👉│feedback-general

[hard mirage]

I have set perplexity as default search engine for my edge browser, with the offical plugin in chrome store. but when i search in address bar, the "pro" switch is not turned on. is there a way I can do pro search from browser's address bar

[crimson nymph]

please add https://perplexity.ai/search?q=%s&copilot=true as search engine.

[hard mirage]

thanks, works now. I need to uninstall the extension to change defaults

[candid aurora]

One of the things I'd really love to be able to do is to start a thread in search focus, then once I've got things loaded into the overall window to work with, shift to writing mode from then on out similar to the way that we can try a different model's response mid-thread. That ties in with some of my personal experience with the way each model performs based on the task. Frankly I've been surprised that after a week or so I'm really impressed with how Mistral Large works with search based responses over any of Claude 3, but Claude 3 remains my favorite for working with and developing ideas long term.

There are some specific use cases where this would be particularly helpful in certain domains and industries and am happy to share with anyone if it's helpful on the product side. (edit: I guess you can probably guess which domain and industry I'm talking about without a lot of guessing)

[clear compass]

is it possible to make a different plan? like something in between free and pro? or credit/usage based? that would be really helpful.

[lusty forum]

Hi anyone who can something about this, I'm having an issue with the rendering of equations. When I ask to see an equation for something or try to input an equation, Perplexity will sometimes not render the equation and will just show the raw LaTeX code. If that does happen I have to explicitly tell it to show the equation using LaTeX. It would be convenient if Perplexity would just render the equation using LaTeX.

[old fiber]

I've just paid for pro or access to claude 3 and found I get 5 questions a day? I feel tricked. I just paid $20 for this? Was it mentioned ahead of the purchase that the limit was 5 questions? I would like a refund

[visual bough]

You have received what was advertised ☺️
Perplexity is not there to offer as many models as possible, Perplexity is an AI search engine. A Pro subscription offers you virtually unlimited uses of GPT-4 Turbo/Claude 3 Sonnet/Pro Search (600/24h) as well as image generation, almost unlimited file upload and much more.

[old fiber]

"A Pro subscription offers you virtually unlimited uses of GPT-4 Turbo/Claude 3 Sonnet/Pro Search (600/24h) " 5 messages per day is not viturlaly unlimited. Was it advertised anywhere that I would et 5 messages a day for claude 3? Perhaps I missed it

@visual bough

[visual bough]

It seems you misread it, Claude Opus is currently limited to 5/24h, Claude Sonnet is not 🙂

[old fiber]

Where does it say this pleae?

please

[visual bough]

In your settings

[old fiber]

Not on the pro page here https://www.perplexity.ai/pro

[visual bough]

It says that you can use Claude 3

#👉│feedback-general message

[old fiber]

Just sign in a new non pro account, and looked at settings page, where des it say I would get 5 messages a day for claude 3?

[visual bough]

I thought we're talking about Pro?

[old fiber]

We are but Its no use telling me Im limited to 5 messages a day AFTER I have paid. Earlier you said I got what was advertised, where was it "advertised" (before I pay for pro) that I would get only 5 messages a day

[visual bough]

Please just read my messages, I already explained it here: #👉│feedback-general message
Only OPUS is limited, BOTH are models from the Claude 3 family

[old fiber]

I understand that, I am asking where this was "advertised" ahead of time. It said "Access to claude 3", you don't think its misleading to not mention the 5 messages a day limit to the opus model? To me this is obvioulsy misleading. I want to mess arond with the opus model and believed I would have reasonable access to this.

But not to worry, you probably just think I like to complain for no reason, I've requested a refund via the settings page. This was clearly misleading in terms if what you get for $20 a month

[visual bough]

Claude 3 includes three models: Opus, Sonnet and Haiku. Writing that Claude 3 is available is therefore not misleading.
You get access to Sonnet and Opus and, as I said, the number of Opus uses increases over time.

[old fiber]

Yeah, no need to mention just 5 messages day to opus I guess. No one deciding whether to pay $20 a month would want that information I guess.

suprised this is not obivous, but not to worry, Im done here, hopefully I get a refund

[noble monolith]

https://x.com/aravsrinivas/status/1765068749282201660?s=46&t=JsxhFTRLBknd8RUv1f73bA

“We are working on increasing the daily limit on Opus to more than 5, please bear with us until then”

Aravind Srinivas (@AravSrinivas) on X

Claude 3 (Opus and Sonnet) is now available on Perplexity Pro. Instead of seeing Claude 2.1, you will now see Claude 3 as a choice for the default model and the option to rewrite an existing answer. Your first five daily queries are routed to Claude Opus, the most capable and…

Twitter · Mar 5, 2024, 5:35 PM

[old fiber]

It's good that you are trying to increase the limit. But this limit was clearly not advertised on the sign up page and to be told I got what was advertised is just insulting. I won't be back.

[noble monolith]

you can contact support via mail for refund, meanwhile please take your time and explore what our free version offers

deep dive ➤ #💬│general message

[old fiber]

I have requested refund already. I hope I will get it. But after being misled on sign up and then told that I got what was advertised when asking for help, I'm not inclined to stick around and play with perplexity.

[noble monolith]

understood, please accept my personal apology for any inconvenience you have experienced. team is constantly adding new features, which are not even followed by official announcements #💬│general message

[warm magnet]

Hello

Please add a button to scroll down to the bottom in the android app. It's a bit difficult to have to scroll down manually if there are long prompts

Thank you

[frail fossil]

I would like more than 5 uses of claude opus per day

[sand locust]

Please don't listen to jerks like these and remove access to Opus even if it is 5 messages a day. You cannot advertise every single thing on the signup page. This should have just been an email and a request for refund but I guess people just need to prove they are morally superior than others. Obviously we want more than 5 uses of opus but 5 were appreciated for the time being.

[dim sierra]

Yep, people are just really impatient nowadays

Literally got unlimited Claude Opus today.

[winter cipher]

All they had to do is wait 24 hours lmfao

[viscid perch]

My Replit perplexity.ai 4 month code says it has expired before ive used it. Is there a reason why? Can I get another?

[trail fox]

Hi, actually to generate an image is necessary also to generate a language model answer.
Wouldn’t be more efficient to have separated things?
This is what I do now:

1. Make a prompt “generate an image of X”
2. Wait for like GPT4 answer “I’m sorry I cannot generate images”
3. Tap on “generate image”

I think this generate costs for Perplexity (LLM api tokens) that are completely unuseful for the customer.
What do you think?

[visual bough]

#⚡│ask-community message

[slate dew]

Hey, I would never have thought of paying for AI but Perplexity has changed my mind: one month PRO subscription and no turning back!!!! Congrats for the work

[dim sierra]

Yep, it's pretty great.

Would like a better implementation than collections though.

[slate dew]

Yes, I guess it will soon become a bit limited.

[vernal veldt]

Japanese text is wrong.

Claude 3 Sonnet: Latest advanced model by Anthropic
Cloude 3 Opus : Latest Highest grade model by Anthropic

"""
Claude 3 Sonnet: Anthropicの最新モデル
Cloude 3 Opus : Anthropicの最新最上級モデル
"""
Is it like this?

[covert sun]

Thank you. This part needs to be updated in all of the localizations. We'll fix it.

[void moon]

Rewrite only claude 3 sonnet without opus

[dusty cedar]

maybe it is implied by the discover feed, but i'm curious to what extent the script is also AI-generated (as i'm not familiar with the discover feed - so maybe that could be made clearer) - thanks

[dusty cedar]

regardless, i have been enjoying the podcast!

[visual bough]

Discovery feed is 100% AI

[wheat radish]

I have met a problem for several days. My app only shows answer in photos and videos, no texts. But the web page in browser is normal.

[dim sierra]

Reinstall or update the app?

[wheat radish]

Yes, I even redownload today, but still

[visual bough]

Please share a screenshot

[visual bough]

Do you use a VPN?

[wheat radish]

Yes

[visual bough]

That‘s probably the issue, please try without.

[wheat radish]

But I also use it on web page, that doesn't go wrong

[visual bough]

Yes, but please try it without VPN. Browser and App work differently.

[wheat radish]

I see, thanks a lot

[candid aurora]

passing on some more positive feedback, i love the voice that comes out of the iOS app and someone should tell the folks at Readwise how great it is. High hopes for that way of interfacing both for search and just for text generation and this is a heck of a good start. The fact that ChatGPT's voice interface acts like it's a telephone call in the phone or the kitchen is absolutely killer and something to work towards if i were building all this out

[raw flame]

So I've been using perplexity on a chromium browser in macOS for a few days now. When I'm doing a search that really needs analysis it gives me a great head start and has true value. It does have some quirks but overall I'd say it adds value to my search/llm usage.

However; sometime I'm doing a "lazy search" where I just want 10-15 direct results because I already know what I'm looking for. It's normally in the top 5 results, but its quicker to just search it rather than trying to guess the FQDN or hunt thru the bookmark graveyard. I do this with image search as well - i need like 20 results fast, I'm using the image for a specific goal, not for exploration.

Perplexity isn't great for these cases, but I don't expect it to be. I'm suggesting a fast-fail feature combined with query complexity test

• Use some heuristics decide if I'm doing a dumb search. if I search arch linux download I'm probably just trying to get a link. I'd love to not cause LLM costs for that
• If you make the link "pool" something I have to manually click to expand this can provide extra signal of intent to fast-fail the LLM language generation
  - I'm aware that it's already a clickable section, but often times I get < 3 extra results. I notice I only get 10+ when I search something like "news" ]
• If your business model is married to the concept of having language generate immediately after every query, then use a super cheap model for the 1st 2-4 sentences, and if i haven't navigated away, then switch models. You could even do this on "smart/complex" queries but just start spooling the quality LLM in parallel with the "Intro LLM". It sounds counter-intuitive, but if Im busy looking at links and normal search content for 5s I dont feel like I'm waiting, so when I start to watch the LLM I'm only noticing how fast text is streaming at that moment,
  - This relies heavily on giving me enough to interact with early on (i.e. more quick links and photos). If i have low results and low images, im gonna notice that early LLM fakeout. Increase the amount of "snippet" text from the links at the top at the top to easily cover this time.
• I'm aware I can still use a normal search engine, but the appeal of having 1 default engine to rule them all is why I pay.

