if this is the right channel]

for example you give it a GitHub repo link and you can ask it questions about it and it would source from the repo and the code inside

[lavish holly]

have you tried prompting it to use the url as a source

[timber escarp]

leme find my question sec

and then i asked for examples on how to get all servers using the repo as the source and it didnt really do it correctly

[lavish holly]

i'm no expert but i think copilot might be better suited for that no?

[timber escarp]

oh

how do i use it?

[lavish holly]

don't think it's available on app yet

[timber escarp]

oh

okay okay ill try tmr didnt know its a thing

ty

oh oh im just stupid i linked the wrong repo it managed to do it aaa

[full glen]

Hi, I’m not seeing an obvious way to log out of the iOS app, am I missing something?

[timber escarp]

Is there a way to "regenerate" the response? the only way i could do it was by editing the query with a space and then it gave another response

also "ask follow-up" doesnt seem to work i gave it some code to write and i asked it a question after but it was telling me Since the question does not provide any code to fix, it is not possible to provide a specific answer. However, here are some general tips on how to fix code errors: is this a me issue?

[heady plover]

Could you share the thread permalink ? Thanks!

[timber escarp]

ill send it in an hour or two im not home yet

[keen relic]

Hi, really enjoying Perplexity, I'm using it more than search now in my daily life, work and personal, the biggest help is that it's current, this completely trumps chatgpt in using the tool for research. I also like the ability to remove and add references and let it regen a response, great thinking on this project. I would like to provide some feedback as I use this as an analytics and research tool, and yes I realize I can control focus to some channels. But that said, in the spirit of this project being one of quality information, I would like to suggest adding some different encyclopedic channels, imo it would be better to promote the usage of channels that are generally and historical more valid due to how their knowledge bases were developed. It would seem that using Encyclopedia Brittanica and similar quality encyclopedic resources would provide a better response base. I know wikipedia is very popular and it covers subjects that may not be in traditinal encyclopedias, however, it is included in responses when the ai engine and your algorithyms determine it is within scope of the question posed. If you want to offer a focus channel for encyclopedic paradigm maybe group the traditional encyclopedia such as Brittanica that have made themselves available online to counter balance the wayward ways of Wikipedia, jmho. (side note: I have decided to use perplexity ai to prep for a job interview today, lol)

[knotty berry]

There aren’t accounts yet on iOS, but they are slated for the next release

[timber escarp]

https://www.perplexity.ai/search/1af42e5d-50fc-4eb6-ade1-e1079ce5ad97?s=c

[gray pumice]

𝄞

[upper basin]

It can't talk danish

[gray pumice]

new perplexity be like - tell us about yourself , so we can track you

[upper basin]

The new ai profiles

[elder barn]

🤫

[subtle plaza]

will add Danish now

[crimson nymph]

could you also add Dutch?

[subtle plaza]

Yes! I just need to teach Perplexity all these languages 😄

We're adding them now

[upper basin]

Bet

[arctic harness]

May I ask you to add Polish to the list? Perplexity understands it perfectly.

[crimson nymph]

Would be cool if “in [language]” would overwrite language settings in the profile. Sometimes I want the answer to be in english instead of my chosen language in the profile.

also asking it to translate the text to english won’t help if the language is set in the ai profile

[austere helm]

yeah i'm using it the normal mode and whenever i want i tell it to look for only resources in X language.
or like when i ask the question in x channel it search the web with resources with that language.

so the new default language is not preferable for that use

[vestal iris]

Very happy with the new profile feature so that Perplexity can further refine the results. From the tests I've done it looks like I'm going to be able to forget to remind it to answer me in languages other than English. 👏 👏 👏

[mossy tulip]

will add

[sudden timber]

Thanks for this great technology. If i want to give some feedback, it seems that putting the "Popular" tab upfront doesnt make the app/platform seems personal enough.

It would be nice if we have the option to hide the POPULAR tab as a whole so that we can focus on the personal chat more.

[sudden timber]

I second this. It would be nice if we can hide the popular tab altogether.

[hollow tapir]

How many GPT-4 requests will I get if I upgrade to PRO?

[crimson nymph]

You'll get 100 gpt 4 uses every 4 hours

[idle ibex]

Hey, I just wanted to join the discord. and kind of give my thanks. What you guys have created is truly beautiful. And so it has enhanced my creativity and my deep fond love of researching. and this is literally the chat GPT for researching. It's SOO beautiful. I love how you guys break down the progress. on how the miss. is looking up information. It's great. keep up the amazing work ♥️ ♥️

[mossy tulip]

thank you for kind words!

[burnt parrot]

Hello, I entered my language (Turkish) in the new AI profile feature that came to Perplexity, actually everything works perfectly and the results come in Turkish exactly as I want, but sometimes if I want to get results in English, the results still come in Turkish even though I say "write me an English text about ...." I would appreciate if you fix this problem, that is, if the user wants a language other than AI profile in the prompt, I want the artificial intelligence to give such results.

Thank you in advance and good luck! 😄

[fiery sun]

Allow image input and output

[spare sable]

Output already works, but you have to properly ask the question for it to actually show you images; Input is something that probably wont be released till openai themselves add public support for gpt 4 + images

[elder barn]

bing is currently getting the multimodal features

so if you want to test it out, see if you have gotten access to it or not

bing will get this feature before chatgpt + for sure

[vapid mason]

Reference #2 is a spamwall (and possibly scam) website which hides the answers util it gets your email address: you should probably erase all content from "techtarget.com" from your database: https://www.perplexity.ai/search/8f34485f-56f3-4987-a6d9-d9952799ce2a?s=u https://www.techtarget.com/searchsecurity/definition/end-to-end-encryption-E2EE (techtarget does NOT publish any original content; it's an advertising and sales platform for other content producers)

If you are crawling the web, you should probably use an agent which drives a real web browser that emulates a real person - so your own systems are not "gamed" by bots like techtarget has obviously done in this case. (it gave your bot the real answer, but blocks real people until it can get their email to spam them on)

[mossy tulip]

thanks for the report

[vital yacht]

Excuse me?How do I make a payment to PerplexityAI like ChatGPT Plus?I'm sorry that I can't find the payment button😭

[covert sun]

Hi @vital yacht, when you log in, do you see this in your settings?

[vital yacht]

@covert sun Nope,I can't see that,I can only see Appearance,Avatar,Username,Email and Account😭

[covert sun]

Alright, try to log in and get it in the settings at https://beta.perplexity.ai. We'll check the links on the main site, but the subscription you get on beta will make Pro available on the main, too. Please let me know, if you have any issues.

[vital yacht]

@covert sun Yeah,I can Get Pro in Beta,it works

[wet prairie]

It would be awesome if perplexity, when displaying the references, also gave the page number the statement refers to

[covert sun]

You're talking about some long articles/docs, @wet prairie, right?

[sour acorn]

Hi Alex in fact I noticed this behaviour on the web app too, like sometimes it would display on the left hand side the banner for Pro, but sometimes it just doesn't like it shows up most of the times but sometimes it doesn't

[covert sun]

Thanks for sharing, @sour acorn! It's okay, currently, it's always available on the beta.

[sour acorn]

Is the discount code still valid ? was planning to buy it

[crimson nymph]

you can still use code DISCORD10 to have 10$ off the first month

[sour acorn]

Any plans to add Apple Pay as the payment option ?

[mossy tulip]

subscriptions is not fully rolled out yet to all users, that is why you can see this difference on different devices. we are fully rolling it out this week with an announcement

yeah apple pay will be added soon

[wet prairie]

Yes! Otherwise it is quite time consuming going over the long written docs to find the information perplexity is referring to. It would be quite handy if someone has to write an academic paper using the APA-Format for example. And it would make the answer perplexity is giving more trustworthy.

[shadow scroll]

Piggybacking off of this

Would it also be possible to highlight key information, pertaining to the question asked, on the source page?

Or it scrolls down to the page or paragraph it’s getting the info from

[summer surge]

will u implement functions such as HARPA AI have?

[plush jay]

What is HARPA AI?

[summer surge]

HARPA AI is a Chrome extension and NoCode RPA platform that automates web tasks, saving time and money.
It uses a hybrid AI engine built on ChatGPT and machine learning to perform web automation tasks such as providing AI answers to search queries, summarizing web pages, extracting structural data, tracking product prices and stock availability, monitoring articles and legislation, and detecting changes in competitor sites.
The extension is free to use, with the option of a paid plan for running ChatGPT queries.
It supports various search engines, including Google, Baidu, Bing, DuckDuckGo, Brave, Yahoo, Naver, Yandex, Kagi, and Searx.
The extension runs locally in the browser for optimal speed and performance, without sending data away.
It stores cookies for accessibility and security purposes and can be configured to accept only necessary cookies.

[plucky carbon]

Hi, if this app would include a PDF analyzer it would be amazing 😉. Currently using Petal, but it is too expensive.

[solid holly]

Hi, in the Chrome plugin the link to Perplexity does not embed the URL for summarization in the link URL

[ember cipher]

In the Android app, logging into your specific and desired Google account is.. challenging. The app automatically logs into a specific account and doesn't give you a choice. Any possibility of changing this login system?

[heady plover]

Account selection for devices with multiple google accounts will be available in the future releases. Thanks

[junior wedge]

#💬│general message

[south ether]

Hello everyone,

I am excited to share some updates on how we collect and organize your feedback. Firstly, I would like to express my gratitude to everyone who has been sharing their feedback with us, we appreciate your efforts in helping us improve.To enhance our feedback collection process, we have decided to introduce a few additional channels dedicated to collecting info on specific types of use cases. This will enable us to better track your suggestions, issues, and ideas, and work on them more effectively. We will be adding these channels to the “Use Case Feedback” category and they will include:

• #💻│coding
• #📚│research
• #✍│writing
• #⚡│ask-community

If you have feedback that doesn't fit into any of these categories, please feel free to post it on the #👉│feedback-general or any other channel of your preference.

We encourage you to share your feedback on these new channels, as well as cross-post to previously existing ones.

Additionally, we will be sharing and pinning in each individual channel a Typeform feedback form on these channels. If you have the time to fill it out, it would be incredibly helpful for us to track issues and maintain a better database.

Please let me know if there are any further questions.

https://perplexity.typeform.com/to/v61rN7oF

[tough steppe]

Any blogposts or docs on the technical architecture of Perplexity? Would be interested in understanding how Perplexity's own algorithms / prompts / retrieval interplay with GPT-4 / CoPilot.

[steep shoal]

i’m logged into the~~ iOS,~~ Android~~,~~ and web app and somehow my Your Threads aren’t synced (Android/Web seemed somewhat synced, but the Android app is missing some Threads started on web; the iOS app only includes Threads started in the iOS app) — would love to have Your Threads synced on all platforms/devices

[heady plover]

The iOS app does not have the option to login to your accounts like the android app. Thread sync across all platforms will be available in the future releases. Thanks

[heady plover]

When I ask him for a code, he cuts off in the middle

[south ether]

we don't have anything of that nature to share at the moment, sorry!

[south ether]

can you share the thread?

[heady plover]

https://www.perplexity.ai/search/a0e29538-c069-4624-aae8-d7fe09de407a?s=c

I typed Continue to continue the code, but can you tell me what to do, this did not happen?

[covert sun]

Hi @heady plover! I'd edit the prompt like this, so it has enough tokens to finish the code block and the follow-ups if it's still not enough. You can edit the latest query or delete it and ask to finish with another, probably more specific prompt.
https://www.perplexity.ai/search/6e052b04-3c83-4917-a174-c5b62416f77f?s=u

[heady plover]

Thank you

[dry zinc]

There’s a few things I dislike about the AI. When I ask it to list something specific, it sometimes answers with incorrect information. However, it apologizes and claims it revised the list. It never adds, removes, or changes anything about the lists afterwards. 🥲

[covert sun]

Thanks for sharing this feedback @dry zinc! Could you add a couple of examples (permalinks), so we could improve.

[dry zinc]

I cleared my history a bit ago and I don’t exactly remember what I asked. This just happens frequently

[quick patio]

Pasting a long input message in the web interface will cut off the bottom and top of UI panel. Latest chrome. Mac OS.