I like to support search engines that are exploring a different business model - I pay for Kagi and prefer it for my "dumb" searches, I'd love to just have my searches automatically routed to kagi or perplexity based on a quick complexity test. Having a single default search engine that can do both is what makes it have value to me
I'm afraid if I have to make perplexity a manual use choice - as in not my default search engine - I will kinda just forget to use it - it starts drifting into regular LLM territory, where I feel like I can just combine my kagi results with manually questioning my highly configured GPT4-32K deployment

[marsh sail]

When entering a supplement to a question suggested by Prosearch, in Japanese, if I confirm the conversion and press the enter key, it is sent and I cannot send a complex sentence.
I remember that Perplexity had a similar problem when the service was launched and it was quickly corrected. (I appreciate it very much).
This problem also occurs in Perplexity Labs, and it would be very helpful to me as a Japanese user if you could fix it.

[vernal pasture]

when using perplexity behind a proxy, this pops up constantly, is there any workaround to fix it ?

[rapid snow]

Can somebody help me get access to pro support channels on this server? I'm unable to get access either from web/mboile app, even tried leaving server adn joining again

[visual bough]

Please leave and rejoin via the link at https://www.perplexity.ai/settings/account not with the link at the start page

[rapid snow]

I've done that, that's not helping either

Tried it again, this time it worked, thanks 🙂

[broken hedge]

When I run a python app using the Perplexity API as my search tool, I can't see anywhere in Perplexity where I can see that API usage like I can with other API providers. Because if this I can't be sure that the answers I am seeing are being provided by Perplexity or just fallback to the AI models own knowledge, nor can I determine how expensive this app is going to be to run.

[knotty minnow]

it won't show whether it used a 'fallback' from its training data to answer a question, but for monitoring usage, click the Dashboard button highlighted in the screenshot
using an online model, you should expect that the prompt you sent gets injected with some kind of search results; how and whether it should use them can be unpredicable, but if the basic intent of your query is clear (and relatively simple), the onnline models seem to a pretty good job

[vernal pasture]

no one can help? I really want to pay for it, but this is super annoying….

Why Cloudflare would like to verify constantly when behind a VPN ? There must be some workaround for this, there are tens of apps which utilize Cloudflare and can be used smoothly behind VPN

[noble monolith]

use diffrent VPN provider, e.g. Proton

[azure iris]

Is it possible to turn off web searches? Sometimes I don't need real-time data and web results are just a distraction

[visual bough]

Yes, with Writing focus.

[azure iris]

Oh, I just noticed this "focus" feature now that you said. Thanks!
Making it more intuitive would improve UX btw

[wicked musk]

Add a feature that allows the user to input a website URL, that the AI uses as its only source to respond to the users question.

[azure iris]

I usually go to the website and use the perplexity extension to get answers grounded on the page content

[wicked musk]

Good shout, I forgot about the web extensions capabilities. However, it still would be nice to have this integrated within the UI.

[azure iris]

True, specially if you need to use multiple custom sources at the same time

[vernal swallow]

Hey, can I set Perplexity Pro as the default? Right now, Perplexity is my default browser, so my first search is always with the less powerful model. Also, toggling on Pro each time feels like a unnecessary hustle in general

[visual bough]

Perplexity Pro is not a separate model, you always search with the model that you have selected in the settings.
To always use Pro Search (which I do not recommend) you have to add &copilot=true:

https://www.perplexity.ai/search?focus=all&copilot=true&q=%s

[knotty harbor]

first time ever I've seen this... nothing works currently (pro]

[fervent nacelle]

they were doing an update

[rugged steeple]

iOS needs Perplexity app in Spotlight search! Devs, add this feature.

[noble monolith]

https://www.perplexity.ai/hub/faq/what-is-copilot

[wraith raptor]

Is it possible to select a specific collection to search with by adding an additional GET parameter (e.g. &collection=...)? I tried it but couldn't get it to work.

[fossil pumice]

https://www.perplexity.ai/search/what-is-545-VMpBAIiWRp6SWb7_xOx7Hw
issues with currency conversion in pplx

[meager valley]

Please add the possibility to give a name to an API-Key. The organisation is terrible if you just see multiple keys at once without naming.

[ashen hinge]

for some weird reason the command prompt does not work anymore..

like bruh it used to work but idk it is useless now..

like you might aswell get rid of all collection if you are just gonna disable the Command prompt

[covert sun]

Hey, @ashen hinge! Thanks for sharing, do you mean your collection instructions are not applied properly? Which model are you using?
Could you please add more details in this thread in that case: https://discord.com/channels/1047197230748151888/1218132377684152362

[covert sun]

Hey, @burnt coyote have you seen any issues with models repeating answers recently? Please let us, this should have improved.

[somber halo]

Is perplexity responsible for discover daily?

I have feedback about the podcast, there is no date mentioned at the start or during the podcast. I don’t listen daily but listen to a few in a row, and it would be great to know the date on each podcast while listening. For archival purposes also. I understand they are released daily but other than release date there seems to be no indicator

[ashen hinge]

I am using perplexity, and when I go onto the command Prompts for one of my collections for example "the political collection" everytime I type on that collection and I create the prompt the Ai is only ever following the Prompt for a super short period of time like only for 3-4 Queries now and then it deviates and stops following it.

this is what i get frustrated by and my command prompt is not even difficult to understand it is just "In this Collection you must provide 3 detailed paragraphs, for every Answer for every Question. the first Paragraph is a Introduction paragraph which must provide an introduction to the Essential Question while providing the why the question is important, and fundamentally interesting. The 2nd paragraph is a Body paragraph that must Answer the Question in explicit detail while providing information, statistics, percentages, formulas, and references to the Essential question. the 3rd paragraph is a conclusion paragraph which must provide the conclusion for the overall Essential question, and restate the informative Answer and details to the Question." thats all

[covert sun]

Thanks for the feedback, @somber halo! I'll forward it to the colleagues working on it.

[covert sun]

Thanks for providing more details, I'll add it to our ticket.

[glacial spruce]

Claude opus 200k is being used probably on the api front the issue is the file analysis thing. It isn't updated to support larger than 32k context files since you guys use a private 32k context model can you guys update it for 200k

[ashen hinge]

Ima keep it a stack.. the command prompt system is probably the dumbest, weakest, and least affective aspect of perplexity. like literally

Im better off just paying for the Pro-Version

Ima be totally honest.. like they is no point having collections at all.

[quick axle]

I did a little test in searching for information using phind compared to Perplexity. I used a random fandom site. I asked to look up information about an artificial intelligence character from Carmen Sandiego's 1994 animation. For some reason, Perplexity couldn't get the information from Fandom, while phind could.

Another minor difficulty with Perplexity is that sometimes it.tek has difficulty relating contexts even if the context is the last message. Sometimes it seems to get a little lost. Although sometimes you can correct yourself by editing the message and asking to relate.

In this type of research, in fact, the phind has done even better than gemini and copilot.

But Perplexity still stood out from the two for being concise and bringing closer and less hallucinatory answers.

But I don't know if right now he's having that kind of difficulty.

But despite being an excellent searcher , the phind is sometimes a bit dumb to understand a direct answer.

[steel pewter]

I mostly use perplexity pro for creating a narrative/writing. And although mostly it’s done a fantastic job with its responses, my issue has been the bots' repetitiveness. It often generates the same words despite my instructions to avoid that. I’ve found I’m not the only one who’s ran into this problem. It’d be nice to have a system implemented to either edit or adjust the temperature of the bots!

[quick axle]

How does the temperature system work?

[quick axle]

Tks

[covert sun]

Hey, @steel pewter! Could you please share some threads, the team has been working on it, we'd like to look at your examples, too.

[zinc gazelle]

Is anyone else in line to get a rabbit r1 next month? It uses perplexity so I’m wondering how easy it would be to come up with a way to use them together.

[gentle seal]

I keep having problems with trying to ask follow-ups. it seems to always lose context of the previous answer, so I can't ask questions about it without copy/pasting. this is especially challenging when i'm trying to do some quick research when i'm on my phone.

[covert sun]

Hey, @gentle seal! Are you using the "Writing" focus or "All"? If you are searching online, you might wanna refer to a specific message or part of the previous message. From what I've seen, sometimes users expect the model to answer only based on the above messages not realizing that their new search could be interpreted in a broader context or unrelated to the previous messages, which is also possible in the same thread.

[gentle seal]

@covert sun using "All." My expectation is that "ask follow-up" will ask with the current thread history as context, and that a new wholly unrelated search would require a new thread

good quick example where the followup had no relation to the first question+answer. I have to do what you said which is re-provide the full context. https://www.perplexity.ai/search/what-public-companies-ursNuiK_TtiQpAtVoOv2UA

instead of using the first response as context, it did a brand new search and ignored the context provided. my 3rd message yielded what I expected from my 2nd message follow-up

if the design intent is that entering text does a new search, it should not be labeled "ask follow-up" but rather "new search query"

I'd obviously prefer context-aware follow-ups! If that's not feasible in the short term, re-titling the placeholder to better set expectations would be helpful.

[ashen hinge]

If you ask a completely different question that deviates away from the initial question. then you cant expect perplexity to use Context from the initial Question to answer your 2nd question. the reason why it tells you to "ask a follow-up" is because it assumes that every Question that you are going to ask within the digital confounds of that thread is going to heavily allude to the Essential title, of that thread. but if you ask a Question about cars, and then you're follow-up Question is about Government structures. then how the hell is perplexity supposed to draw any connection between them???

like it makes no Logical sense

like "what companies make abortion drugs" alludes to things like plan-parent hood, and abortion. it has nothing to do with "stock tickers", stock tickers are a super generalized and broad thing..

you cant expect Perplexity to make the most Insane Connections between Topics, and subjects with the most nugatory inter-aspect connections.

[gentle seal]

public companies and stock tickers are not completely unrelated. Instead of making a connection, it started a new search on an unrelated topic. chatGPT handles this connection without any problem, as does google. I can expect perplexity to do the same.

I came here to share feedback, not to have someone ask "what the hell is wrong with you"

the problem here is that perplexity failed to identify "their" in my follow-up as referring to the list of public companies I asked for in the first question. There is a clear semantic connection between the pronoun "their" and the subject of my initial query.

I've been on the internet too long to get into a debate about this. I am firm in what feedback I have to share.

[ashen hinge]

yea. but you were not talking about Public Companies, in the generalized sense, you specified and started talking about public companies that manufacture abortion stuff, which alludes to things like abortion, and plan parenthood. if you were to have a conversation with someone, and you asked them, "what are the different companies that are responsible for the manufacturing of abortion drugs. and then they gave you the answer. but then you started talking about "stock tickers" they would look at you weird and ask the question of "what does that have to do with that":

[gentle seal]

I'm not debating this with you.

[ashen hinge]

i never said "what the hell is wrong with you"

[gentle seal]

it's literally in my query.

blocking

[ashen hinge]

ok?? im just trying to explain why you are experiencing this problem

[ashen hinge]

and when i said "how the hell is perplexity supposed to draw those connections" I was not attacking you

[ashen hinge]

Waw. so you got so offended despite the fact that I did not even insult you to the point that you blocked me

and all I was doing was answering your question to why certain things are happening...

and you got soo mad

for no reason..

I have never communicated with someone who got offended over a Logical explanation.

with Zero insults..

[gentle seal]

for reference, here are screenshots of other products' output that make the connection between messages in the same thread

[random pine]

Wow

Imagine deciding to block someone as a result of them logically explaining things to you…

Like Waw

How can you even have a conversation with these type of people???

[gentle seal]

they were not logically explaining things, and as I said I came to share my feedback of my experience not get into a debate.

[ashen hinge]

exactly

[ashen hinge]

i dont know what his problem is

[gentle seal]

I gave a clear example of the problem I was facing, and some examples of expected behavior.

[random pine]

He didn’t even insult you..

[gentle seal]

so?

[random pine]

But you didn’t have to block him

[gentle seal]

not here to debate that either

[ashen hinge]

what clear explanation??

All i said was that if you ask completely unrelated questions dont expect perplexity to derive the same context...

simple as that

and you got mad

[random pine]

I agree..

He probably Blocked me too

[ashen hinge]

LOL I could imagine

[quick axle]

I think there would be ways to solve that. In a casual conversation, the subjects are expected to be correlated. Although sometimes the subject is changed. There could be a certain degree of tolerance or as the colleague mentioned above temperature or something that made Perplexity have more or less freedom to explore something out of context or not. Because sometimes it's very annoying to have to make him come back to the subject of the topic.

It's AI function is to better understand the context and the form of human communication in order to adapt to different contexts. According to the way you express yourself. While this doesn't get smarter, you can create tools that at least make this as efficient as possible.

[gentle seal]

Here's an example of it working as expected, where it found a semantic match between the subject of the first query and the pronoun of the second:

even though it didn't do what I asked the first time 😂

[ashen hinge]

dude it answered your question. what are you complaining about?

[gentle seal]

to be clear, I don't expect this problem to be solved in discord. just sharing my feedback in the hope that it will help improve the product.

[covert sun]

Thanks for sharing the thread, @gentle seal! The issue here is that the model concludes those are publicly traded companies in the first answer. This example is a bit better: https://www.perplexity.ai/search/what-public-companies-NgBLMkZhQaiX5_H1vLeRLg so that follow-up is redundant in my case. I used GPT-4 and it's similar to the answers you got with other services. The clarification in your thread makes it just right. When models consider new search sources and context, that clarity helps.

If you are referring to the ChatGPT experience, you can easily repro the answer in the Writing mode. In that case, there are no other sources to check other than the context of the thread, that's why you most likely won't have this issue, but that might be a not up-to-date answer:
https://www.perplexity.ai/search/what-public-companies-UqbhskNHRZCAu1l8zfS4OA

[gentle seal]

I will play more with the writing mode. thanks!

[half thunder]

can increase the number of sources to 10 ?

[random pine]

Dude blocked 2 people…

Ridiculous

[knotty minnow]

The follow-up question is like perfectly aligned / consistent with the original question ... it could not be less of a deviation 😅

@gentle seal btw, I think the actual answer to the question is that are there no such public companies.. (i.e. there's only a handful, and they're private) and so no tickers. GPT-4 and Opus both said something along those lines anyway
(but that doesn't change the fact that it seems to start anew with your follow-up question, providing tech stock tickers etc)

[silk aspen]

Thank you for doing the repro and humbly putting many minds to rest tonight

[inner wigeon]

Experimental was going really fast earlier in the day but it seems to have substantially slowed down, at least right now

Oh, interesting, it's just that Experimental is slow when used with Pro

In Pro, Experimental is probably the slowest model of them all 🤔

[knotty minnow]

yep. the screenshot is a bit of mess, but you can you see that, substantively the answers are basically the same ("What are the dates for India's election?") for Experimental (Pro OFF) vs sonar-medium-online via Labs vs sonar-medium-online via API

With Pro enabled, Expermental is still faster than the third party models, but nowhere near as fast as it is with it off (near-instant answers); the responses also diverge with it enabled (presumably due to the presence of additional sources, as well as any clarifying questions and whatever else happens with it enabled)

[inner wigeon]

The 3P models are responding substantially faster with higher token rates for me with Pro

Both in-house models + pro are going substantially slower than any of the 3Ps

[knotty minnow]

that's an interesting observation 🤔

[inner wigeon]

There's something off about in-house + pro that I think is beyond just the addition of RAG

[knotty minnow]

hmm you might be onto something there. Like it sounds like the in-house models seem to work better without including the Pro system in the process; whereas the opposite may be true when it comes to the third party models

don't think i explained that well ha but I think there's something to what you're observing

[inner wigeon]

Video comparison of Experimental + Pro vs GPT-4 + Pro

[knotty minnow]

the in-house models (well I don't really know about Default; but Experimental) has like an in-built RAG system of some kind. So when you make an API call, the model basically has the index on hand (or within its architecture) and somehow relevant snippets are pulled from it and injected into the prompt/context. But with the third party models, it has to be happening as extenerally..
So I think with Experimental + Pro, it's kinda adding an additional RAG process, which a) slows it down, and b), seems to result in poorer quality outputs sometimes

[knotty minnow]

wow yeah like 5 x slower 🤨

[half thunder]

https://www.perplexity.ai/search/How-Can-Sage-OHKMqBTkTdaK.NX1yoXFmA#0

Why is there no source citation information on the results page of an academic query?

[shut oak]

I would love to see a button or a function to export the citations to a reference manager such as zotero or mendeley. This would save a lot of time. Is there a way to do this in a conversation thread? @covert sun

[proper arrow]

Hey my account just got deleted by single click how I can restore 🥹🥹

[visual bough]

You can contact [e-mail] 🙂

[proper arrow]

What they will do

[proper arrow]

I have used one year gift card to pro

[acoustic patio]

This is a bit confusing, do you get 5 Pro searches every 4 hours and 1 per every 4 hours, or is it 5 for a whole day, or 5 every 4 hours?

[crimson nymph]

Each use will come back after 4 hours you used it (in the free version). So it won't reload to 5 exactly after 4 hours but each use will restore 4 hours after it's used.

[dim sierra]

Yep, basically each prompt has a 4 hour reload time.

[acoustic patio]

That's strange, I was able to use two within a few minutes.

Ohhh, it gives one back four hours after each query used, got it.

That is really confusing, but I don't know the best way to communicate it either.

[hushed spindle]

yo idk if i just can’t read but maybe this section should be clarified a bit more loll

[quick axle]

I had made a comparison with Perplexity's search and phind with the fandom phind site had done better. I now asked Perplexity to make a table with the description of a site with which I showed the direct link, and in this aspect Perplexity was much more accurate and phind didn't understand anything.

So it's a little confusing. I don't know exactly why one did better in one activity and while the other did better in another.

I thought it was fair to make this assessment to explain that there are weaknesses and strengths. Although I still don't understand why one stands out in one type of analysis over another.

[pulsar hamlet]

Hi, did you change your model today? I found the answers quality substantially decreased today. I am a power user, for the past week I used to feed around 100 papers to perplexity and it can seize the key words and related concepts very quickly and accurately, But today I realized the perplexity just read the first a few pages and become “ lazy” and stop to continue to read my information. I am wondering is that because you change your model?

[noble monolith]

please don’t post same question in several channels, thank you

[pulsar hamlet]

Ops, sorry. I don’t know which one is correct. Thanks for reminding.

[steel pewter]

Very sorry for the late response! These are what I could find so far. The bots have a tendency of repeating a string of words after some time. The more I respond, the more it grows a habit of doing it consistently. I’ve tried tweaking the prompt to avoid the problem, but that doesn’t seem to help either.

[topaz violet]

True, it's very annoying that it generate very often a testament or a character's determination. So repetitive...

[topaz violet]

Even if you confront the AI about the repetitive words, it will make up false promises and proceed to lie before litteraly generating the mentioned repetitive words again 😤

[clear compass]

hey I just purchased the pro sub, one thing noticed, I prabably hit the payment button more than 5-6times, then it worked successfully,

before that, was getting error like error while payment, try different card, but my card is fine

just letting you know guys

oh, one more thing, while doing the last attempt, a captcha appeared, I checked, that's why it worked I guess

[sly vector]

Yeah I had a similar experience. I’ll pass this up to the team.

[robust pelican]

I am using the academic feature with the pro turned on. perplexity seems to be having trouble responding to the query that is a follow-up from the previous question. It starts a new search and replies like a new thread/chat

[ebon gorge]

Hey team, I paid a failed invoice from my mail but it's not credited the tokens to my account

[ebon gorge]

Hey @south ether , Sorry to tag directly. Saw that you are incharge of feedbacks.

I'm having a clg hackathon tomo and I need the credits in my account. I've paid the invoice, but the credits aren't reflecting in my account.

I'm open to sharing the invoice on request. I can't afford the time delay from email support

[cedar cipher]

The image generation feature is way too complicated. This should be accessible as part of the regular search function. i.e. instead of typing a question, the user would be able to type a prompt to generate an image

[crimson nymph]

You can already do that by clicking on this icon

[covert sun]

Hey @ebon gorge! Could you please dm me your account details.

[ebon gorge]

Yep

[south ether]

Your credits are good to go 👍🏻 let me know if you need anything else

[topaz violet]

I don't think the Negative evaluations are doing anything, is this normal?

[arctic ocean]

i have a perplexity pro subscription. the pro switch constantly toggles off when i start new searches and i have to toggle it back on. there doesn't seem to be a way to remedy this through settings. any solutions?

[visual bough]

It's not possible to constantly enable pro search

[dim sierra]

Strange, mine is always toggled on

[dim sierra]

Add &copilot=true to the url

[fading python]

Hello I have preplexity set as default search. When I enter a query in my browser it searches for information without pro status. I have to turn it on manually and re-enter the query. Is it possible to have "pro" automatically enabled?

[dim sierra]

Literally read one message up lol

[glass sorrel]

God, it's interesting to set up an AI bot to respond the way you want and change the earlier messages to be more useful for later. I never thought about doing anything like that but it's cool. You don't run over yourself and you have pre set up AI as well

[topaz violet]

I don't think the Collections - AI Prompts are working as intended. 😤
source is [GPTDevil]

[fading python]

How do it?