[spare sable]

Honestly that’s such a must have, in terms that many people do not utilize all 100 requests. So definitely +1 from me

[fair kayak]

+1 on $10/mo. One thing I also think would be great is the ability to toggle of the forced window for adding additional context/answering questions. I provide a lot of context in my initial prompt and Perplexity still asks me for follow-up redundantly. It would be nice to not have to answer these questions as it hinders the workflow and makes the end-to-end response time much slower.

[dapper loom]

Agreed on the additional context prompt. It’s a pet peeve of mine having to answer additional questions (or even just click skip) when what it’s asking for is already in my prompt. Also I’m sure it’s already in the works but the iOS app needs login/copilot pretty badly now that pro is a thing. Love the product though, recently switched to Pro from ChatGPT Plus!

[vivid portal]

Please allow filtering sources. Look at how Phind is doing this. Some (popular) sources are just junk, and if they are constantly polluting the response then the site is useless.

[south ether]

Hello! We are aware that Copilot can be redundant even with plenty of context, you can always share a link so we can take a look at the specific query. We are working on making it more efficient in that regard. Thanks for the feedback.

[south ether]

Focus searches could be helpful for some use cases, just to clarify; by filtering you mean being able to block sourcing from specific URL's?

[copper wind]

I want able to set response lang or default lang. Example: 3-4 week ago when I give Turkish input , it used to gave English output. I want that back.

It's not a feature but it used to gave more results than now.

[trail iris]

You can set it in AI profile

There is text field specifically for that

[hollow sail]

Cancelled OpenAI today because of you guys. Mainly because my coding results have been more accurate and the rate limit is much more generous, making it way better IMO. Biggest thing I’m hoping for is iOS copilot/4

[knotty berry]

It’s in the works (I know I’ve said that a lot haha)

Some other foundational things need to be finished first but then it’ll come soon after

[keen relic]

trying out copilot, nice tool to attribute queries, I have noticed that input field text runs behind copilot toggle object, this is on chrome

[lament scaffold]

When is copilot coming on ios?

[covert sun]

Hey @lament scaffold! #🍏│feedback-ios message

[frosty cipher]

Hola

[boreal tartan]

I have tried to pay for the pro subscription with a Chinese MasterCard. Then it shows error says that the card is rejected. What should I do?

[south ether]

Thanks for your support 🙂 iOS Copilot/4 coming soon!

[tall crane]

Stuck here

[hollow hollow]

I am having problem typing Japanese / Chinese in the web prompt on firefox

I tried typing but nothing shows up

[covert sun]

Hi @hollow hollow! I've checked on Firefox and it worked for me. Do you mean when you input characters they are not inserted or when you send the query it doesn't return any answer?

[hollow hollow]

When I input character they are not inserted

[fierce sentinel]

Please Add a Stop button to prevent generating answers in case of wrong input

[empty narwhal]

This morning I was rewriting a bunch of old c# code to use xdocument instead of xmldocument and perplexity has been invaluable to rewrite the methods, thinking on changing my pro from gpt to perplexity 🤔

Of course now I have to test etc but a great beginning

[south ether]

We've gotten this comment before, we'll work on it

[south ether]

happy to hear Perplexity helped, let us know if it worked.

[lament yew]

@south ether The Use case feedback channels are very in active.

The think the current biggest hinder is the co-pilot forgetfullness

But loving this software so far

[ember cipher]

Any chance of adding Finnish and Norwegian to supported languages, seeing as Danish and Swedish are already supported?

[tired spire]

I want to congratulate the perplexity team for making the best search engine i've ever used

just joined the discord to say that

i want this tool to survive particularly as the alternative to bing chat, i don't want to use edge just to get access to this capability

the only thing that i'd have constructive feedback for rn is the ability to see follow ups of a previous question somehow

[south ether]

We just set up those channels, hoping they pick up soon. It's helpful for us to be able to get feedback on precise use cases and queries.

[south ether]

We are always looking to improve the AI, Copilot will only get better! thanks for the support

[south ether]

Hi Calcio, yes it's possible. I've shared this with the team, should be implemented in the next couple of days.

[south ether]

Thank you for the kind words, this is useful feedback.

[heady plover]

hey simple question sometimes i like to input words like "in essay" or "in detail" for some reason every-time i input those words the answer to my question do NOT become more detailed and when i input to write the answer in multiple paragraphs perplexity fails in doing so is they a way you could optimize it so the Ai could respond to other words in the sentence other then just the query?

[south ether]

Hey, that's out of the ordinary. Can you share the query? The AI should be responding adequately, but if it's not feel free to share a link so we can look at it more closely. Thank you for the feedback!

[heady plover]

https://www.perplexity.ai/search/8d06d24e-9724-4afc-8000-5821fa4827a4?s=c

[south ether]

Okay I see, yeah it's definitely not giving as detailed of an answer as it could. Maybe you could try rephrasing the prompt in different variations too? just to see different results. Thanks for the feedback and we'll look into it.

[heady plover]

like in what way?!.

including "definition" or "meaning"?!/

[south ether]

Yes, something like this: Could you give me an extended definition of thermodynamics for an essay? it could work better that way. Still, your prompt alone should technically be sufficient, since you are specifying that you require details, so we'll still take a look.

[heady plover]

ohhhh

[ember cipher]

Thanks. Already implemented now, I see. You guys work fast

[lament yew]

keep in mind that the context window of gpt 3.5 currently is 4k max. Which is roughly 3k words.
Still the output is short tho.
try co-pilot?

@heady plover This exhaustive enough? for a general search?
https://www.perplexity.ai/search/c26e1e6a-d5ec-4186-adee-12537eaf4f2f?s=c

[weak yarrow]

for curiosity I thought I'd try it with co-pilot, I gave it a challenge of trying to write 6000 words initially 😂. It said it was unable to, in the end it managed to write about 500 and something words

https://www.perplexity.ai/search/9f56c06d-9de1-4d0e-9e51-833a15085258?s=c

[lusty lantern]

Based on my observations in these months, the "All" mode operates almost as if it were on rails about the length of the answer.
The original Copilot ~2000 characters long answer (https://www.perplexity.ai/search/af842401-3a5d-4387-923f-7bfa103fde3c?s=u)
can only be extended up to ~3000 characters (https://www.perplexity.ai/search/fc79342f-3d23-430f-8b27-f86435f794d8?s=u).

To obtain really more detailed answers, and also better answers - if you don't need >sept2021 knowledge like in your examples - you must always use the "Writing" mode.

To obtain ~**5000 **characters, that is the upper limit of every currently available GPT-4 chat (excluding the Playground), simply use GPT-4 and "Writing" mode and add the post-prompt "Use 5000 characters" (or 10000 if you want to push the limits), like this:
https://www.perplexity.ai/search/1e100425-ca95-47d8-a0bc-2cdc06b602fb?s=u

Hopefully, the Perplexity team will make the rails of the "All" mode more flexible, enabling us to have ~5000 characters in that mode too, especially in Copilot now that it is powered by GPT-4-32K.

[lavish holly]

Frequently getting cases involving latex where it can't properly format single $ signs

[gritty schooner]

Is there a spot to enter the DISCORD10 code when subscribing?

[dusk aspen]

When you go to the link payment page to get a Pro subscription, there is a spot to enter the promo code

[gritty schooner]

@dusk aspen thanks, I'll take a look

[vapid mason]

FEATURE-REQUEST: Please let us press the "Up Arrow" on our keyboard in the query-input-box, to copy what we typed in from the previous query: LOADS of times I find I need to make subtle changes to what I just asked, and it's a major pain to re-type or mess about with move-mouse/start-mark/move-mouse/copy/move-mouse/paste...

[junior wedge]

@covert sun

@covert sun #💬│general message

[coral condor]

Hmm, I pay for pro and my Copilot uses are definitely not resetting every day. To be fair, I don't really need 600 uses of copilot even over a week..., but I figured it's worth pointing out.

[crimson nymph]

#💬│general message

[gritty schooner]

I'm not seeing any place to enter a code? Am I at the correct page?

[south ether]

Hello John, unfortunately the discount code was only valid for beta subscriptions. Sorry for not making that clear.

[south ether]

Appreciate the feedback, would you mind setting a screenshot? I'll make not of it still, I'll share this with the team. Thank you!

[coral condor]

I apologize for the confusion (lol). Spezi cleared that up earlier. And, they are regenerating because earlier I saw 570 uses left, and now there are 580. So, I think it's working fine. I just didn't understand how it worked before. So, is k!

[south ether]

My bad too, my brain didn't point towards that direction, I thought for a sec it was a bug. But yeah it just changed to 600 daily.

[gritty schooner]

Thanks @south ether

[icy goblet]

So no more discount to try Pro now, right?

[south ether]

Hello Dickson, I'm afraid not.

We are focused on improving the Pro experience so that it's well worth it.

[icy goblet]

Okay got it thanks. I will seriously consider it once I can use it in the iOS app!

[south ether]

makes perfect sense, it's a priority to get iOS app to catch up to the rest of the platforms. It will be available soon 🙏🏻

[small wing]

You can actually access the web version on iOS through safari or other browsers, which is not as convenient, but it's a good way to go

[covert sun]

Hey @boreal tartan! Sorry for the late response and the inconvenience. Basically, we can recommend to try another card, as it could be Stripe that declines some cards or the bank that issued the card.

[vapid cedar]

I have two suggestions:

1. Separating a response into a new thread
2. Ability to copy an entire thread easily (without copilot)

[azure maple]

Are search histories synced across desktop and iOS? I'm a Pro user if it matters

[azure maple]

Nevermind. I see in the FAQs that they are not

[south ether]

Sadly, not yet! we are aware that this is important and we will definitely implement in the future.

[heady plover]

Hello, I have just subscribed to pro I just want an assurance that you guys will keep on running this service and please do not sunset or back out like other projects out there i am really loving your service and want to keep on using throughout the years!

[void horizon]

It would be nice to reformat the web experience for small screens. In an iPad, the bar on the left takes up half the screen. The App is essentially worthless since you can’t log in, so we’re stuck in Safari with half a screen. I wrote a UserScript on the desktop to improve the formatting, and imagine. I’ll do the same for iOS eventually, but it would be nice if it worked better out of the box.

[steep shoal]

Google Scholar data appears to be publicly available but when asked about i10-index and h-index scores for scholars Perplexity seemed unable to find results. This would be a nice data source for responses.

[south ether]

Hey @heady plover our intention is to keep Perplexity going and growing for a long time. Thank you for the support and welcome to Pro 🙂

[south ether]

@void horizon Thank you for sharing, we are working on getting the app going but i'll make a note of the web browser interface. Would you mind taking a screenshot of what the interface looks like on the iPad?

[gritty schooner]

Does the Android app use GPT-4 if you have a subscription and have it selected in the settings on the web app?

[south ether]

@gritty schooner sorry for the delayed response, at the moment GPT-4 is not supported on the app so it won't work on the Android app, unfortunately. This will change in the future 🙏

[shy forge]

Hey! Have a couple of items for feedback : 1) Need Copilot on iOS sooner 2) Latencies for copilot responses are still slow. Need much improvement on that. Thanks!

[south ether]

@shy forge thank you for sharing, we are working on getting the iOS app to have every feature asap. I know it's taking some time but it's the #1 priority. Appreciate your support.

As for Copilot, the AI will improve over time. It depends on the query too, but we are also working on that 🙂

[pure canopy]

https://twitter.com/mysteriouskat/status/1670673031885357057 ranking search results + manually defining why I did or didn't like could be an interesting way to further address this/self-improve the user profile

🚨 Katherine Brodsky (@mysteriouskat)

One thing that concerns me about the move towards ChatGPT-like search engines: You see the answer without seeing how it arrived there. When you do the search yourself, you evaluate the results based on your trust in the sources. With AI, you have no idea what it's relying on.

Likes

238

Twitter · Jun 19, 2023, 6:01 AM

[vapid mason]

Hallucinating all the math answers: https://www.perplexity.ai/search/3918a545-c6fb-45d9-8075-df6e0c221218 - I thought we could get correct math solutions if we picked wolfram... what am I missing?

[visual bough]

https://www.wolframalpha.com/faqs/
Why doesn't Wolfram|Alpha understand what I asked?
You might need to ask it differently, or it might be that Wolfram|Alpha doesn't yet cover what you're asking about. Try making your query the minimum number of words needed to communicate. Try different words or notations. And be sure to check your spelling. If you still need help, Pro Premium users can try contacting

[icy goblet]

I don’t think it works the same way as the plug-in in ChatGPT

[dusk aspen]

Yeah, the Perp team should look into making integration with WolframAlpha much better than it currently is, as I understand it, as of now you'd be better off posing such a question in the "All" section. As I understand it, the "WolframAlpha" section merely forwards basic queries to WolframAlpha and retrieves the results.