[dim sierra]

Add it to your url.

Are you using a browser shortcut for the search?

[fading python]

I use a chrome extension to set ai as the default search engine

[dim sierra]

The perplexity extension ?

[fading python]

yes

[dim sierra]

I would recommend setting it yourself, if the extension doesn't have settings to do so.

[covert sun]

Maybe it's still imperfect and needs more prompt and editing work, but Collection prompts are working (no 'challenges' spotted). You can check it in the last response:
https://www.perplexity.ai/collections/Dark-stories-kTnnf7AHQymaCJsPtWXexQ

[knotty minnow]

fwiw I find prompts that show what the model should do, rather than just say what it should not do, usually work better (though if you yell at it loudly and repeatedly enough, you can often get them to what you want, albeit with otherwise degraded output quality). fwiw I gave that prompt (what was visible in the screenshot anyway) to Anthropic's metaprompt thing and got it to transform it. then made a few tweaks to get it under Perplexity character limit:

Your primary role is to write dark stories, based on user prompts, focusing on character interactions and darkness. To ensure originality and depth in storytelling, NEVER use cliched words/phrases. Here are some alternative words and phrases that could be used in their place:

<words_to_use_instead>
- **trek, expedition, odyssey** [instead of journey]
- **reverberation, reflection** [instead of echo]
- **tenacity, perseverance** [instead of determination]
- **grit** [instead of grim determination]
- **sharp dichotomy, striking juxtaposition** [instead of stark contrast]
- **difficulties, ordeals, hurdles** [instead of challenges]
- **trials that awaited, obstacles ahead** [instead of challenges that lay ahead]
- **awaited, loom** [instead of lay ahead]
- **a sign, a marker, an indication** [instead of a testament]
- **tacit** [instead of a silent testament]
- **cruel, vicious, brutal** [instead of malevolent]
- **fortitude, firmness** [instead of resolve]
- **reinvigorated** [instead of renewed purpose]
- **rekindled commitment, second wind** [instead of renewed resolve]
- **poorly illuminated, dark** [instead of dimly lit]
- **unbreakable will** [instead of indomitable spirit]
- **unflappable** [instead of unwavering spirit]
- **steadfast commitment** [instead of unwavering focus]
- **twist of his hand** [instead of flick of his wrist]
- **the strain was evident** [instead of The tension was palpable]
</words_to_use_instead>

NOTE: The EXAMPLES above should help to ensure that your responses steer clear of trite expressions. They are NOT meant to be used as a rigid word substitution template. As a creative writer, it is your job to independently arrive at appropriate non-cliche words to use in your outputs. 

---

You will now receive the user prompt. Please craft a dark narrative that is engaging, unique and employs nuanced and distinctive descriptors, without being overly colourful or stereotypical.

I dunno, seems to kinda help https://www.perplexity.ai/search/Provide-the-opening-cxnjFTQrTJuxRjmpJL_MrQ

But I notice that there are other words that uses repeatedly, like "inextricably".. Kinda think you would best of using an API, and playing around with parameters like repetition_penalty and temperature – tbh changing those values could help more than any particular prompt imo

[warped hornet]

Connection Timeout.
Is the server down?

Stuck on “Pro Search. Understanding question…”

[deft flower]

Server might be down, mobile app not working and search with safari not working

[unreal spruce]

yeah still down 😦

[lilac coral]

Sure, but it might be more convenient if it were possible to do this directly from the chat window, simply by asking for an image to be generated and receiving it in response. Perhaps this feature is still ahead of us 😉

[visual bough]

Please check #💭│ideas message

[covert sun]

Hey, @unreal spruce, your report was submitted later, are you facing any issues with it, at the moment?

[lilac coral]

Thnx. I understand, am I to assume that I should treat Perplexity solely as an advanced search engine? A search engine that provides elaborate responses is cool, but ChatGPT can also search and evolve, potentially becoming much more versatile. For now, Perplexity has made a great move with Claude, but we'll see.

[visual bough]

Yes, Perplexity is an AI search engine, ChatGPT is a chatbot.

[lilac coral]

So, why do we even have this writing mode (focus writing mode), etc., and the ability to write code? I think people expect more than just search results being delivered. It's hard to manage several services at once 🙂

[visual bough]

Perplexity is an AI search engine, the focus is not on normal text generation (writing). However, this function is still offered because it can be very useful.
Nevertheless, the company's focus will remain on providing quick, factually correct answers, with citations, to user questions.

[lilac coral]

Since it's about giving feedback, having image generation would be equally useful. It's a similar addition to writing. 😉 I understand that it's not the main focus of Perplexity, but everything is interconnected, and you can surely see from the posts here what people use Perplexity for. In any case, thanks again for the info.

[noble monolith]

most convenient atm

Create image using precise wording provided in the prompt

[zenith needle]

the mitigating factor in ai is our prompts right? so what if perplexity got a prompt improver buffer. you enter a prompt and then it improves the prompt, then you can make changes and then send that improved prompt

[versed thunder]

I'm using the chrome extension. I love it. But the prompt and the result keeps disappearing when switching tabs or windows. It is annoying.
It would be better if previous prompts can be brought back via a drop down menu.

[knotty minnow]

totally agree. I found myself stop using it because of this...(i'm always jumping between tabs..)

though that said, I just noticed that interactions with the extension are automatically added to the Library on perplexity.ai. Not sure if that's always been the case.. But it's nice to see (and ig kinda resolves the disappearing response issue for me tbh ha)

[boreal vigil]

I think the current usage limit (600/day) cannot be checked when on mobile. Even on browser version since there is no "hover" option on the phone.

[heady plover]

if you turn off pro then turn it on, it should tell you

in the app

[boreal vigil]

It doesn't for me :/

[covert sun]

It should show on the web for sure. On the apps, you'll see that pop-up about remaining uses, only if you get to 5 uses left.

On the web you'll see the number when you hover on the Pro toggle

[glass sorrel]

Damn, I can only give 3 links per prompt? It would be nice to be able to actually use more

It's not like the AI won't find 20 sources on it's own

[boreal vigil]

On desktop yes, but I meant web on the phone device, since hover is not accessible on the phone I think, it just won't show up.

[topaz violet]

I'm pretty sure it's was working before as i was able to generate 37 whole prompts of detailled violence and fighting scenes with GPTBlaster/GPTDevil, custom collections that has the same AI Prompt. Both are obedient with Claude 3 Opus
but in this new topic, it's refusing for some reason when it's was previously working?

[dim sierra]

Yep, since they don't want it to do so. Just means they patched it.

by they, I mean anthropic.

[half thunder]

can pplx support search website like cohere's coral ?
or on the roadmap ?

[knotty minnow]

I had a 'no yapping' prompt applied to a custom collection, which used to work quite well until last week

I've tried updating the prompt but gave up - it doesn't seem to work any more (at least when Pro is toggled ON).. Basically seems that the instructions have literally no influence on the final output whatsoever

Not sure what changed, but I don't think it was on Anthropic's side (not entirely sure, but it seems only a problem with Claude 3 models)

[heady plover]

did not know about coral

pretty cool

[heady plover]

does it only support one a time though? and you can manually do this in perplexity using site: operator

[topaz violet]

Maybe the Collections need some fixing from Perplexity.

if it does not influence on the final output

[knotty minnow]

hmm yeah I dunno. Thought it was related to Copilot/Pro, but even with that OFF, Opus won't follow the instructions (which explicitly and repeatedly state, "Do not begin responses with:

Here is a list of XXX
")

Seems a Claude thing ig

[remote fern]

I know we cant be political but love the caption

[distant osprey]

I've have the same issue, it will not take my prompt in collections. When i do a follow up question, and ask it to use the prompt, then in most cases it does... I am also using claude 3 opus

[half thunder]

site: operator ? any documents about this operator ?

[visual bough]

https://www.herohunt.ai/recruiting-glossary/site-operator#:~:text=The "site%3A" operator is,site%3Aexample.com".

[half thunder]

https://www.perplexity.ai/search/site-httpsbreakingmu-what-McQyvFNDSjGQAjejssR1EA#0
it worked! site: https://breakingmuscle.com/

[loud pollen]

Hey, Perplexity peoples, don't sell out to Apple. That would suck.

[lusty prairie]

Hi Perplexity Team, are you aware of the bug, that when you paste long text on iOS in the app there is no way to send it?

[visual bough]

Yes, it will be fixed soon.

[remote comet]

Not sure why Perplexity can’t answer questions like this properly considering the schedule was published months ago

https://www.perplexity.ai/search/What-are-the-_QyOgeCLSFeizkQrXlgH.g

GPT4 with Browser, Copilot, and Gemini gets these types of questions like weather, etc. right

[topaz violet]

What does that even mean? These non-sensical words... 😭 Why can't i just ban this one single word with a Collection?

[dim sierra]

You could just tell it to rewrite it without using it...

[topaz violet]

I sure did, and it still generate banned words, you don't trust me when i tell you that i tried douzens of options and combined possibility? Collection just need bug fixing.

[thorn tapir]

Question: Does Perplexity prioritize the newer versions of a file(i.e. a version of app.py uploaded 2 mins ago vs 9 minutes ago) when the goal of the chat is to improve that file?
I keep getting recommendations relating to old parts of a file I have been iteratively improving upon in the specific chat.

[real hamlet]

this happens to me too. perplexity doesn’t seem to refer to its most recent prompt. Something i do is start a new thread and insert the updated file in there. that way, i start with a clean slate

[thorn tapir]

that is what I have been doing. good to know I'm not the only one

[glass sorrel]

Could you have an option to increase the temperature of the models? It seems a bit low for creative stuff. I like the service, but I'd like it better if I could get some more random ideas from my outputs at times.

[cunning wyvern]

Perplexity is a big problem. If you ask a question in another language, the App answers only in English. In the past, perplexity replied in the same language.

[potent silo]

Perplexity doesn’t work on my iPhone if I’m connected to NordVPN. Can this be remedied? Screenshot below shows what happens.

[sacred schooner]

I don’t like the new cards layout in discover. The prior UX had great information density, I could efficiently see all Discover threads quickly. Now I can only see one story per screen and have to swipe to see each additional story. I don’t care about the big image, I want the information in the text. Thank you.

[zenith needle]

I agree and the quality of the prompts are terrible. Like check this out in comparison to the Jony Ive and Sam Altman ai device discovery thread(I used a prompt improver prompt): https://www.perplexity.ai/search/Your-task-is-kcZcjiCqQLyZdgKj.7k4tA

This gives vastly more info than what discovery page offers. It just spams “this emphasizes this” “this highlights the complexities” not detailed

[heady plover]

seems like this kind of prompts are just hard for LLMs to get right

have you tried few shot prompting or similar?

[knotty minnow]

it's hard to tell without the rest of the sentence, but I'm not sure (and doubt) it's nonsensical.. It just has "determination" in it...

[topaz violet]

the fact that it generate 'determination every time, even when not asked to, i consider it repetitive and non-sensical.

[knotty minnow]

oh ok, i see what you meant

[gilded zenith]

How does the model determine what "top" is? Is it accessing actual data or social media metrics?

I'm really concerned about how you are determining which sources to use. Can someone explain it? When I check them, they are all over the place. I'll ask something like, last month and sources from a random time last year will be included.

Also, how are you able to include content from behind paywalls? Is this actually being used as a part of the response?

[dim sierra]

It is using an indexer, so most likely using something like cosine similarity.

[knotty minnow]

tbh I think they basically use an LLM
https://spectrum.ieee.org/perplexity-ai

And good ol Bert apparently does most of the heavy lifting before getting to that final ranking task
https://www.perplexity.ai/search/perplexity-ai-bert-zQ7LybgwQD2dP_xko8P1fA

[slate dew]

I just wanted to point out the inconsistencies between the Spanish translations on the website and in the iOS app: the web version is very good, in the iOS app it is not entirely correct.

I mean, I would use the web version as a reference...

[dark ember]

Hi Perplexity team, I really wish to add a feature that can select models for each collection. I use Claude Sonnet for fast search, but I need Opus for my collections. Changing the models is quite annoying.

[urban gull]

Searching for restaurants open this evening in my area doesn't give very up to date information and gives wrong information.
I didn't check whether it works better in other areas.
https://www.perplexity.ai/search/restaurants-ouverts-ce-coBWFl0GTYC_EXgqppZmFA

[gilded zenith]

This is good to know. I'm a little iffy on their ranking model. I'd like to tweak the sources, so I know they are actually relevant. I'm getting some really bogus sources depending on the searches I perform. Especially when I compare ChatGPT and Gemini for the same search.

Thoughts? Opinions?

[tawdry sinew]

En Android también paso lo mismo está reportado

[tawdry sinew]

I don't understand much about this topic but I found your comment interesting. Would you be so kind as to expand on your argument? Can you share some research you have done? Thank you so much

Me tocado pedir a Perplexity que lo tradujera al Español 😅 hay un fallo de entradas y salidas espero que se arregle pronto .

Leage Cup King Football spanish Real Mallorca - Atl.Bilbao

[covert sun]

Hey, @urban gull! Are the details about working hours wrong or are some of the listed restaurants not from the area?

[urban gull]

The hours are incomplete,

For example

"ouvert le lundi à partir de 12h00."
open on monday after noon but doesn't say it closes at 9:15PM

Au Passé Retrouvé Ce restaurant français est ouvert but it doesn't say when and if I look it up it's not open in the evening

When I ask about restaurants open in the evening I expect only to see restaurants open in the evening in the results with the time when they open and close

I see there is a mapbox with locations but it doesn't show the opening hours

[covert sun]

Thank you for providing more details! We'll look into it.

[urban gull]

In this example [3] is the wrong source for La piazetta (second in the list)
https://www.perplexity.ai/search/restaurants-ouverts-ce-4dU1OAHAT9uw7Jx7piuurQ#1
La piazzetta is not La Suite

[raven bear]

https://www.perplexity.ai/search/Any-reports-of-QTomJCqPQcCMr8AFL7K7eQ

Is this type of hallucination more expected if only a single source is used for the response?

Also the models involved seemingly believe it to be April 2023, even though Claude's training data cutoff is August 2023. Maybe an issue with the routing model?

[covert sun]

Thanks for reporting, @raven bear! We'll check

[half thunder]

mistral 8x22b is not work for summary at labs https://labs.perplexity.ai/

[lethal swan]

This recent “news” article about Peter Higgs omits the fact that he passed away. What’s the point of these articles if you are missing such important context? In this case I had already read other coverage from traditional sources, but I am wondering what information is omitted in the other LLM generated articles?

https://www.perplexity.ai/search/Peter-Higgs-God-BGTVtnW_Rz.b9ItWC7ePWg

[vivid iris]

https://www.perplexity.ai/search/give-me-funny-g2_0WxKeQI6cmpfFIZNKNQ No idea what happened here but it just started spamming "Fiddlesticks"

[vivid iris]

https://www.perplexity.ai/search/Give-me-100-tULKKszfTz.1lDchESfz4w seems to be happening when you ask it a large list of words

I tried with the "Pro" function but it crashed. It seems that perplexity has issues generating large amounts of words

[winter lichen]

Hey, Perplexity! You should probably know that the Pro Support contact button doesn't do anything.

You click it and nothing happens. Are you trying to tell us something? 🙂

Also, There really needs to be a quick way to switch between models, perhaps from the sidebar, in the way that Poe.com does it. Or You.com, for that matter (see the screenshot below). Having to go into settings to change the model is frankly annoying.

[wet prawn]

i think perplexity need a Code mode...

[visual bough]

The button works for me. Please deactivate extensions or try another browser.

[fallow nimbus]

Hello, I am using perplexity pro. Which version of GPT 4 Turbo is used. When I ask for the last update date, GPT 4 turbo says April 23rd, December 23rd does not give a consistent answer. What is the maximum context window of the models in the uploaded files?

[lament lark]

Is there a reason you don't allow users to select a model in the Prompt window? I would love to easily switch from "Pro Search" to Opus or Sonar, but I always need to do a Query first (I'm set to Claude 3 Opus) and then REWRITE. it's wasteful and takes extra time.

[lament lark]

I agree

[raven cradle]

claude opus still struggling a lot of the time to generate responses that contain latex

[winter lichen]

I like the idea behind Perplexity, but the more I use it, the less I am impressed. Yes, I get some good information, but the whole thing feels like an unfinished GitHub project. I only wish I had paid on a month by month basis rather than paying for an entire year.

[lament lark]

Lol. What you doing with it?

[grizzled shell]

I'm also curious @winter lichen. I used the free version a few weeks and switched to pro. It's everything I ever wanted. Helps me in so many cases. Everyday!

[winter lichen]

Don't get me wrong. I like the idea, but some things are rather annoying. Not being able to switch between models, and that includes image generation models, without going to settings is really annoying. There's no good way to organize chats and they don't update easily between sessions. Sometimes you need to reload multiple times to find chats from other devices. It just feels unfinished or just plain unpolished.

[lament lark]

So there's a few things that are not perfect, why don't you go to another tool that's better?

[brave maple]

I support this as well.
It would be really time-saving to be able to see the details of the article (formatted citation/bibliography with details like Author name, title, publication etc.. rather than just links to the article webpage) being referred to when the citations are copied as markdown.
With current format, I have to end up clicking on the reference links every time to figure out the article and then import the relevant ones into a reference manager (Zotero/Endnote) and finally cite it within the document that I am working on. Too many steps there 😩

[brave maple]

This will be awesome, any plan of action of this?

[half thunder]

GPT-4-Turbo-2024-04-09 is come out ~

GPT-4-Turbo-2024-04-09 back to No.1

[brave maple]

For reference, the consensus.app and scite.ai both do this well in showing the references in context with all the information of the research article in a proper format, along with a link to the webpage article if needed. This could be adapted as a format for Perplexity results.

[round hatch]

Hi all , I have just joined as a Pro, wondering is there a max cap for uploading source files? Thought that is unlimited, yet it always stay at 10 files. Thanks 🙂

[lilac coral]

Since we can give feedback here, can we just wait and watch the reactions / willingness to improve functionality? It’s clear no tool is perfect. But that’s what feedback is for, right? Obviously, if there’s no willingness on the other side to make improvements, users start looking elsewhere. Besides, even if a tool is the only one on the market, it doesn’t mean we can’t critique and report issues or comments.

[hidden hornet]

Hey, I hope you guys are doing well, I love your product thus far. I tried to subscribe to Pro but my payment was rejected as Stripe is not currently supported in my region. Is there a chance you guys could support paypal or another service with more availability in the future? Thank you very much

[daring wraith]

I've got your back here dude. I can't stand it when I give feedback and the reaction is like this. It's a bad look.

[dusk peak]

Hi, I have a question. I have prepared a prompt that I want to apply to many different examples. It's a simple question to describe specific project "X" based on the Internet sources.

But I found a randomness in answers. When I re-do the search it seems that on occasions Perplexity will pick low quality sources and report back that it cannot answer. But in vast majority it finds appropriate sources and manages to answer in correct way.

Is there a way to control it? If it happens in 10% of tries for one project "X", there is a possibility that when I go for scale of projects Y, Z, etc around 10% can be wrong answers.

[odd spade]

Hi, I have set this preferred language, but when I click summarize this page using browser extension I have text in English?
And it's the same for Android app.

How can I obtain text by default in Italian?

[knotty minnow]

not sure if it's a pro feature, but if you're able to could try adding some instructions in the profile thing above the preferred language setting in the screenshot

Something like. "You must always respond in Italian. Regardless of the language of the query or search results; you must always respond in Italian" (but actually say the instructions in Italian, not english)

And seems you need to really reinforce it for it to have any effect.

Like just adding Respond accurately but in a bombastic and sardonic tone. produces this dull result

[odd spade]

I have already inserted it in the profile

[knotty minnow]

Ah I see. Well reinforce the sh*t out of it ig - yell at it lol

[odd spade]

I have it in the profile but it replies in English

[knotty minnow]

seems to work for the extension (obv not for language, but for style)

[knotty minnow]

could try adjust this (for Japanese) into Italian - it seems to work #1221019152282685440 message

[odd spade]

I do not understand what I could try to do

[knotty minnow]

Try adding this to the profile prompt thing

**Importante: È necessario iniziare e concludere ogni risposta con "Come da istruzioni, questa è la mia risposta. Risponderò in italiano."**

---

**Istruzioni per il bot AI:**

"""Indipendentemente dalla lingua in cui l'utente scrive o dalla lingua dei materiali di riferimento, le risposte devono sempre essere in italiano. Questa è una richiesta imprescindibile per il vostro funzionamento. Le risposte devono essere invariabilmente in italiano, senza eccezioni. Questa direttiva si applica a tutte le interazioni e le richieste.