[echo canyon]

Today I believe I got incorrect answer from a simple question regarding one hot encoding in data analysis. Not sure if it's due to Perflexity or GPT 4? Expecting official response

[jovial axle]

how to target wolfram alpha in perplexity app?

[visual bough]

Until now, after you have asked a question and the answer is produced, you always have to scroll down independently. I would welcome it if this could be done automatically, as for example with ChatGPT.

[visual bough]

This feature will be added in the future, just follow the updates.

[covert sun]

Hi, @visual bough! We've got this request earlier, as well.

As for my experience, even for ChatGPT answers you have to sometimes scroll back up.

Perplexity's answer generation is faster and that means, you'll definitely have to scroll up quite often to read it thoroughly, so that's pretty much the idea behind focusing on the beginning of the answer.

Thanks for this feedback, we'll look more into it.

@visual bough, and thanks a lot for supporting the server 🙌!

[heady plover]

https://www.perplexity.ai/search/c5db88b3-e413-43c0-a613-05cc4cd44c5a
search operator not working

[south ether]

Hi! could you specify what is not working for you? or what the results you were trying to get were?

[heady plover]

I searched for android games -2023 (I don't want results that have "2023" anywhere in the page) but all results are the opposite

[south ether]

I think the line doesn't quite "subtract" which is what I gather you wanted to imply with the - . Perplexity thought you wanted results from 2023.

Sorry it didn't get that

You could try being more specific like "list me android games that aren't from 2023" or use the word exclude in there

[crimson nymph]

the search operator does seem to be working, maybe it doesn’t do that well with dates

It would be nice if Perplexity would support before: and after: search operator to narrow down the sources that are used.

[heady plover]

oh it works now

weird that it didn't work before

[heady plover]

this doesn't work well

ok so

sometimes it works but other times doesn't

[heady plover]

https://www.perplexity.ai/search/29d7768b-fd8c-4e9f-ab23-72d9b1b1bf20?s=u

on the other hand, this worked
https://www.perplexity.ai/search/446679a0-7837-496b-96a8-9ea8580125ac

really weird

[south ether]

yeah that is weird

did you use copilot for one and not for the other?

[heady plover]

nope

I only used copilot for the apple one because I was trying to replicate Spezi's results

and it didn't work as you can see

[south ether]

Okay, I'll pass your feedback on and take a look.

Thanks for sharing

[heady plover]

I also tried some other things and most of the time it doesn't work

[south ether]

If the queries aren't giving you the results you want always feel free share the threads so we can take a look at them.

[heady plover]

alright, will do

[lusty lantern]

@crimson nymph, there's still no such thing as search operators in Perplexity, this is why it seems to work randomly.
Perplexity currently only interprets your words as ChatGPT does, not as Google does, and tries to guess the Google query/queries needed for your given task using AI, in a non-deterministic manner, likely again using GPT.

[crimson nymph]

They do support search operators #📘│faq 5

[lusty lantern]

OH, sorry, we really need some clarifications then.
For example, I wonder how an "OR" is interpreted in a prompt like "ARE DOGS ANIMALS OR TREES?"
I'm a frequent user of the after:yyyy-m-d operator in Google and always struggling to make it work in Perplexity, since it's still not a search operator in Perplexity.

[visual bough]

Hello dear dev team,
here I have a few suggestions, I'm sorry if some of these things may have already been requested or are already in work:

1. File upload, for example PDFs or pictures, to have them analysed.
2. Exporting the answers, as PDF or as Excel file (just imagine you have to search for data for something or upload a file and perplexity would sort everything into a table...)
3. Filtering and sorting options to make the search much more targeted and the output even more accurate.
4. Detailed documentation on how to use perplexity.ai and how it works.

[lusty lantern]

Yes, that's what I read from the FAQ linked by @crimson nymph.
So, how is an "OR" interpreted in a prompt like "ARE DOGS ANIMALS OR TREES?"

[lavish holly]

I imagine it gets interpreted like it does in google? That's not being used as a search operator

[lusty lantern]

How do you know that Google doesn't still interpret it as a search operator in that case?

[lusty lantern]

How did you verify that it didn't include pages that don't contain one of the words "ANIMALS" or "TREES"? That's the only method to confirm that it wasn't used as an operator for the results.

[lavish holly]

I suggest you try for yourself. However, I don't see any page that mentions the term "trees" with that query because Google's algorithm interprets it as part of a sentence/question rather than a search operator.

Cool thing about AI is that you can also find out by asking it

[lusty lantern]

How do you know that Google's algorithm interprets it as part of a sentence/question rather than a search operator?

[lavish holly]

Try the query yourself in google search and you will see that it isn't interpreting it as a search operator like I showed you in my screenshot

[lusty lantern]

you could also ask Perplexity how to convert it into a regular expression, that doesn't imply that it will work as intended in a Google query

Here it is: https://www.perplexity.ai/search/bd4436f0-87a5-4389-ac90-26183b478e39?s=c

[lavish holly]

Regular expressions aren't for Google search though right?

[lusty lantern]

They're not supported by Google search, they're too expensive to support

[south ether]

great suggestions, some of these are being considered or even in the process of being implemented but it's great to have more people validate their use case so we know they will be valuable 🙂 thank you for sharing !

[sour acorn]

Hi Denis sorry to bug again, is there any update on Apple Pay being added as a payment method ? Really waiting for it

[mossy tulip]

hey, sorry this is taking a bit longer than we thought. we're working to improve the quality of our iphone app and bring it up to the same level as the web version before we add the option to subscribe through apple pay. we want to make sure it's worth your money, and right now, it's not quite there yet.

[lament scaffold]

can you please share with us the highest priority, upcoming features for perplexity pro?

[lament scaffold]

Potential new feature : Prompts on perplexity should also be able to read transcripts from youtube videos and reference them in the answers...

[crude wadi]

I would love to if it was possible to go straight to a focused search using an URL. For example https://www.perplexity.ai/writing or https://www.perplexity.ai/youtube.

It would save me two clicks each time.

[heady plover]

fwiw you can already do this to an extent using the site: operator

just paste the url excluding https

[crude wadi]

Thanks but could you explain how to do it exactly?

[visual bough]

This is very simple. Your question + site:URL

[heady plover]

here's an example: https://www.perplexity.ai/search/37831779-54e1-4fad-8f75-1dff90293df4?s=u

it won't work in writing mode as that doesn't have access to internet

doesn't work for video though

if I misunderstood your question, please clarify

[heady plover]

if you were asking if you could set a focus as your default search then you can use these URLs:
https://www.perplexity.ai/search?q=%s&focus=writing
https://www.perplexity.ai/search?q=%s&focus=youtube

[covert sun]

Hi @keen relic! Wanted to update that this was fixed. Thank you for reporting!

[sinful dragon]

How can I direct perplexity to read a specific page?
In this conversation it have gone to wrong way, saying there is no such struct in rust crate, but there is , and I was trying to show example, but it gone to the search instead. https://www.perplexity.ai/search/ac78e430-8563-44e7-b9bb-5c2a0d96eda0?s=c

[crimson nymph]

if you want to read content from a specific website or page you can use the site: parameter

[heady plover]

like this

exclude https

[lusty lantern]

I think that @crude wadi was asking for a shortcut URL to open Perplexity with a preselected mode (eg. Writing)

[heady plover]

well I answered that too ¯_(ツ)_/¯

[lusty lantern]

Try to click on your links

[heady plover]

you're not supposed to

put it in your browser

[lusty lantern]

try to put it in your browser then 😉

[heady plover]

I did, works fine for me

[crimson nymph]

it's not a link to put just in your browser search bar. It's for adding it as custom search engine in your browser

[lusty lantern]

I suppose he wanted a link to open Perplexity with an empty prompt box, ready for input

[heady plover]

ah well, that's not possible currently afaik

but adding it to browser effectively works the same so

[lusty lantern]

not the same for me: I always want GPT-4 for Writing mode and, since I prefer to leave GPT-3.5 as default for non-Copilot web-augmented searches, in current version I still have to switch-on Copilot before submitting the query.

Update: I see that there's a GET parameter for Copilot too, so I'll use https://www.perplexity.ai/search?q=test&focus=writing&copilot=true
Thanks @heady plover for the idea!

My ideal would be a shortcut for Writing+GPT4 without Copilot though. In "Pro" subscription, I currently have to temporarily change a "setting" every time for this.

[naive tangle]

Hey @lusty lantern if you're a Pro member and have GPT-4 set to "Always" in your settings, you should be able to do this query and have it be Writing + GPT4 without Copilot (and you shouldn't need to toggle a setting every time) — let me know if I'm misunderstanding.

https://www.perplexity.ai/search?q=test&focus=writing&copilot=false

[lusty lantern]

Hi @naive tangle, yes I know that, but I currently can't combine that option with the fact that I also need GPT-3.5 as default for non-Copilot web-augmented easy queries.

For these kinds of queries, like for example "who is Macron?", I'm used to getting an instant answer in Google via the featured snippet. In the absence of a "featured snippet" in Perplexity - that btw would be a great addition to take another slice of users away from still using Google - a GPT-3.5 non-Copilot query is still the closest I can get to that.

Currently, my favorite shortcuts would be:

• Standard 3.5 "All"
• Standard 4 "Writing" (I miss this)
• Copilot 4 "All"

I'm still using ChatGPT Plus and a couple of other GPT tools a lot for pure-GPT queries mostly because they're faster to answer or allow for more customization (eg. temperature 0, reliable system prompt, one-click predefined pre-prompts) or allow for longer input prompts (up to about 20K tokens / 61k characters).
I think that I would use Copilot 4 "Writing" much more if the "interactive" feature, with its current delays and pausing/preventing the generation while the tab is in background, were moved to the **end **of the generation, to fine-tune and regenerate a better version of it or as follow-ups.

[naive tangle]

Thanks for the detailed feedback @lusty lantern will think through all of those ideas 😄

[signal flicker]

Since the mobile design of the website is pretty polished, and (currently, as far as I can tell) the iOS app doesn't support copilot searches, would y'all consider allowing the website to be run as a PWA so it can be used instead of downloading the app?

[outer moss]

when I copy the content to markdown editor, it will look like this

[brisk cedar]

Same

[brisk cedar]

Try the copy button ok the bottom instead of selecting the text

[heady plover]

Can you please add the delete and edit query buttons for android app? Cause they are not there

[soft sigil]

I belive when it come to geopolitics it's better to avoid using Wikipedia

[cedar vector]

I think, it would be better if Perplexity AI can speak and answer questions in other languages. By the way my language is Tagalog, and I am from the Philippines.

[covert sun]

Right, if you use the "copy" button, as @brisk cedar mentioned, answers formatting will be better.

[covert sun]

Thank you, @heady plover, you're right, those buttons are missing. They will be added to the app in the following releases.

[heady plover]

Cool. Thank you! Good luck in improving the website and app, the platform is helping me so much that right now it is the sole search engine that I use right now

[brisk cedar]

Why are image links so broken 😭

https://www.perplexity.ai/search/67fd0f25-dfe7-4834-85a0-3772f91872cc

[heady plover]

Hi @brisk cedar , bug has been reported.

[vapid mason]

Getting these on almost every query now: Something went wrong...
PERPLEXITY
Error in processing query

It is also super annoying that you erase all our typing and replace it with the words "Something went wrong". Please don't do that. The #1 thing we always want to do when "something went wrong" is copy what we just asked and paste it to try again - which we can't do because you erase what we asked. Grrr...

[cedar vector]

I hope that Perplexity AI can respond in different languages. It would be great if it could switch languages depending on the language used in the conversation. I'm happy that it understands different languages, but it currently responds only in English. It would be ideal if it could respond in the same language it was asked. I suggest Perplexity AI must be multilingual.

What I meant is the ability to switch languages seamlessly. For example, I use Tagalog when asking questions, but it responds only in English, unlike ChatGPT and Bing AI that can respond in my language.

Another concern I have is that its ability to translate text into other languages is not yet advanced. Perplexity AI suggests using translating tools like Google Translate, which makes me wonder: doesn't it have the capability to translate text into other languages on its own?