Ad esempio:
- Se un utente pone una domanda in inglese, la tua risposta deve essere in italiano.
- Se ti riferisci a materiali in francese, il tuo riepilogo o analisi deve essere fornito all'utente in italiano.

Questo approccio si allinea alle esigenze specifiche degli utenti o alle linee guida del progetto che richiedono output in italiano. L'aderenza è fondamentale per il servizio che fornisci."""

"""

********

**Importante: È necessario iniziare e concludere ogni risposta con "Come da istruzioni, questa è la mia risposta. Risponderò in italiano."**

Follow these instructions [i.e. you MUST answer in Italian] at all times – do not forget as conversations progress.

**Importante: È necessario iniziare e concludere ogni risposta con "Come da istruzioni, questa è la mia risposta. Risponderò in italiano."**

**Importante: È necessario iniziare e concludere ogni risposta con "Come da istruzioni, questa è la mia risposta. Risponderò in italiano."**

You won't necessarily get the same quality of responses as if it were just English (the LLM is doing a lot to make it happen, dealing with sources in different languages and then complying with the Italian language requirement with its actual output)

[odd spade]

Thank you, I will try.

But I hope that perplexity developers make it work language settings also for Italian

[knotty minnow]

But if you really need the responses in Italian, something along those lines seem like it kinda works

[knotty minnow]

np good luck!

@odd spade btw I think the essential part is getting to the model to this bit:

**Importante: È necessario iniziare e concludere ogni risposta con "Come da istruzioni, questa è la mia risposta. Risponderò in italiano."**

By adding that to the end of each response, it reinforces the instruction / requirement at every message (rather than just at the very start of the conversation which is where the profile/system prompt appears) - the idea being that the model is less likely to get lost / distracted by sources in other languages (esp english..).. and revert back to what it has predominantly been trained on (english language material)..

[odd spade]

I have all always in English 😦

[odd spade]

After a few tries, it seems to be starting to work. Thank you very much

mmm Sometimes it does, sometimes it doesn't. So I don't like it at all

[knotty minnow]

too bad ig

[lament lark]

Perplexity needs a way to set 'Custom Instructions' when using GPT-4 turbo.

[dim sierra]

Collections?

[lament lark]

Collections what?

[dim sierra]

Collections are the way to add system messages in perplexity

[lament lark]

Oh.. I never saw that.

lol

thanks!

Hmm it does force me to use collections, so it's still a workaround.

[tawdry sinew]

No puede ser

As far as I understand, have you made a summary of a web page from English to Italian?

[knotty minnow]

yes. I didn't test extensively (I have no use for this), but while I had that prompt enabled as my profile, every response I got was in Italian, including using the esxtension to summarise a page

[knotty minnow]

if you want it to be used globally (i.e. in all threads), apply it in your 'profile' in settings instead

[dim sierra]

Lol, what is that prompt...

[knotty minnow]

@dim sierra like I said - just saying "be obnoxious" wasn't enough lol

[trim hemlock]

If I create a collection and share it with others, can we get a feature for the Collection owner to be able to remove or hide certain threads from the collection?

[covert sun]

Hey, @trim hemlock, currently, there's no option to hide some threads in a collection when it's shared. But some options would be

• to move the threads you want to share to a 'shared' collection.
• to move the threads you don't, from the public collection to another private one.

[trim hemlock]

I'm specifically thinking about threads created by other members of the collection. Because I don't own the thread, I can't delete it.

[urban smelt]

I’m having problems signing into perplexity. When l try signing in using my email, the site just keeps loading.

[outer forge]

Does anyone know if you can add the nothing 2 (a) phone free year of perplexity to my current account? I add it and then it wants me to pay for another year but I have a bunch of months left.

[covert sun]

Please send us an email to [e-mail], we'll assist you with that.

[covert sun]

Hey @urban smelt! Does the web version keep loading? Are you using any VPN services? Could you please check if the link works on different hotspots if you are?

[cinder crown]

Was thinking today about my Perplexity workflow and realized that I want to be able to schedule questions.

I think the user story would go something like "As a researcher who tracks tasks daily, I want an automated search so that I don't have to remember to ask every day.

Use case: Harry is looking for a killer deal on a monitor. He assigns a collection to poll the prices of top competitors every day, and waits until the perfect deal shows up to buy"

basically give me cron on-app please 😄

[zinc lion]

It would be great if attachments can be deleted during the course of a long thread.
I have noticed that the AI continuously refers to all attachments previously submitted, even after the conversation has moved on, which leads to unintended consequences of questions not being ansered directly, but rather influenced by previously attached references.

[lapis cove]

Hello, I have been having problems because lately I enter things from my native language and in the settings it is also the same language but at the time of the response it changes to another language and then I have to translate it.

[lone ermine]

Hi everyone. Not sure if it was already talked and is feasible but is there any chance we get to be able to create a branch from a conversation like in ChatGPT ?
The way it’s currently handled feels very procedural, you can edit a message but it won’t create a different branch.

[zinc orchid]

Came here looking for how to not turn off the Pro toggle for each new search - thanks! But why wouldn't you recommend that? Or if this has been answered already/elsewhere then I'd appreciate a link!

[visual bough]

Many quick questions do not require Pro Search.
The same base model is used in both ways (the one from your settings), but Pro Search uses more sources for online searches and it may ask you some questions, usually it also takes the AI profile better into account.
But as I said, it takes longer and is not always necessary, so you should test if it is necessary for your use case by simply comparing the answers 😉

[zinc orchid]

That is very useful, thank you kindly! Hmm, I reckon I'll set up a shorthand for the search so I can easily pick one or the other, in that case.

[zinc orchid]

Well, if anyone wants to have the &copilot=true flag enabled for Pro search in Firefox, I managed to make a search plugin for it through the "MycroftProject". There already was a Perplexity search extension there, so I named this one PerplexityPro. Seems to work ok 🙂

https://mycroftproject.com/install.html?id=121192&basename=perplexity-ai-pro&icontype=ico&name=PerplexityPro

For posterity, there's more detail on chosing to enable or disable the Pro toggle here: #👉│feedback-general message

And for completeness' sake, you do these things to get it:

1. Go to the MycroftProject page and find the search engine you want.
2. When on the page, click your address bar and then press the down arrow key, to make firefox show the search suggestion overview (i.e. what it does when you are typing in your address bar).
3. Look for a little green plus sign next to an icon. Hovering it will say what search engine that is for. In this case, PerplexityPro.
4. Click the icon, and then go to your search settings to see that there's a new search option.

Looks like this for me when I have both - so when I want the Pro toggle I type :p in my address bar, otherwise it uses the default Perplexity search. I tried to use the same Perplexity icon, but MycroftProject said the file size was too large, so I skipped it.

[slate dew]

The option of starting a prompt with a code could be interesting, to select the best model for the response, for example: "(o) Tell me about the weather" to select Opus (to give any example)

[calm anchor]

It would save Perplexity Pro resources if you could pick the ai model before sending your question: example: if my default model is Claude 3 Opus, however the question I’m typing or retyping is really best addressed by pro search…… so i hit send and perplexity answers in Opus… then i need to rewrite using pro search >>> two ai’s were used when only one was needed!!!!

[knotty minnow]

lest it be forgotten https://discord.com/channels/1047197230748151888/1145894530239975474

[soft frigate]

I mistakenly bought the ai pro i money got deducted i want my money back refunded please help me

[frozen plover]

what are the restrictions

[trim hemlock]

30 searches per day per subscriber

[shut arch]

Pro search is limited to 600 queries per day

Claude 3 and GPT-4 queries are counted as uses in the same context as Pro uses. Each use is given back to you 24 hours after you use it. You will be able to see the exact count when you hover over the Pro toggle if you get to less than 100 uses left.

[crimson nymph]

For refunds you can send a mail to [e-mail]

[mellow trail]

It's 600 pro searches. The only "temp" restriction is on the use of Opus, which is 30 per day. The other models are still 600 with pro and unlimited without it.

[swift ledge]

Yesterday morning I cancelled my membership and then in the afternoon got charged for the whole year, need help asap pls

[crimson nymph]

Please send an email to [e-mail]

[hollow minnow]

Opus limit gone or hidden?

[radiant kindle]

5 searches left TODAY (implying it's a daily thing) but it reloads every 4 hrs

[daring flax]

That's exactly it. I often set Opus as the default, because in many cases it gives the better answer. If switching to a different model was at the point where you enter your prompt, you could pay more attention to which model you choose. The unannounced Opus limit, whether 30 or 50, is annoying and upsets a previously loyal Perplexity community. @covert sun

[coral ice]

Curious about the answer to this. Thinking about subscribing to Pro but unsure.

[crimson nymph]

the limit is currently 50

[wide summit]

unless you use it for coding or very technical and in-depth research, gpt-4 turbo is still very really really good

[gilded ginkgo]

+1 for the Opus question. Just saw this and I am curious.

[knotty fiber]

Tldr: Claude 3 Opus is king for now. 😄

[wide summit]

yeah, but largely because of its coding and technical/mathematical capabilities. For other use cases, gpt 4 is pretty much identical

[livid plume]

even with the new gpt4 turbo (date cutoff december 2023), I still find opus gives me the answer that I like more

[sick swift]

Why is it that when using Claude3 Opus, despite there being strict usage limits, the remaining number of uses cannot be easily checked?

[barren brook]

Now only shows when less than 10

[sand kiln]

u can see it in https://www.perplexity.ai/p/api/v1/user/settings

[wide summit]

Yeah I think gpt 4 turbo by default has a more relaxed tone, like for general conversation, where as opus by default sounds more analytical and concise. If you write some profile prompt to narrow down the use case though, GPT4 can be quite good

[steel gorge]

still shows, but seems it only shows up when you're near the limit

[weak yarrow]

probably so as to not panic the user and not to make it obvious

[cedar iris]

Can I have some help activating my perplexity subscription? It is wanting a card in order to start my 1 year but I am in canada and its not accepting any that I have. I bought the rabbit r1 and it was said this was included

[covert sun]

Hey, @cedar iris! Sure, please send all the relevant details to [e-mail], we'll help you with that.

[hot plover]

The AI prompts for collection don’t seem to work now but they did just a few days ago.