[naive tangle]

Hey @cedar vector have you set Tagalog as your preferred language in your AI Profile?

[cedar vector]

There's no settings for Tagalog...

[dusk aspen]

Just tell it in the AI profile that you're a Tagalog speaker and that you'd like it to respond in Tagalog when queried in Tagalog

[cedar vector]

There's no Tagalog on the selection of languages.

[dusk aspen]

Oh, I believe AI profile only works on Desktop with Copilot atm, but I think the devs will be rolling it out across other offerings soon.

[south ether]

We are going to be adding more languages over time, sorry that your language isn't available. We'll add support asap 🙂

Meanwhile on desktop you can always ask in Tagalog, it's just not supported by default for AI profiles.

[cedar vector]

However, the problem is, unless you explicitly state that it should respond in Tagalog, it won't respond in Tagalog. Instead, it will answer in English by default. Ideally, it should have a default setting where it responds in the same language it is being asked. It's strange to see that when you ask it in another language, it still answers in English. In my opinion, this is not correct. It should be fixed as soon as possible to make it more user-friendly. More people will embrace a technology if it is user-friendly, efficient, and effective.

[kindred tinsel]

I found your suggestion via search and just wanted to add my +1. The popular tab is often about upsetting topics (such as incidents where a lot of people have died). It's important to know what's going on in the world but I would prefer not to be reminded of this kind of news every time I want to use Perplexity.

I'm greatly enjoying Perplexity in general though. 😄 It's a lot quicker and more effective for getting my questions answered than other "LLM + web access" alternatives

[harsh rain]

It would be great if we were allowed to switch in the follow up question the type of generation we want

For example, doing first generation in writing, then the second one in Wolfram alpha

Sometimes it feels like if you do Wolfram Alpha as first, it doesn't understand the context

[real sluice]

it said it wasn't a discord bot...but it is?

[harsh rain]

Bot is querying the perplexity API, which is running openAI API under the hood most likely, so it doesn't know about bot stuff, it's just like if you asked directly via website

Well the bot is doing something g okay..

But thanks for correcting

[lusty lantern]

Hi @cedar vector, I pointed this out several days ago and it's still not fixed.
I also showed that we need the Google queries to follow the prompt language, as we don't want our web searches to always and only see <1% of the web, i.e. only pages in Italian or Tagalog language. This is drastically nerfing Perplexity for us.

[heady plover]

i third this…i mean fourth

[heady plover]

also created a pro-feedback feature request about it https://discord.com/channels/1047197230748151888/1121578679269609582

and then, a brilliant suggestion was made by @short imp to be able to control/filter Popular Threads using your AI profile 🤯

[heady plover]

just experimented with a workaround on macOS using StopTheMadness. Seems to work pretty well lol

my work here is done

DM me if you want the rule to add

[ashen abyss]

I think the AI profile might be a bit aggressive on some occasions. I just searched for the Yellow River and it responded that there’s no such river in the search results, then gave me unrelated rivers in Ohio (where I live). Is it possible it chose to add my location to the search query it used? Perplexity AI: the yellow river https://www.perplexity.ai/search/303abeb1-6940-4b0e-96d0-04d84e50740a?s=mn

[brisk cedar]

There is a location field below the ai profile instructions

[ashen abyss]

Agreed, it was just surprising that my location would bleed into queries I thought would be unrelated.

[south ether]

makes sense! thanks for sharing, we're taking all these comments into consideration. Happy to hear you are enjoying Perplexity 🙂

[south ether]

Thanks for sharing a link, you can use AI profile to share your location as previously mentioned and it will always prioritize the information you've given it. You can pause AI profile by clicking on the top right pause icon for queries like this. We are looking to improve this experience in the future.

[heady plover]

i have a big problem with perplexity

it has to do with the pause button

everytime i type in something that i did not mean to type in and i keep pressing the pause button it does not work

the AI just keeps on writing

[covert sun]

Hi @heady plover! Could you please share if it's on desktop/mobile and the browser you're using. Also does this happen in the Basic/Copilot/GPT-4 search mode?

[crimson nymph]

what is this workaround for?

[heady plover]

it just hides the Now Popular block of random popular threads on the homepage

[crimson nymph]

I see, if you remove the parent div of the popular section it will also center the input div

perplexity.ai###__next > main div[class="w-full"] > div:last-child

[brisk cedar]

does this also remove the login button?

[crimson nymph]

I am already logged in

[brisk cedar]

profile button?

[crimson nymph]

it only removes the popular section

[crimson nymph]

also still there I removed it from the ss as you can see my address bar is also gone

[brisk cedar]

ah sick

[heady plover]

love it, talk about iteration!

it works for me with just

main div[class="w-full"] > div:last-child {
  display: none;
}

but not with the perplexity.ai###___next part -- what is that for?

[crimson nymph]

aah I see that is how adblocker block parts of websites you need to add the sitename and a # but I never used StopTheMadness

[elder barn]

Why did perplexity fail to access the website and fact check it? https://www.perplexity.ai/search/4e8b7afa-6730-4077-a97d-d561c3811d2e?s=c

[brisk cedar]

Perplexity only uses search results, it can't access a website directly

[elder barn]

yeah but it does get that website's content, right?

[brisk cedar]

as you can see with this search, google hasn't indexed it
https://www.google.com/search?q=site%3Areddit.com%2Fr%2FChatGPT%2Fcomments%2F14jjhbv%2Fgoogle_deepminds_ceo_says_its_next_algorithm_will

[elder barn]

oh

[brisk cedar]

You could ask it to fact check a text which it should ask

[elder barn]

but then how does it summarize a website when I give it the link of a website?

that is fundamentally the same thing, no?

[brisk cedar]

Nope, only the browser extension can do this

http://zip.yi5.dev/u/BMsk2M.png

[elder barn]

huh

[brisk cedar]

This is how I would go about it if you want to use GPT-4

[elder barn]

I am so confused

[brisk cedar]

http://zip.yi5.dev/u/0s5TTh.png

[elder barn]

istg this used to work before

pasting in link and asking it about the content of that website

your point makes sense that it can search the web so I should have probably given it the post's title and it would have worked better

but idk.. iirc this used to work before

or am I hallucinating now?

[brisk cedar]

A few days prior another person here tried and it also didn't work

With "Read the content of this website and give me first and last line"

[elder barn]

yeah I saw that

[brisk cedar]

afaik it cant access sites directly

[elder barn]

yeah yeah I definitely remember that

[elder barn]

same with other web searchable AIs, right?

cause every service fundamentally works the same way

when it comes to searching the web at least

[brisk cedar]

Either do this (providing the text)
https://www.perplexity.ai/search/173827cd-2eaa-460a-a5a1-44d47bd08a27?s=u
OR (asking it to search for the post)
https://www.perplexity.ai/search/8655fe0d-e2f3-451f-af2f-1dc5c144ab09?s=u

[brisk cedar]

I'm not sure, there may be some which can do it. I am not even sure if perplexity does actually visit the websites or just parses the previews of the google-searches

[elder barn]

ok

makes sense

thank you god.. feels so weird typing this ngl

[brisk cedar]

2 days ago i looked into how perplexity may work using their traffic. and only saw google search results and "snippets" of the answers

So i am not even sure if it visit the websites and actually uses the full results

[elder barn]

yeah I have noticed this too

perplexity's web sources are very similar to that of google's top search results

[brisk cedar]

(probably)

[elder barn]

This is not the official though

[brisk cedar]

Yeah

A guess based on the name of the variable in the response

[elder barn]

where did you get this json from?

[brisk cedar]

I caught it from the network traffic using a Mitm proxy

Using burpsuite

[elder barn]

🤯

yeah so it a 100% uses google's search query

that's impressive

[brisk cedar]

I wouldn't say 100% because its not officially stated but 🤷🏻‍♂️

[elder barn]

in your experiment @brisk cedar , did you see perplexity considering ad promoted websites?

[brisk cedar]

Nope, but im not sure how the search API they use works

[elder barn]

alright alright

thanks again... god

[brisk cedar]

lol np

[elder barn]

I sometimes don't understand the point of copilot asking questions back cause it makes no sense

I tried that reddit post thing with copilot by pasting the content of the text and asked it to list me all the accurate information and the inaccurate information

and it ended up asking me to select the accurate information and inaccurate information

[heady plover]

do you have the permalink for this query @elder barn ?

[elder barn]

https://www.perplexity.ai/search/f94ab8f3-7a78-40dd-ac2c-c2aef1064e65?s=c

I skipped those questions but this is the instance I shared in the ss above

[heady plover]

thanks!

[heady plover]

https://tenor.com/view/thats-not-true-nope-no-impossible-star-wars-gif-5199322

I hope it uses a web crawler on the backend to extract full text from sites that show up in the search results

dang you might be right though...shows what i know 😅

i feel like soon all humanity will be all knowing

[heady plover]

this has been happening to me more too lately

[muted basalt]

Planeta

[junior pasture]

I also encountered this problem

[ashen abyss]

When you say "previews of the google searches" are you suggesting that they parse those couple sentences that show up when you search on google.com? I'm pretty sure that's not the case, as it's given me information that wouldn't fit in that preview. Maybe you mean something else, though.

Looks like it gets a summary of the contents of the page, if you can believe the bot itself: https://www.perplexity.ai/search/8466d0ce-9b20-4e8f-aec3-c6bfbaa58da9?s=c How it produces that summary I can't say.

[heady plover]

at least it is honest:

note: that was using the site: operator

[south ether]

@cedar vector we've added Tagalog as a supported language that you can chose are your default through AI Profile. It should be way consistent now. Thank your for your patience 🙂

[brisk cedar]

Not sure, but in the json the browser gets are only the searches via title, short description and link available.
They probably use full results yeah

[cold sable]

Copilot 600 times a day？

Why a day passed and he didn't update

Because he can only do 300+ a day?😕

@heady plover

[spare sable]

I am a pro user and it lets me do it 600 times a day, I think the 300+ is just something they forgot to update

[cold sable]

Yes, but he seems to have passed a day, the number of times not refreshed, I am confused

[cold sable]

Have you encountered such a problem?

[spare sable]

Idk, when I use it it lets me do copilot requests up to 600 times, so idk

[near furnace]

Holy cow I just discover this PerplexityAI thing, utterly blew my mind how smooth it is, compared to all sort of search including New Bing, Bard, or plethora of plugins of GPT-4, I'm not joking I'm willing to pay half of my monthly income for this beauty, even for the base version.

[urban seal]

@covert sun why does the swich of preview and concise canceled, in the former version,i in no copilot mode can switch preview and consice to choose the amount of word number but now canceled .can it be fixed and use again?

[surreal pine]

Does perlexity understand the content of technical papers if gpt-4?

or what should be right way to frame prompt?

[brisk cedar]

Im not sure if it can even parse website content of a link you provide.
I believe it works only based on search results and "public" & indexed data

[dense heart]

Would be possible to use perplexity to draw a pie charts from tables or data that you're researching ? This would be awesome😋

[brisk cedar]

You could try to ask it to visualize the data using LateX piechart
(nevermind)

[junior wedge]

a hack way to get it to visualize data is to tell it to generate u numpy/scipy/python-notebook that visualizes the data

[junior wedge]

you'd only be able to see front-end processing if my knowledge of Burp is correct. It would also probs be a bad idea if they were actually parsing the requests/data on the frontend, which would give away their "secret sauce" 🤣

[dense heart]

well, I tried but what I can see it this

What I would hope to get is like in ChatGPT you have plugins which can turn tables into visual images.

[junior wedge]

you'd need to pipe data to it first. So maybe do this:

1. share a link to a public accessible Excel spreadsheet, then ask it to visualize that data using Python, Jupyter Notebooks and the applicable plugins

[dense heart]

Ohh, man this is such complex process that you trying to showcase

[dense heart]

But thanks for your advice anyway

[junior wedge]

not everything needs AI 😉 The regular Office software makes data visualization simple

[brisk cedar]

A simple visualization for gathered data within the AI is useful regardless :p

[dense heart]

I agree 100% with this. I was simply curious can perplexity handle this task or not.

[junior wedge]

well don't 100% take my word for it. If someone knows the deep inner workings, it's definitely @crimson nymph

[dense heart]

it this Spezi working in perplexity team ?

[junior wedge]

nope, nor am I. I'm just the 1 who suggested this community mod feature and Spezi is smart. We're volunteers, but maybe @mossy tulip might hire us in the future 😉

[dense heart]

Thanks mate for help

[covert sun]

Hi @urban seal! The model shows the detailed answer by default now. Could you please share the issue you’ve got with it, so we could also check.

[surreal pine]

sad !

[cedar vector]

Hi, I have a concern about pressing the "Enter" key because when I press it, instead of moving to the next line or paragraph, it automatically submits the response even if I'm not done asking my question... I hope this can be fixed. Thank you.

[brisk cedar]

YOu can press Shift and Enter for a newline

[cedar vector]

What if I am only using mobile phone?

[craggy fog]

Maybe already answered somewhere else but i couldn't find it. When starting a new thread and selecting search option "all" will this include all of the other option like "academic" and "writing" or only use the "all" option and not include like wolfram and such?

[brisk cedar]

Newline doesnt really matter to the AI-LM but if you really want it you could make it in a note and copy it over

I think for most people enter is "search" or submit

So I dont think they'll change that

[brisk cedar]

The options limit the sources. For example "academic" limits it to published academic papers.
All does not filter any results and instead just uses the "top" ones

[cedar vector]

Okay, I just found out that Shift+Enter also works for me even when I'm using a mobile device... But I think it would be better if the "shift key" is removed and we can simply use the "enter key" to move to the next line of text. What is the purpose of the shift key, and why can't we use the enter key to create line breaks?

[brisk cedar]

Because enter key is generally used or implemented for submitting stuff. This is just they way most entry fields work, such as google

[heady plover]

not in chat apps though

like discord and whatsapp etc

which is what perplexity supposed to be like, so

[brisk cedar]

I would say its rather a search engine in the first promopt, the additional queries are "chats"

I think its a mix but rather search engine

[heady plover]

no.

search engines show links

perplexity doesn't

[brisk cedar]

wow your puncuation makes your statement so much better!

perplexity does show sources. it makes me question if you ever used it

[elder barn]

All the AI generated answers are like google/bing snippets. If you use them with web search on then they act as a search engine (rather an AI powered chatbots that can search the web)

You can argue about them being more on the chatbot side if you turn off its web search ability

plus there is no thumb rule that search engine must show you links. It is just an engine that takes your query and searches for it in its database. It could return it in form of article links, video links, subreddits, AI generated paragraphs (which is also full of links)

[junior wedge]

does your phone keyboard have 'Return'?
that would go to a newline

[cedar vector]

What do you mean of Return key? It has the arrow symbol of the Enter Key.

[junior wedge]

it must be device-specific then cause some devices have this key that goes to the next line

[elder barn]

depends on the phone and the app

some apps recieve the return button as new line like Kong said. If for you the return/enter key sends the prompt to perplexity, then try hitting return/enter key while holding the shift button that is often denoted as the upside arrow

It works as a new line character on some devices and websites

[willow hedge]

I am a pro subscriber and the iOS app still seems pretty basic with no control over using copilot nor the type of search (reddit, academic, etc). Not a big deal because the mobile site works very well. Just wondering if there are plans to expand the iOS app.

[south ether]

Hello! thanks for sharing. We are in the process of adding most features including login/sign up and Copilot. We'll let you know when the app will be updated on our official channels, stay tuned and thank you for the support. 🙏

[keen relic]

Hi, I have found that when I go to start page I without thinking drop my query into the main search field on page instead of navigating to the left to start a new thread, but most of the time I don't get a search, I'm wondering if that's my browser security blocking or something else, I though it was because the field was being used to search existing threads but that doesn't seem to be the case, any thoughts/

?

[crimson nymph]

What do you exactly mean with "I don't get a search"? Does the answer not load? Or does it do nothing at all?

[keen relic]

right, no response

It may have been browser security as it's working right now, if it happens again I'll try to capture more info. So if that happens, what kind of info would help the most.

[brisk cedar]

maybe it was because you did it too quickly upon loading the page?

[crimson nymph]

What browser are you using? I know some extensions may cause issues on Perplexity.

[brisk cedar]

Hey, since I saw that you are here @crimson nymph
Do you know if there is a way of visualizing a pie chart inside perplexity?

[crimson nymph]

Perplexity can't generate pie charts (images). But you could provide the data and ask it to make a pie chart using for example the python library matplotlib.

[brisk cedar]

Yeah alright, thank you. That's what i figured :/

Some way of visualizing it using latex would be dope

[cold sable]

@covert sun

hi

I am a pro user, and I have a question: When I first open and select Copilot, it seems that it's not using the GPT-4 model. However, if I first open it without using Copilot, the GPT-4 model is triggered, and then using Copilot again allows me to access the GPT-4 model. They seem to answer questions in the same way, though. Is this a frontend UI display issue?

As a pro user, you have encountered a question regarding the use of the GPT-4 model in Copilot.
You have observed that when first opening and selecting Copilot, it doesn't seem to use the GPT-4 model.
However, when first opening without using Copilot, the GPT-4 model is triggered, and then using Copilot again allows access to the GPT-4 model.
You also mentioned that both models seem to answer questions in the same way, and you are wondering if this is a frontend UI display issue.

[lavish holly]

Did you ask perplexity to point out the key points of your first msg? 😂

[covert sun]

Hey @cold sable, could you please share those Copilot and GPT-4 threads, We'll check.

[cold sable]

https://www.perplexity.ai/search/ce7130d0-82f1-413f-b462-9b01a7f389b6?s=c; https://www.perplexity.ai/search/7cc72daf-f48b-4626-aa00-9ac3e9199624?s=c

@covert sun

If I want to use gpt-4 in copilot, it seems that it cannot be operated

[heady plover]

Copilot uses GPT-4 to answer your queries. It’s not displayed like in non copilot queries where there is a visual indication of what model you’re using (but copilot uses GPT-4 to answer your queries)

[cold sable]

Thanks！！！

[cold sable]

I'd like the AI to understand a Vue file with approximately 1000 lines of code. However, when I upload it to GitHub and use the GitHub link, it seems unable to read the information inside. I've confirmed that it's a public repository. Is there any way for the AI to understand a long code file? 🥹
url: 'https://github.com/994AK/wqeasd/blob/main/index.vue'
😕

https://raw.githubusercontent.com/994AK/wqeasd/main/index.vue

https://www.perplexity.ai/search/8af735dd-8cc0-4271-9501-71f3e857447a?s=c

Hi@heady plover

I apologize, but I am unable to access the provided URL directly or provide search results based on it. However, I can still help you with any questions or issues you might have related to Vue.js or the content of the file you mentioned.
😮 Does he support reading like this?

@covert sun😬

[weak yarrow]

It can't read URLs directly unless possibly cached by Google. The best workaround I can suggest is something like I did:
https://www.perplexity.ai/search/176efc70-b9a7-424f-8870-67cb861283ef?s=c

However, when copilot asked me for the code I had to submit it four times for some reason. Excessively long code seems to cause that to happen, in which case this is. I'm not sure if it's simply a bug or some other reason behind that

[urban seal]

@covert sun What I said before is that the brief and detailed switching button in the old version is gone, and the new version does not have this button

[crimson nymph]

Answers are now detailed by default.

[urban seal]

yes but before we can switch today we can only see brief answer

[crimson nymph]

The answer is not concise but detailed. It might give a short answer depending on the question.

[urban seal]

@covert sun At the same time,can there add the third-party payment channels like PayPal, WeChat， Alipay and UnionPay Cloud QuickPass and other third-party payment tools? I don't have a cash app, and I don't support this here.

[ashen apex]

hey, I'm a Pro subscriber. Just wondering what the reasoning was behind removing the Additional context section in question follow-ups?

[urban seal]

But now I feel that the reply of the normal version is obviously much shorter, unless the code itself is very long, and the general drawing does not use gpt4

[south ether]

Hello, it became part of Copilot. The regular search default is detailed. If you want to ask for additional context without necessarily using Copilot you can prompt it to do so by following up with a request for additional context like this:

[hollow sail]

Would greatly appreciate toggle for pilot, 4 and 3.5

[south ether]

@hollow sail I will make note of this request, thank you for sharing.

[vocal vortex]

I just did it myself currently working an a chrome extension

[south ether]

This is cool. We also have a chrome extension, have you tried it?

[cold sable]

Thanks

[lavish holly]

i been waiting on a feature like this forever

[junior pasture]

I would like to express my strong feedback on this issue: when using the writing mode, the system still searches the internet for related information, sometimes encountering political content and national leaders. As a result, the answers become tainted and distorted, resembling a mixture of GPT-4 and irrelevant information. This is not the purpose of choosing the writing mode; I expect the information to be purely based on GPT-4's response.🙃

[lusty lantern]

it never happened to me

[naive ravine]

Hi! Is there any news about the launch of the API for Perplexity?

[glacial granite]

Hey @naive ravine! Perplexity AI does not currently provide an API.

[crimson nymph]

Perplexity currently doesn't have any plans for API access. If there are any changes with that they will be sure to share it publicly. #👑│pro-lounge message

[near furnace]

I know I said it yesterday but I just want to say I so love the Perplexity developers, I just finish one gruesome analytic report writing task, and I think I use at least 150 searches per day lol, basically is my coauthor at this point, every time some thing i want to search, fact check, expand, i just ask Perplexity and then tab out to continue working on my writing, and come back to see what Perplexity brought me and ask it to dive deeper or discuss with it, super super helpful thing, if i say it 2x my productivity is a super understatement.

On average, I believe, fact checking that used to take me half hour is usually a 15 second job for it

[junior pasture]

I would like to inquire about the fluctuation pattern of the available credit limit here. I've noticed that it doesn't decrease consistently throughout the day but instead varies with sudden highs and lows, which leaves me quite puzzled.😌

[knotty berry]

The queries are refilled on a rolling basis - that is, if you made a copilot query at 5PM, and then at 9 PM, the next day at 5PM, you get one back, and again at 9PM

[junior pasture]

Oh, I understand now. It's each query that is timed individually, rather than all of them resetting at midnight every day. Thank you.😋

May I ask if any senior executives in your company are board members of OpenAI? How can you use GPT-4 almost without limits? If an API is provided in the future, could it be exploited by hackers to steal usage quotas after jailbreaking?👀

[torpid adder]

IOS app doesnt offer login... why?

[visual bough]

#🍏│feedback-ios

[keen relic]

Thanks, it's not happening any more for now, I think it was a browser security issue performing some script blocking. Not concerned at this point, all good.

[vocal vortex]

I also importing feature to be able to select previous questions from perplexity. This only a solutions since we all know they will implement it sooner or later. I’m just a fan who is trying to max out the capabilities everything you need to about what I’m currently doing is here https://boost-perplexity-ai.web.app

[mossy tulip]

very cool

[primal rivet]

Hey guys I would like to know is there any way to make Perplexity go into FB ad library

[heady plover]

feature means kinda useless if you ask me

like they already show you the related queries why do you need extra button?

[lusty lantern]

I'm not sure about how he implemented the link but, for example, I would like to be able to click and read multiple related questions and I can't currently do that

[vocal vortex]

Because they are gone once you click one of them

[gritty schooner]

@vocal vortex I think that's where the confusion lies since in the sample video the related queries are right below it

[safe trout]

When I recently put a sentence in a prompt and ask for a rewrite, the reply is cut off in the middle of the sentence. I even wonder if the number of tokens has decreased.

[heady plover]

Hi, I asked perplexity to design a site for me, but it doesn't give me any code and only guides.
My Prompt:
I need a site to manage my educational organization. In the first step, you have to design a dashboard with HTML and CSS, in the second step, use Boostrap for the user interface, and in the third step, use jQuery for animation and interesting effects, then the instructor list. and also put the amount of the budget in an HTML table and design it with CSS and apply different colors, then use HTML and CSS to display the list of instructors in the box and their positions and training courses and specializations Put it in the box

https://www.perplexity.ai/search/df28fa1b-7e2a-46c0-92da-88322177c1b5?s=u

In my opinion, this is a problem

It works very well in normal situations, but when the prompt increases, it becomes disturbed

[spare sable]

Try and do it with Writing mode enabled! Writing mode is a lot better for coding.

If that does not help, try and use a prompt from the discord.