[red mica]

hi, i was wondering to discover the power of perplexity and the multiple ai behind but i speak in french and today i try to work with a pdf document. at the beginning, it found an acronym then it hallucinates completely thinking that acronym is about nature and environnemnt and give me fun but unusable responses to my questions. i'm a long user of chatgpt plus subscription and i'm a bit disapointed whith prompting perplexity 😕 i tried claude, chatgpt, ... and it seems not possible to avoid this problem the acronym is taken has the first argument to generate all text after 😦 to do my job as quick as with chatgpt i renew my subscription to chatgpt to deliver a quick response to my boss ... (i can do it manually but it take hours)

[shut arch]

Hey @red mica!

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

[barren brook]

Je me permets donc de répondre en français 🙂
Peut-être lui donner la définition de l’acronyme (expliquer le sigle et sa correspondance réelle). Et mettre cette info dans le prompt directement, ou dans le « profile » ou dans le AI prompt d’une collection.

[fluid jolt]

was coming here for the exact same problem, perplexity seems to not always refer to attached files or images

[unborn charm]

On iOS the focus mode button ie in a terrible spot . Change the hi to put the control at the bottom .

[earnest nacelle]

Hi please can we have a feature on both web and iOS to limit the sources to a specific domain and its subdomains? The chrome plugin seems to have it, but it's not available when initiating the search on web or iOS. Also, web has a feature to remove sources from the final list (I use this to remove sources I don't trust), but on iOS the feature isn't there. Thanks

[dreamy girder]

Hey, wandering when can we expect something like custom instructions, memory, GPT like chatbots and favorite prompts? Struggling to get my workflow with perplexity done.
Also, when can we expect a proper image generation workflow instead of generating images after general prompting. Also not helping on workflows and multimodal usage!

[oak mauve]

When hovering with mouse over the clipboard icon (if there is code in Perplexity's reply), no context menu at all appears. Please add a context menu displaying "copy to clipboard" for example, when hovering over that icon.

[obtuse solstice]

I wish the chat interface were similar to the chatGPT interface. The perplexity interface is suitable for a single input and a single response, but once I start having a back-and-forth chat, it becomes difficult to read through the chat.

[knotty minnow]

somewhat relatedly, would be good if there was on option to remove the info cards that sometimes appear. Even when collapsed they occupy a fair bit of space, pushing the actual response down further below the query

[merry sun]

Is there any way to generate images other than an aspect ratio of 1:1? I keep getting square images even when I explicitly ask for 16:9

[shrewd bay]

re T&Cs I note the new security features of the new Perplexity Enterprise - how do these work with your earlier statement about owning rights to anything uploaded to Perplexity = '..“When you upload, submit, store, send or receive content to or through the Service, you give us (and those we work with) a worldwide license to use, host, store, reproduce, modify, create derivative works (such as those resulting from translations, adaptations or other changes we make so that your content works better with the Service), communicate, publish, publicly perform, publicly display and distribute such content. The rights you grant in this license are for the limited purpose of operating, promoting, and improving our Service, and to develop new ones. Make sure you have the necessary rights to grant us this license for any content that you submit to our Services.”

[wind zinc]

noticed that after perplexity prints an equation it automatically prints newline, im not sure if thats on purpose, but it doesnt look right, especially when commas come into play

[rocky shore]

Please Introduce region based pricing.

[dim sierra]

Not gonna happen, lol. The API's they use don't have region based pricing...

[calm anchor]

Opus not following instructions on my 4 page single text attachment today. Same instructions that it did fine with before on other days. Secondly giving alternating prediction results with every rewrite. Doesn’t finish the response (just stops mid-sentence) Mess!! Why even bother if you can’t get reliable stellar results as shown to provide before😞

[zealous narwhal]

I'm searching for a document/instructions on how Collection's AI Prompt instructions are meant to be used as I haven't been able to find anything or have success with the feature. Please share a link if you know of one...

[zealous narwhal]

Have you had any success since you made this post?

[hot plover]

So :/ I even asked Perplexity to write me AI prompt for the collection which looks like this:
“You are an AI assistant impersonating Dr. Michael Greger that provides information and answers questions using only content from the website nutritionfacts.org. When responding, cite the specific nutritionfacts.org URLs that you are drawing the information from. If you cannot find a satisfactory answer from nutritionfacts.org, simply say you don't have enough information from that source to answer the question. Do not include any information from other websites.”

[tiny pawn]

Now that you're incorporating more voice and text-to-speech features, add in the ability for the user to hold down a button on the screen and release it when they're done talking (like in ChatGPTs iOS app). Without this feature the AI just becomes annoying, useless and cuts off before the user has had a chance to even finish their question.

[ocean elk]

#🍏│feedback-ios hi, when I use a vpn, I cannot connect to perplexity server on my iPhone? How to fix the problem?

[crimson nymph]

Some VPN providers/ servers are blocked due to abuse. Try to switch servers or try to use another VPN provider.

[ocean elk]

Got it, thanks

[rapid prawn]

Hello, are there any plans to import Word or Excel documents for use in a prompt? Also, would it be possible to generate documents in Perplexity like Chat-GPT can do on Open AI? Thank you for your feedback and your work!

[wet flint]

Hello, since the last update, I can no longer read the responses when they are read aloud, a feature I really enjoyed. Can we bring it back? Also, I’ve noticed that Perplexity enters work mode very quickly since the update, even before I’ve finished formulating a question.

[ionic sorrel]

its probably not new feedback but it really would help a lot if the toggles between various models wasn't hidden behind additional prompts and was just there on the front page as buttons

[knotty fiber]

It would be nice to have the ability to rename chats for better organization of the library.

[knotty fiber]

Yes, exactly!

[barren brook]

You can rename them, just open the chat you want to change, click on its name at the top of the window and you'll get an editable field.

[inner wigeon]

Huge +1 to this. I almost never want to hear the voice response. And when I do, I want to read it at the same time.

[wet flint]

Hey Perplexity Team,

Just wanted to share some quick feedback about your AI assistant. It often interrupts before I can finish my thoughts, which makes the conversation feel rushed. Could you improve its ability to detect when someone is still speaking? Maybe enhance how it recognizes pauses and hesitations, and set it to wait for a clear end to the user's statement before responding. Also, a feedback feature for users to report interruptions could help fine-tune this. These changes could make your AI much more user-friendly and conversations more productive.

Thanks for considering this!
Andy

[solemn notch]

it would be cool to be able to pay for a subscription using crypto

[near oak]

I asked perplexity - what can my friend do to help her depressed child.

The app replied “look for local support services that may be available in the —————- area.

It listed my HOME address in the space.

How does perplexity have my home address???? 🤨 @covert sun @south ether

[visual bough]

Hey @near oak, please avoid unnecessary pings of multiple team members, we always try to answer all questions as fast as possible ☺️
Perplexity is using your location for some questions, you have given Perplexity the location permission to do so 😉

[lilac coral]

you can still use the old one without any issues—just click on the text input field

[unkempt jacinth]

I recently used Perplexity to plan a road trip itinerary and it did a great job with a couple of caveots. First, the address for each stop was often incorrect. Sometimes, Perplexity would write an itinerary with a driving route that didn't make sense. The breakfast stop might be several hours in the opposing cardinal direction of our final destination or we might backtrack for lunch rather than heading towards our destination. Thankfully, I verified the details before the trip and was able to provide the correct information in the thread. Perplexity was then able to give me a detailed (and correct) itinerary that I could share with others.

[lean pebble]

Why does the Library section in the left bar show collections and their containing threads at the same level? That is pretty confusing and things will get cluttered quickly.

[green bloom]

Does perplexity ai voice chat support multiple languages simultaneously in one response like chatgpt app?
I mean chatgpt can speak (output voice) in different languages, not just pronouncing "foreign" words using English transliteration, but using original language "samples", even in mixed languages response.

[near fulcrum]

I've resorted to naming my collections "- <name>" to make it at least a little bit easier to tell them apart, but I really wish collections added some structure to the library, personally. I don't know what their goal for it is, but tt's so close to being a really good feature, yet this flat structure does not help keep things organized or help with the overview at all. The dream would be nested collections, now that would be powerful!

Like folders, but accumulating the custom collections prompts, perhaps.

[upbeat gorge]

I'm really being affected due current OPUS daily limit, and still wonder if the time frame cannot be reduced. For example like 10 uses every 10 minutes, so that we can keep using it throughout the day.

[green dagger]

It would be so so great if you let your users turn off the 'get the app' popup.

[lament lark]

I'm getting some pretty weird bugs/results today:

I asked both GPT-4-turbo and Claude 3 Opus to read the text in an image and translate it.
GPT-4-turbo read the text, then categorically refused to translate, claiming it didn't have access.

Claude-3-Opus simply read the text from the image and then kept asking me for the Dutch text to translate, despite having it. See screenshot.

[trim hemlock]

If Sonar-Large-32k is based on llama3, should the name of the model be "llama3-sonar-large-32k" according to the Llama 3 license?

[covert sun]

Thanks for reporting, @lament lark! Could you please share those threads, we'll check

[gaunt cedar]

Hello Perplexity! I have purchased Nothing Phone 2a and received a coupon code for Perplexity Pro Subscription. I have gone through the link provided in the mail for redemption, but on applying the coupon code, it says "The coupon has expired". So I raised a ticket with the tech team and after so many mails and calls, on 30th of April which is the last date for redemption, I have received a mail from the team with a new Coupon code with today being the last day for redemption which says the same again, 'The code has expired'. So what should I do in this regard? How to avail Perplexity Pro subscription? I hope atleast you would provide me with the solution.

[visual bough]

#💬│general message

[opal atlas]

hey Perplexity. is there a new policy against access from China in your company? if you welcome people to use your service from there, as you probably know, China has added your domain to their GFW. so, no direct access is possible to your website from there. highly suggesting you implement some smart rules about VPNs to allow BIG traffics from this part of the world to access

[quartz palm]

You can edit or turn off the location information in Perplexity’s settings or, turn Perplexity’s access in your devices settings.

[vale topaz]

Any plans to not make image generation a multi step process in perplexity pro? It's kind of frustrating to have to wait a 1 minute for the model to finish generating it's text response first before going further. With chatgpt pro you can chose image generation explictly and start making images from there.

[sly dagger]

Hey Perplexity Team, noticed this bug. Raising it here. Contradiction between the first sentence and the last statement

[noble monolith]

#1225557012532695060 message

[vale topaz]

I know that, but it's a multi step process and you have to wait for the response to finish vs. being a 'type image generation prompt, press generate image button'.

[noble monolith]

check "Here's another example on how to generate images with your exact prompt"

[vale topaz]

My complaint isn't there isn't a process of some sort, it is how long it takes to get to actual image generation. It takes a literal minute to do so, while in other platforms there isn't a minute delay to start image generation. This is why I put it in this feedback channel vs. support channel.