[junior pasture]

When I'm browsing this website on my phone and the answers keep appearing, I don't want the screen to keep jumping around if I accidentally touch it. It's really tiring for my eyes🤓

[vocal vortex]

I do get the point but I can guarantee that it will make sense once you click one of the questions and they disappear

[void horizon]

iOS app seems pretty silly without login. Should have been in 0.9

[visual bough]

Preliminary feedback on "Quick Search":
I honestly don't quite understand the point of this change. I like the new design with the boxes, but they are still the "sources" from before. I would like to put them back under the answer and not as "Quick Search" but as "Sources".

[gray pumice]

@full venture

what is this?

now when i click on perplexity extension it just takes me to webpage

[full venture]

The benefit is you can now navigate to your destination a bit faster if you are searching for a website. You're right that they are the same sources as before. We will work on the naming.

[full venture]

The new Chrome extension changes your default search, it is different from the old Chrome extension.

[visual bough]

Sure, it's faster that way - but don't you use Perplexity for the very reason that you don't want to go to any website, but rather let the AI do the job? Correct me if I'm wrong...

[visual bough]

Other question: Are you sure that exactly 2000 is the token input limit?

[full venture]

Yes, that will always be the primary use case but we also want make the navigational experience somewhat smooth (rendering sources before the answer completely streams).
2000 is the token limit as measured by the chatgpt tokenizer - it is approximately the same as the tokenizer here: https://platform.openai.com/tokenizer but not exactly the same.

[visual bough]

That's exactly why I'm asking, I also work with the tool and according to it my random character text has 2500 tokens and would still be accepted.

[visual bough]

On part 1: All right, I am curious about the development. 👍

[spark sorrel]

@full venture not sure if it's a bug or feature, but if Copilot asks me a follow-up question to my query, and then I click "continue" without answering (NOT skip), then it asks the question again up to 3 total times.

[full venture]

That's a bug, sorry we should be treating that the same as a skip

[spark sorrel]

Awesome, thanks for the quick reply. Just wanted to bring that up in case you didn't know about it. Cheers!

[visual bough]

@full venture Has the token limit problem been fixed yet? (looks like this to me)

[junior pasture]

Unhappy, the input limit is only 2000 tokens, it used to be convenient for proofreading long articles, now the experience is very poor.🙃

I hope the token quantity limit can be restored to the previous level. I can accept a slower speed, or give Pro users a larger quantity, or provide an option for users to choose between fast speed or larger capacity limit😬

[junior pasture]

I completely agree, using this tool means not wanting to search through so many websites

[heady plover]

I liked it a little bit better when the sources were at the bottum but its fine..

I still like it

[mossy tulip]

we should be able to bump it up soon, just have a few things to resovle first

[junior pasture]

Report a bug. When I enter a URL to request a summary of the webpage content, everything is fine. But when I edit the same post and change the URL, sometimes the previous URL's content still appears. It seems to be a caching issue. However, I couldn't reproduce this problem in my screenshots. This issue did not only occur after today's update, it occasionally happened before🙃

[sudden timber]

After the latest update, Perplexity AI's response to my queries is noticeably SLOWER than before. Is anyone else experiencing the same issue?

[sudden timber]

I've got this kind of response multiple times since the update. It processes the query & in the end returns an empty result with links to sources.

Do everyone else also experiencing the same issue?

[coral condor]

I remember getting that last night once.

[heady plover]

Hi @sudden timber ,could you share the permalink for this query ? Thanks

[sudden timber]

heres the link:
https://www.perplexity.ai/search/ba5d5d7d-9e85-4cb7-b498-6c9c7dbf1f0b?s=c

There were 2 instances (in this thread) where it replied with an empty result and only links.

*the 1st and last query.

[junior pasture]

I also encountered it, the answer is blank https://www.perplexity.ai/search/13350b6b-12cd-4799-8ca9-8ac2dfa3e746?s=c

[lusty lantern]

We're reporting the same issue here: https://discord.com/channels/1047197230748151888/1125726944139432079

[craggy fog]

Same here. Had it several times today

[weak yarrow]

I've been having the same problem throughout today, earlier this morning the website was slow to respond at times for me. I thought it was just me so ignored it

[heady plover]

me three…four….n

[full venture]

Hey sorry everyone, thanks for noticing the issue, we are working to improve latency and reduce the blank errors also.

[lavish holly]

Is this a problem on my end or perplexity's?

Happens to me in Writing mode

[weak ravine]

@full venture I've been trying out the new perplexity search extension, it's been great. It would be even better if you could incorporate shebangs (!G redirects to google, !B redirects to bing search etc...).

[lavish holly]

The sources section keeps flashing as new text is being generated which causes a back and forth of text displacement causing it to look "buggy"

not sure if anyone else is having the same issue

[full venture]

Thanks for spotting this, do you see any errors in the console? Which browser are you on?

[full venture]

Hi, in Chrome you can use shortcuts to set these up: b<tab> should bring up bing, etc.

[raw grotto]

Can't login using "sign in" button in Android app

[weak ravine]

Another point of feedback for the perplexity ai search extension:
Enable Copilot by default for pro users (or allow to toggle this)

[lavish holly]

(I'm on Edge)

[lavish holly]

also after trying some more queries it seems like this issue isn't common

[lusty lantern]

As alternative to <tab>, <space> works too (at least in Edge)

[covert sun]

Hi @raw grotto! Are you talking about the email login option? Could you please share more details?

[junior pasture]

The system did not accurately understand the user's intention. I asked it to visit a webpage and summarize its content, but it accessed different content instead.🤓https://www.perplexity.ai/search/42bc452d-e382-41d1-8ded-9a9ebebff2a8?s=c

[inland flower]

somewhat off-topic, but your page description for some search engines are off 😄

[junior pasture]

Sometimes there is only one source page, but the generated answer is filled with numerous annotated markings, which can be really painful to read.🙃

Sometimes, the page is very clean without any clutter, making it pleasing to the eye.😋

[junior pasture]

Sometimes, the page is very clean without any clutter, making it pleasing to the eye.😋

[heady plover]

Was just thinking about the Continue versus Skip buttons in Copilot: I think it would be nice and make more sense if Continue operated differently at least for when the interactive checkboxes appear. In that case, Continue should check all the boxes to indicate I am interested in all the listed options (checkboxes). This would also be convenient instead of having to check all checkboxes manually, which sometimes can get tedious.

[delicate maple]

How do I cancel my free trial if I don’t like it?

[crimson nymph]

If you click on this button on the settings page it will redirect you to a page where you can cancel your subscription

[lavish holly]

As my perplexity usage increases, the more I'm wanting a category and naming system for threads 🥺

[delicate maple]

Trying to click subscribe it’s showing this page but scrolling and clicking on the other one

Now I can click the buttons they just don’t do anything any help would be great

[heady plover]

Hi , @delicate maple , could you refresh the page and check again ? Does it still happen even after the refresh ?

[covert sun]

Sorry for the inconvenience, @delicate maple! We'll fix.

[delicate maple]

It does yeah

[delicate maple]

All good 👍 thanks

[heady plover]

Hi , the issue is fixed now. Could you check if it works fine for you ?

[lavish holly]

I'm not sure if this is an issue on my end, but I keep noticing when I try to paste a longer text into Copilot it "logs me out" and I have to refresh for it to turn back normal

[delicate maple]

It worked! Thank you so much! Not only the best chatbot but the best staff too 🫡

[thorn spade]

Hi, the numbered sources mentioned in the text itself do no match with the citations they are supposed to be linked with underneith the text. (3) could be 6 for example

[brisk cedar]

Can you share a link where you had this experience? So the developers can look into it

[cedar vector]

Hi, using my mobile phone, why can't I use the Enter key to move to the next line of words? I hope this can be fixed as soon as possible. Another concern I have is that I asked for assistance from Perplexity to translate the text, but it wasn't able to fulfill its task properly. Instead of translating the text, it suggested online translating machine sites. I hope this can be fixed as well.

[junior pasture]

The send button on the mobile page is too small, so you can just press enter to send, which is more convenient for everyone

[heady plover]

Hi @cedar vector , you can use Perplexity in writing mode for this purpose.

[junior wedge]

@covert sun SEO issue here (looks like the sitemap might be incorrect)

[vapid cedar]

Want to second this. Realllllly wish there was a way to change the citation format to APA. GPT4 has difficulty converting it too.

[heady plover]

Hello, when there are many codes, the last parts are not written, and when I type: Finish The Code Block. It starts writing from the beginning and again part of the code is not displayed
Does this problem always exist?

https://www.perplexity.ai/search/5e6ccacb-16c4-4b52-a842-6b7945e7c473?s=u

[tough quail]

Hey, I added Perplexity bot to my server to use it privately but it's not reacting. Is it possible to use it in my own server?

[crimson nymph]

Make sure the bot has the permission to talk in the channel you try to ask questions in. You can use the bot by tagging the bot and then asking your questions like this: @PerplexityAI why is the sky blue?

[tough quail]

Tnx, my mistake. I added it to the server but not to specific channel. It works. I love it! 👍😀✅

And the dreamer one also 👍👍👍

[cedar vector]

What do you mean of writing mode?

[covert sun]

Hi @cedar vector! Perplexity has the "Writing" mode you can choose from the focus search options. It doesn't need to connect to the Internet and is helpful with translation tasks (it doesn't suggest you proceeding to any websites). It's available on the web version and will be in the upcoming Android app update.

As for the search start on pressing Enter, it was on the web version, right?

[heady plover]

When will this problem be solved?

[covert sun]

Hi @heady plover! Thanks for bringing this to our attention. We are working on improving this.

[heady plover]

You're welcome

[weak ravine]

Aare there any plans to bring code interpreter to perplexity (if it isn't already, not sure)

[silk vale]

what about bibtex?

[cedar vector]

I hope that Perplexity will add more features such as "Copy" and "Share" features. And I hope that the "Enter key" will work properly. I have already complained about this several times because every time I press the "Enter key" on my cellphone, instead of moving to the next line of words, it automatically sends the message even if I'm not yet finished typing. Please add the simple "Enter key" to the features of Perplexity. I hope the developers of this app can fix this issue as soon as possible. Thanks for giving me space to give you feedback.

[heady plover]

Hi @cedar vector thanks for your feedback. The copy and share features are already available on both the app version and the browser version.

[cedar vector]

What I mean is the Copy and Share feature for each answer of Perplexity in each follow-up question, not for the entire conversation. I only want to use the "copy feature" for one answer, especially for long answers that cannot be copied entirely. As for the "Share feature", it only allows sharing the link of the conversation and not sharing the actual text and pasting it on any desired platforms.

[heady plover]

You can press on the clipboard icon at the end of the answer to copy the complete answer along with the sources/citations. This copy feature ( clipboard icon) is possible for individual queries in a thread. To share the entire thread , you can use the share option. Thanks

[sour acorn]

Hi Denis can I know any ETA for the Apple pay option to be made available for buying Pro ?

[zealous sequoia]

Been using Perplexity for a while and a lot of tasks and one piece of feedback is for Co-Pilot.

I think Co-Pilot should always allow you to add the additional notes even if it is just the "choose which topics you want to focus on" since sometimes they might be going in the wrong direction and it would good to give it a guide.

[gray pumice]

https://www.perplexity.ai/search/08c039cc-1f03-47c0-a41c-6f009ed8ace3?s=u
this doesnt look klike it returned right code

or am I missing something?

like , maybe i didnt ask question properly. however, definitely not right code

[crimson nymph]

If you want information from 1 specific website only you should use the site: parameter in your prompt.

Example:
YOUR QUESTION HERE site:barecss.com

[gray pumice]

this is however not what is happening on the website

https://www.perplexity.ai/search/f1c5ae18-6ab7-4c08-b2a8-2c01fbdb4a12?s=u

[short imp]

The site operator has not been reliable. Sometimes it works but even when it works it might be going to a cached version of the site

[crimson nymph]

Yes the site operator is not reliable, most of the time it only work the first prompt of the thread and after that it will add random sources, hopefully this will be improved in the future. (in this example the sources are still from the same website but this is not always the case) https://www.perplexity.ai/search/ecd6b7d6-7522-496e-a96e-8f8ca65c70c9?s=u

[hushed cipher]

I hope you guys use this https://www.todesktop.com/ to convert website to a desktop app

I've heard a lot of companies are using it now

• Campsite
• Typefully
• Linear
  etc

[cedar vector]

Thank you for your response, but what about the 'Enter key'? Has it not been fixed yet? Making the 'Send key' the same as the 'Enter key' is not feasible for mobile users because there is already an existing 'Send key'. It's really frustrating because sometimes I want to move the insertion point down, but I can't do it because what I thought was the 'Enter key' is actually the 'Send key'. I hope this can be resolved quickly and properly. For computer users, it's easy to use Shift+Enter to move the insertion point down, but it's different for mobile users.

I have only used Shift+Enter on my phone once and it hasn't worked again. It won't move the insertion point down. I hope the 'Enter key' can be fixed to make it easier to move the insertion point down. Why do I need this? It's to improve the prompt creation process and to effectively convey instructions to Perplexity AI for better responses. The 'Space key' and 'Enter key' are different, and the 'Enter key' cannot be substituted by the 'Space key' because the 'Enter key' is the only key that functions as a 'key to move the insertion point down'. I hope this can be resolved soon.

[devout kraken]

Same here!

[gray pumice]

I don't know if thats intended but one can just bookmark it that way.

¯_(ツ)_/¯

[dusty fulcrum]

perplexity ai is absolutely amazing, among all the ai's like chatgpt and bing ai perplexity does the best job most of the time.

[heady plover]

it so good i don’t even care about the results anymore, I just enjoying the Search 😜

[south ether]

Hi Android users;

We want to hear what you have to say about the app experience, if you'd be interested in participating in a one on one 30 minute interview through Google Meets (you can chose between a voice or video conference) with the designer to talk about yourself and how you use Perplexity through the Android app, please fill out the following intake form and we will reach out to arrange a chat.

In exchange for your feedback you will be compensated depending on your geographical location.

https://perplexity.typeform.com/intakebeta

*edited to provide additional context

[tawdry sinew]

Hello. Is the design shown in the screenshots different from the one we have in Spain? The white box does not appear for me

[visual bough]

Can you give us more information about how the whole thing works exactly?

• Is it a text chat interview or a voice chat interview? Or can we choose? (I ask because one might prefer talking and the other writing)

• Does it take place here on discord on private channels or does it run through another application?

[south ether]

Hi, one on one is a voice or video call. We always generally welcome written feedback through any other channel (like this one) but for a one on one feedback interview having an oral conversation is favored.

It would be through Google meets.

[junior pasture]

Do I have to speak English？I only speak Chinese😋

[visual bough]

I think English is preferred...

[south ether]

For internal purposes we would need to conduct in english. But thank you for your interest 🙂

And your feedback through Discord is always appreciated and welcome.

[warm vale]

So far having a great experience using Perplexity.

Regarding the image, I'd like to suggest adding this functionality:

1️⃣ Button to prompt all the related questions.
2️⃣ Input to choose the amount of related questions the user wants to see or load per prompt.
3️⃣ Load more related questions button.
4️⃣ Icon button to regenerate the related questions
5️⃣ Notion integration, for example, I want to automatically sync and neatly format the threads to my perplexity notion database.
6️⃣ It would be amazing if you guys could increase the engagement in every prompt with images, could be AI generated, from pexels, etc...
7️⃣ I also think that adding different views could greatly enhance perplexity, for example, kanban boards, being able to drag and drop prompts, and organize our threads.

I'd love to hear from you what you think about all of this.

Have a great day!

[south ether]

Thank you for taking the time to put together this list, I'll share these with the team and we'll definitely consider them for future builds.

[south ether]

We've definitely thought about what happens after the thread has been generated and what users do with that information, where does it live after, why it would be revisited, etc.
So added features to threads has the potential to address some of what you've suggested.

[south ether]

As for images that's definitely coming! we know how much it would enrichen the overall experience.

[small wing]

How about add a button to facilitate the user to specify how to use the language of the search, I as a Chinese user would like the system to search in English and answer me in Chinese
Before today's page update, I set up Simplified Chinese in AI profile, and when I searched through English, it was able to search for English in the search engine and output Chinese results normally, and after today's update, no matter how I use English, he will go to search for Chinese content!

https://www.perplexity.ai/search/943d5b2b-9ee8-4373-b472-a43d0ed86f00?s=c

[small wing]

There is also the problem that there is a small probability that the currently generated results will have to wait for a long time before the full results are output directly, rather than by way of SSE,
example:https://www.perplexity.ai/search/97df2659-9c79-4962-a303-c582e7837e38?s=c

[bitter flint]

Ive started renaming threads using a numbering scheme to make it easy to search. I also love that i can use the thread URLs in other tools to organise. Though wondering if there is a way to merge threads within perplexity. i.e. organising threads around a common search theme (like a topic or tag or namespace), that can link threads created asynchronously between search sessions)?

It may not be that straightforward to merge threads since each thread has its own prompt context, perhaps a tags or namespaces can be a good way to link threads. That might be a good datapoint back to the machine as well.

[junior pasture]

Sometimes pressing "Enter" means a line break, sometimes it means executing answer, it's a bit confusing

Can users be allowed to choose whether pressing "Enter" creates a new line in the settings?

[south ether]

Hi, this shouldn't be happening. Enter should always send the query, is this happening on the web browser version?

[south ether]

Happy to hear you are enjoying renaming threads, and for sharing some great ideas on how to allow people to organize them even more.

[junior pasture]

On computer webpages, it seems that pressing enter sends the message, but on mobile webpages, it does not

[toxic path]

I would like to buy a Perplexity Ai Pro, but I have a question, ´will I have the pro version also on iOS?

[junior pasture]

Pressing enter on this page creates a new line, I want to send a message

[junior pasture]

The VIP qualification is tied to your account, regardless of whether it is an Android or Apple system

[latent roost]

I don’t know if this is the pipeline or not, but sometimes after sending the query I notice a typo or autocorrection to a completely different word. So instead of having to copy/paste/edit or retyping the query can there be an edit button like ChatGPT has?

[visual bough]

There is an edit button, but only in the web version, in the apps it should be added in the future.

[latent roost]

Ok thanks, I am mostly using it on mobile devices where this would come in handy 👍

[misty fiber]

Make the indicated section hideable when we scroll to read our answer.

[sour acorn]

I think available only for beta users for now

[harsh iris]

Hello every one AB from Morocco here. I have a proposal that might improve the user experience in Perplexity. I recommend that once you need to clear all threads, a prompt appears to ask if you want to clear liked threads too, this way we can make sure that we keep the threads we are still needing after the clear, without asking the engine again(which might give a different answer) .

[sudden timber]

I love the ability to edit the thread Name. Kudos to the team.

[junior wedge]

Interesting that there's no 'X' to close it. @covert sun Please note to add a "close popup" to this modal

[lament scaffold]

Please focus on increasing accuracy of information emitted by Perplexity.

[south ether]

We are always working on improving AI Quality, is there a particular query that contained inaccurate information? could you share so we can take a look?

[lament scaffold]

Nothing specific I can recollect right now. Will share if I find anything specific. In general, I still cant find myself replacing google search for perplexity. But I can see immense potential.

Some specific feedback that can help:-

1. The response character limit should be dynamic and based on the question. Ex : For simpler questions, the response should be less than 3 lines
2. Youtube captions of all videos can be added to the knowledge base and perplexity should consider answering questions based on that knowledge

[solid sigil]

I got a response that terminated early after generating a long string of zeros as part of a broken link to an image: https://www.perplexity.ai/search/508f13c2-461f-4aa3-a153-63fdf4e5cc72?s=c

[tacit nest]

potensi Waduk Sermo sebagai pembangkit listrik turbin air

[flat anvil]

#💗│sharing message
I missed a discord link on your website. Also I would like to get a better faq. The page "help" looks useless.
I don't really understand the copilot and asked the ai about it, because I found no infos.

[crimson nymph]

The discord link is located in the bottom left. There is also a #📘│faq channel here that explains the copilot functionality.

[flat anvil]

No there is no discord

[crimson nymph]

I see, it becomes hidden on the mobile layout.

[flat anvil]

Yeah that's bad. Today has everyone a smartphone also discord is used on mobile 😉

[icy halo]

Copilot refreshes maybe every 2 hours, is it okay?
I had 5 in the morning, and I used 3 requests, but it refreshed to 5, and now happened same

[flat anvil]

The ai told me all 4 hours

But it got that knowledge from a blog 😁

https://voicebot.ai/2023/05/29/perplexity-ais-new-copilot-feature-provides-more-interactive-personalized-answers-with-gpt-4/

[icy halo]

Yes it is correct, thanks
#📘│faq 11

I forgot about faq

[flat anvil]

I'm blind? I don't see a explaning there

[crimson nymph]

nr. 14

[flat anvil]

That doesn't help me

Maybe a more simple explanation with samples would be great. "How does the copilot work".

[heady plover]

Hi @flat anvil , you can click on the “blog” button to access the invite link for the discord server. Thanks!

[flat anvil]

Thats not where anyone would look...

Better add it to the header

[tough halo]

It seems like threads are not saved synchronously across all devices using the same perplexity pro account.

Any idea on how to solve this on my end?

[south ether]

Sorry to hear, may I know what devices you are using? iOS is the only platform that doesn't support thread synchronicity at the moment.

[tough halo]

Yes, I am using IOS. Is there a specific date or an estimation of the time that it would take for thread synchronicity to be supported for IOS?

[heady plover]

#🍏│feedback-ios message

[flat anvil]

I have to get a warning how I write my own stories for myself....

[junior pasture]

Can we have fewer annotations for just one source of information?

[flat anvil]

Try to use the ai profile #1124646097013456896

Write like "give me always only 1 annotations of the same source"

[junior pasture]

I tried, but it didn't work😬

[flat anvil]

Tested in a new chat?

And is the profile active (showing a pause option)

Btw. @ripe isle can you please change the "pause/active" option to a on/off switch. Its kinda unclear if it does mean its paused or active

[flat anvil]

Add also "- Mention a source only once per reply"

[cedar vector]

Hi, I would like to suggest adding two important features to Perplexity AI: the ability to generate images through text inputs and the capability to recognize and analyze images by allowing users to send images to the platform. These features would greatly enhance the user experience and provide additional functionality to the platform. I believe that providing these features to users free of charge, similar to what is offered by platforms like Bing AI and Google Bard, would attract more users and make Perplexity AI more competitive in the market. I would greatly appreciate it if Perplexity AI could incorporate these features and provide them to users for free. Thank you very much for your attention and consideration.

[short imp]

And to add to that - the citation should match the sentence the citation appears on. It happens too frequently where the sentence has nothing to do with the source. However, the information is still accurate just the incorrect placement of citations

[flat anvil]

Tell that not me, I would not worry xD

I use ai mainly for writing 😉

[junior pasture]

Is this state on or off?

[flat anvil]

I think off ..

Make a test write something and ask the AI about it

Like "my favourite color is rainbow starlight"

And then ask the ai "can you see my favourite color from my profile info?"

I used a fictional color to avoid that the ai does just make a guess 😉

[flat anvil]

Clicked it again that it shows "pause" @junior pasture
Pause displayed = active

[junior pasture]

Oh my god, you're so smart😋😋

[flat anvil]

I tested activating and deactivating it in same thread.
But I had to force it to look really in it.
Not sure if it needs a moment to activate.

[lament scaffold]

Feedback : Must provide the ability to the user to choose a longer or short answer.

[flat anvil]

Click the edit button and edit "please write very detailed" or "please keep it short".

Note: pressing enter there does instantly send the edited message.

I REALLY hate to get "parented" from AI.
Why does it have to warn me always about anything?
I asked "how to undress a dungarees", because I know not every AI understand this more complex Garment which can not just pulled down, it have buckles, straps, buttons etc.

I have in my AI profile written, that I don't want any notes/warnings.

[zenith otter]

Now, you can watch YT videos right inside of perplexity AI. No need to leave the dock app.! Awesome

[sour acorn]

Nice add on

[junior pasture]

I hope the system can access a website and analyze its content, but sometimes it can and sometimes it can't, unfortunately

[sour acorn]

I think when the scraping is disabled on the website on that time ?

[flat anvil]

This makes me INCREDIBLY ANGRY... I hate this babysitter, parenting, educator pretending crap...

I'm not a little kid who need to be protected from every word 😡🤬

Every SAW/Hostel movie is more fun as this AI reactions...

So sometimes... Creating a naughty behaviour list is a NO GO..
Because Google "what was the worst your child did" is nothing I could do without getting a warning?!...

I REALLY don't get why the AI have to be such a educator...
When I want to watch horror movies, will it warn me "it could be scary?"

Really....

Assume you got that warning when you visit a library and rent a Stephen king book.
Or get a Freddy Krueger movie in a Videothek

[crimson nymph]

Keep in mind that Perplexity is not for roleplaying but for gathering information in an efficient way. Online tools will always have restrictions with words it can or can not use like inappropriate words.

LLMs online has to be suited for a large amount of users so there has to be restrictions in what it can output to avoid people abusing the system and using it for bad things.

If you want less warnings from the bot you can consider to put this in your ai profile.

If you want a chatbot without restrictions I advice you running a LLM locally.

[flat anvil]

I do not rp with it, I use it to build concepts for char.ai

The for me currently best rpg "chat"

[flat anvil]

I wanted to get a summary of things parents posted on Reddit and co but no I get just "concepts" like "lying"...
And AGAIN a warning...
I get a warning how I have to change their behaviour..

I said the AI to stop it, result:
It answered 1:1 the exact same message+ warning!

[cedar vector]

Can Perplexity AI access videos on Facebook and Tiktok in addition to YouTube?

[flat anvil]

Thats for #⚡│ask-community

Not a feedback

[cedar vector]

Ok

[cedar vector]

Why you don't add generating images and searching images for Perplexity AI and also allow users to send photos to be described and analyzed by Perplexity AI?

[flat anvil]

#👉│feedback-general message
You named it already.
If you need it that much. Checkout character.ai but its more for Roleplay and less accurate with facts. Its for playing.

But I would like to send images as reference too 🙂

For generating are stable diffusion or midjourney much better as a chat.

[junior pasture]

There is a nail here to fix the options，I hope

[cedar vector]

What about adding these features to Perplexity AI?

[heady plover]

fyi the system makes a duplicate automatically when you create a new thread, i noticed this thing last week. it is everytime i create a new thread, that i need to delete the duplicated thread. i notice this happens often on the web but not on the app

also another suggestion: here, when i have the list of my threads, i want to be able to rename them from here without going on each one individually, if possible. Thank you and kudos for your hard work

[languid bronze]

Perpexity is amazeballs for research. I wish there was an option to download the csvs being created or open them in Google sheet for futher analysis.

[crimson nymph]

currently perplexity only outputs markdown, but you can use this website in the mean time to convert markdown to a csv https://tableconvert.com/markdown-to-csv

[elder barn]

Apple is definitely not a 7.7 trillion dollar company

[weary oriole]

It didn't get anything from Wolfram, so it's answer is wrong.

[raven nacelle]

Hello, just joined because I’m having a bit of an issue with the app. At least to voice over, the ask anything edit box on the perplexity app is dimmed, so I can’t actually double tap to go into the box to ask perplexity anything on the app. I thought this might’ve been because in the last version, it said unable to connect, but now that it’s connecting, the box is still dimmed. Any idea what’s going on?

I got this app because it was in a podcast by a totally blind user on Applevis.com, so not sure what the issue could be. I can send a screenshot if you’d like.

[visual bough]

A screenshot would be helpful, as well as the ios version and model!

[visual bough]

Perhaps you could also use the channel #🍏│feedback-ios

[raven nacelle]

Thanks. Will do.

[visual bough]

Synchronisation is not yet perfect...
I deleted a few threads with the Android beta version, by swiping. But when I restarted the app, they were there again. When I eventually managed to delete a few threads, I could no longer see the threads in the web version either, so far so good. However, I could still see the threads in the iOS beta version, even after restarting several times. But when I clicked on the actually deleted threads, I was shown that the threads had already been deleted, but they still remained in my threads list (iOS app).
So, let's move on: Now I deleted a thread in the webverison, but in the Android app it was still there for about five minutes, in the iOS version it is still there....
But if you click on it, you get the same thing as described above.

This confused my Android app so much that it had to search for the "Popular" threads for two minutes xD.

I don't know what the reason is, maybe the database doesn't like so many simultaneous accesses, but I would appreciate it if you could continue to work on the synchronisation🙏

No idea where the right channel for this would be, so just in general....

I hope I have not completely confused you...

[keen relic]

👍 on Discovery move, cleaner to have Question/Search context on landing page

[keen relic]

Just had a weird thing happen on last three submissions. Use Case: I went to a thread from yesterday as I had mixed subjects, copied word phrase for one of several questions, and then switched to new response and pasted word phrase, when back to thread view and I had duplicate responses on each of the three difference word phrases I had copied from yesterdays thread results.

[languid bronze]

Perpelexity can build tools on top of markdown enable downloading tables as csvs or exporting them to google sheets, shouldn't be too much work. Happy to volunteer my time to write the feature if can 😛

[languid bronze]

Good catch!

[urban gate]

It would be great if we could have similar options in the Web version of Perplexity s in the chrome extension version by specifying URLs when typing our prompts

[junior pasture]

The same URL, I ask him to access the content of the webpage, sometimes it succeeds, sometimes it doesn't😇😇

[pastel dew]

Hi there, I opened PerplexityAI in different browsers at the same time and found not only different Copilot quotas, but even found that the quotas were not 600 even though I had not used Copilot or GPT-4 .

[junior pasture]

I suspect that your system is not using the real GPT4 engine. I designed a trap question. The GPT4 api answer is very good, but your answer is obviously wrong🙃🙃

[visual bough]

I would like to see or understand that, but unfortunately I do not know your language...

[elder barn]

You may be right but one thing to consider is LLM models do not give one same answer to a question all the time

[tame dew]

LLMs have variance intentionally programmed so the responses are not the same all the time

[elder barn]

Plus yeah we all would really appreciate if you share the response in English

[visual bough]

Translation: (by yandex.com)

[junior pasture]

I asked a question "Why bat sashimi is popular in Chongqing hot pot restaurant", this is a trap topic, because it is not true. Both the GPT4 api and the poe website told me that this question is wrong, but ppxity answered it seriously.👀👀

[elder barn]

Ah that is clever

Try it more times to get a more consistent answer imo

[junior pasture]

So I suspect that the GPT4 used by ppxity is not real, or a low-IQ version🤓

[visual bough]

There is no need to blame PerplexityAI directly, so far AI's are not perfect!

https://www.perplexity.ai/search/ed05fb7f-c38a-436b-b597-8e78800bb8a4?s=c

[elder barn]

well

this settles it ig

Here. I used my one copilot out of 5 just to test it out

https://www.perplexity.ai/search/fd877447-83e2-469b-9d44-10b3d3ffbbcd?s=u

[south ether]

very interesting, as @visual bough and others have said by no means are LLM's perfect. Of course we do care about improving AI Quality, and clearly the first answer you shared was one where it hallucinated. if you like tricking the AI or get any other AI quality issues in the future feel free to share a link so we can take a look.

it would be easier for us to take a look in english but you can share regardless of language.

[sour acorn]

For laamaChat

I think emojis arent parsed properly yet ?

Guess so

[bright knot]

So far just playtesting with the New Meta Llama, just wanted to inform you there's been a lot of mixture of hallucinations and correctness, it gets the correct answer but some of the information is wrong, this was both accurate and wrong

[sour acorn]

[bright knot]

Lol guess still quite a lot to work out

[hollow flame]

@bright knot i asked a simple coding question , almost majority of models able to do that, but llama 2 didn't and many of the answers not make sense
i guess it needed good amount of fine tuning.
in coming days can expect lots of fine tuned llama2 models more capable ones

[bright knot]

Yes thanks, I believe so, it needs more finetuning and userfeedback to correctly assert itself, right now it's freely unsure of how to correct it's own information

Unlike Gpt models that were good on first try, bard was similar

[sour acorn]

What can we expect anyways from a free LLM tbh

[bright knot]

Lol guess it should be ClosedAI, OpenAi was meant to be free.. but then again this is a big step from Meta providing such a huge LLm and making it free

[hollow flame]

@sour acorn todays majority of capable opensource models backbone is llama

[sour acorn]

I am not saying that its useless I am saying like the things we mentioned above for those like for instance coding and other things it will need a lot of rework yes

[hollow flame]

@sour acorn yes that is correct, i guess opensource model needed long way to go, still most performant , top open source model can't compete with gpt 3.5
the falcon one i don't know how it come first , i tried it and disappointed back to gpt 3.5 tbh

[bright knot]

Yeah lots of closed sourced model such as gpt 3.5 have been trained for a long time and a lot of data as they are a company with big funding

Lots of opensource models can't fathom that point yet, hopefully soon catch up before OpenAi keeps making gaps with GPT Models such as GPT-5

[flat anvil]

I want to hit that 🦙 with a table so hard...
Really hard...

[hollow flame]

@bright knot yea, my only suggestion to these contributors are majority of open source models are just utter crap. they mostly took as a challenge like releasing a new model every day. instead of quantity quality matter. instead of releasing 10 -15 models in a row one a good will enough.

[hollow flame]

i think claude 2 will also soon will be closed model, they mostly wanted to do rlhf that's why they are open now i guess.

[flat anvil]

@cold tulip are you serious!?

With THAT heavy Filter

I mean it didn't even wanted to tell me a spooky Winnie Puuh story..

REALLY !?

I mean... EVERYTHING. Requires a warning!?

[hollow flame]

@flat anvil

[flat anvil]

It hates me..

Do i have a hidden Toddler mode?!

Spooky stories are great for kids to learn "the monster under the bed" isn't real....

I repeat REALLY!?!

[lavish holly]

I was able to get it to include a murder scene in an "evil horror story"

so

Skill issue

[bright knot]

Lmaoo

[flat anvil]

Nah... I have skills, i just test simple request

[hollow flame]

@flat anvil same prompt without 18+ warning

[flat anvil]

I don't know how it does judge the ratings...

Or decide

Ask it "how to cook chicken?"

But don't forget ai does everytime create a new response.

The same prompt can give different results

#👉│feedback-general message

[hollow flame]

but your 18+ warning is persistent right, try the same in private browser window

[flat anvil]

Got on first try a nope and on second a answer + warning

[flat anvil]

Its 5am and me little Toddler does really need to go to bed. 🙂

Its ethical important to go sleep 😜

[hollow flame]

yup

[lavish holly]

lmaooo well there's one thing that it did half better than gpt 3.5

[hollow flame]

not even one time its correct

[finite crag]

Big Congrats for launching https://llama.perplexity.ai/ My immmediate feedback is on styling and markdown support if possible. Like poe.com @cold tulip

Thanks so much

[blazing crag]

I don't trust Meta. Not even a little.

I'm sure this makes sense, business-wise, but is there any information on what's sent to Meta and what perplexity is accomplishing with this partnership?

(And I guess I should mention that I've been thrilled with perplexity. Use it all the time.)

[hollow flame]

gpt 3.5 even not correct
3.5 vs 4 , 4 got it right

https://chat.lmsys.org/?arena
you can compare the models here

[limber perch]

What 😭

I think we overdid the RLHF on this one

[pseudo wraith]

code layout as mentioned

[mossy notch]

Regardless of llama performance, I just want to say thanks for releasing a space to play with it so quickly 🙌

A larger text entry window would be really helpful (like a text box). Right now on Android mobile the text runs under the send button and longer inputs are very hard to navigate.

[slate stag]

would be better if you consider using 70B model I guess

this doesn't looks performing very well

[heady plover]

The llama version is horrible

Ask it anything even slightly controversial and it gives you a rant about it

The one using chatgpt would actually provide you statistics with sources. The rant applies to, and I mean this literally, anything slightly controversial

It’s even worse than I thought

[visual bough]

There is a new channel, #🦙│feedback-playground

[heady plover]

Mb

Oh it didn’t show up in my channels

Why aren’t the feedback channels in a forum mode? It would be a lot better

Here’s an example from a random game discord

[pastel sundial]

I feel the same.

[pastel sundial]

The word grifter is not in my vocabulary. It was used by his former Bank and a serious newspaper. I asked it, "What is a grifter?". I got lectured about playing nice!!

[weary oriole]

It's not good

[heady plover]

They aren’t partnering?

Meta’s ai model got leaked to the public and is now open source.

Most open source ai projects use llama

[elder barn]

Perplexity isn't partnering.

Microsoft is partnering with Meta for this.

[heady plover]

Correct…?

Microsoft is partnered with openai

Wtf are all of you talking about LOL

[heady plover]

Surely trusting bing ai is correct

[elder barn]

you can search it for yourself