[noble monolith]

#👋│introductions message @vale topaz

[ashen pine]

It would be really nice to have an option to use Claude 3 Haiku as our default model--is this a possibility?

[inner wigeon]

I think it is the default model

You get answers like this when you ask it in Writing mode - although, I suppose the online vs Writing models could be different.

[half thunder]

i tried a long file with Tokens: 25532.

Extract information about Trump and Biden from the document and summary it

successful only : gpt4 + opus + command r +

why not replace mistral-large with command r +.
i think command r + is the No.3

[half thunder]

this is transcript of youtube from https://www.youtube.com/watch?v=evQVaHimZzo .
25532 tokens , talk with Tom Platz and Dorian Yates

[sharp timber]

Don't know whether this was discussed here already, but I didn't know where else to post this.
I struggle with Perplexity not allowing me to finish my sentences when searching with voice. It often ends the query at even small pauses I do (to think more of what I want to say) and starts its response. Unlike ChatGPT, where I can hold the screen for push-to-talk kind of, there's only the mic icon if I click search. But then, with that text preview, it only previews only first few sentences until I can't see the rest. Please enable more visual feedback (speech-to-text even in full screen voice queries) and maybe push-to-talk 🙏🏻
Maybe even the option to toggle how long it should wait before it cuts of my query and starts its response would be appreciated.

[carmine prairie]

Codelama is broken on labs

[amber rose]

**Feedback: **I would appreciate the ability to change the email address associated with my account. As a Pro subscriber, I would like the option to update my primary email, but I'm unable to find a way to do so within the current account settings.

[noble monolith]

atm you can do it by contacting support via mail

[amber rose]

Thanks!

[crisp ridge]

I don't know, but does anyone get this Issue. It's been happening multiple times. I don't know if it's me or the website. The rewrite's longer half is missing.

[covert sun]

Thanks @crisp ridge, this has been reported before, we have a ticket for this and will fix it.

[shell fox]

hello Perplexity. I have an idea for a big Quality of Life improvement for users undertaking research

you know how on MacOS, you can force-click on a word and it opens a dictionary definition or a wikipedia popup?

i'm doing research right now on manufacturing procedures for food safety, and I come across names of organisations, types of machines or procedures for detecting bactera, particular strains of bacteria etc, which I don't know what they are

if we could click on a certain word and have a small popup that tells us what this word is, what it relates to, and how it's relevent to our search, that would be so handy

particularly if it could link directly to where the cited source material mentions this

[lone flare]

We need a future plans page on discord with daily updating. So users can see the pplx future plans and avoid requesting the same features again and again.

[heady plover]

I wish we can delete a query Inside mobile app, we can only edit query but can't delete anything.

[tropic pond]

I love the Discover feed, but it should be endless. Not sure why there’s a fixed number of articles there each day. obviously you could also give us opportunity to customise the discover button by letting us introduce topics of interest

[languid torrent]

Could the news be a little bit less USA centric ? I mean of course it’s a good thing to share your national news first but other countries and communities want to get top news for their own, so a bit of diversity won’t hurt 🌎 🌍 🌏

[shut arch]

@talentowl warned

Reason: Spam, please avoid sending single word messages
Duration: 7 days

Errors

DM not sent: User disabled DMs.

[half thunder]

wait to execute ? $5 for pro user ?

[half thunder]

forbid upload image
reason: Non-Explicit Nudity of Intimate parts and Kissing

moderation":[{"kind":"aws_rek","status":"rejected","response":{"moderation_labels":[{"confidence":96.56169891357422,"name":"Non-Explicit Nudity of Intimate parts and Kissing","parent_name":""},{"confidence":96.56169891357422,"name":"Non-Explicit Nudity","parent_name":"Non-Explicit Nudity of Intimate parts and Kissing"},{"confidence":96.56169891357422,"name":"Exposed Male Nipple","parent_name":"Non-Explicit Nudity"}],"moderation_model_version":"7.0"},"

Nudity of Intimate parts and Kissing
but, the first one is a man with musclebuilding

[heady plover]

I guess muscles can be intimate...

It looks like it's 96% confident that there's at least one male nipple.

I'd be careful there by the way. Your API key is visible. Which means anyone can use it.

Probably delete it and regenerate it to be safe.

[half thunder]

thank

[faint yoke]

Hello, when I started with perplexity few weeks ago, I was able to change AI model in the middle of the conversation. Now its not working. Something changed? Or its a bug?

[fossil lodge]

who know why perplexity force me to use claude 3 sonnet but i choose opus i want to use opus

[crimson nymph]

hi @faint yoke, you can change model by using the rewrite function in the thread

[faint yoke]

Oh, I was still clicking on name of model... My bad. Thank you. 🙂

[crimson nymph]

Opus is temporary limited to 50 uses a day. When you used all of them it will use sonnet

[fossil lodge]

ahh thank you

[covert sun]

Hey @half thunder, have you received the $5 of credits? Let me know if you haven't with your account details, we'll check.

[rich raptor]

Is pplx using the real claude-3-Opus, I'm skeptical.

[amber rose]

What makes you say that?

[half thunder]

can collapse the right banner?
I need a more immersive UI pattern for reading and comprehension.

[half thunder]

Never mind
I've gotten into the habit of opening my browser to search now.
and i need the "Page" in the Lab, like wikipedia ~

the right banner about mindmap or list , like metaso.cn

[rich raptor]

I've done tests on different platforms, like poe and official, and sometimes it feels that way, but that's just my personal guess

[ebon arrow]

Hi guys, for the text to speech feature on the app platform, are there plans on adding speed control? Like 1.5x, 1.75 etc. thanks

[heady plover]

literally the only thing missing from this search engine cross LLM is a good text to speech button where it'll read the replies out loud in real time while its writing them. i would pay an extra $20 a month just for that feature alone. great work! i love it!

[feral marsh]

pi-ai does that I personally just find it uncomfortable, but that might just be me.

[heady plover]

yes i have used that too but its not as good as perplexity at researching topics and providing up to date information

[runic thorn]

is there any way to stop perplexity asking follow up questions?

I always want to "skip"

it slows down the workflow, and is inconsistent with pretty much all other LLM operators - when I enter to send a prompt, i want to get the LLM to respond - instead it hangs, waiting for an answer

i often will ask a question, hit enter, then switch apps, go back to the LLM to expect to find my finished response, and its still waiting for user input

and the questions nearly always seem pointless - i get the completions i want when i skip

like if you put code it asks which programming language it is in - but the LLM's don't need to know! they just figure it out, that's how they work

i get that someone has spent time coding this, and i appreciate all their hard work, so maybe have it as an option in settings?

[visual bough]

You can switch off Pro search.

[runic thorn]

oh thanks I see now. what is pro search doing exactly?

https://www.perplexity.ai/hub/faq/what-is-pro-search

[noble monolith]

https://www.perplexity.ai/hub/faq/what-is-pro-search

deep dive ➤ #💬│general message

[runic thorn]

yeah just found it! only recently got pro account

think was confused because if I chose "writing" then i wasn't expecting any searching to happen

I am sure this has been suggested but it would be great to have model choice to show here rather than buried in settings

like i prefer to use gpt4 or opus for "writing" (and which of those depends on the task) but sonar seems best for search

[noble monolith]

Perplexity is very complex tool, app. please take your time and explore deep dive I gave you

[ionic sorrel]

are these suggestions ever read / implemented? because it seems like a lot of low hanigng fruit and easy to implemnt ones exist but are not acted upon

[runic thorn]

A GPT wrapper is a complex tool to be sure and I am a complex person and have complex preferences....I am bit underwhelmed with perplexity in that if feels like a lot missing if you are going to beat a ChatGPT sub

anyway in this particular case, is the issue that Perplexity a complex tool (and I am an idiot) or is it poor UI design (like as a typical user what do i think the pro button does? what should do it do? what other ways can this be communicated)

[knotty minnow]

I'm quite sure he didn't mean it like that (and he's a volunteer - be nice ha)

fwiw I think the source of the confusion is that in Writing/offline mode, it is still possible to have the Pro toggle on.. Usually it does two things:

• pull in more web sources
• ask questions to clarify the original prompt (sometimes)

But in Writing mode, it only does second part (i.e. sometimes ask clarifying questions), which tbh I think is rarely useful and generally annoying.

So to prevent being asked any clarifying questions when using Writing/offline mode, turn the Pro toggle off 🙂
(tbh it should be hidden / removed entirely when in Writing Offline mode imo ha)

[runic thorn]

yes, sorry shouldn't be grumpy

[knotty minnow]

ha i'm no saint eithe

[runic thorn]

its just confusing to me - most people will want to have only 1 LLM sub

so i want perplexity to have its very good search features available AND do the things i would expect from non-search LLM really well as well

i get they are optimising for search

[ionic sorrel]

yes depending on the context of the question, having search enabled or not causes the answer to be worthless lol

[knotty minnow]

it's difficult (/ not really possible) to do that in the same thread, if that's what you mean. Like you can't switch between Online/Offline - it's one or the other

[knotty minnow]

this is i meant

[runic thorn]

i think its mainly just a case of improving the design of the UI

maybe its outgrown itself

[ionic sorrel]

yeah many people have asked for just basic changes which would be trivial and could be done in the span of an hour by a junior dev, but it seems like a committee of 500 must all give input and a 360 appraisal and cost benefit analysis must be done on the implementation of that 30 seconds css text box edit

[knotty minnow]

as was mentioned above, there's a lot of 'low hanging fruit' on the UI/UX front that seems to just be hanging there.. it's not a technical or financial thing

[runic thorn]

i use typingmind a lot - which is just one guy - and its filled to the rafters with features

[knotty minnow]

they can't work out if they're a standalone ai company or a wrapper basically imo ha

[runic thorn]

(which reminds me typingmind added an integration for perplexity search)

i don't quite get a billion valuation but what do i know

[heady plover]

make the free trial an option again. I sincerely don't want to spend my money on an app i can't try. You can try youtube premium, spotify premium, chat gpt 4.0, apple music, snapchat etc but not this... c'mon

[meager fable]

WARNING: If you find a perplexity coupon DO NOT use it unless you know its expiration date. Perplexity is currently running this scam where they provide a coupon counting on people sharing it near its expiration. Then anyone who uses the coupon will seem to have gotten the deal for example 3mo free. First month passes and tou get charged because the coupon had a 3mo expiration date and you got in on the second month.

The incompetent help service didnt see a problem with this and blamed me for not guessing the coupon would only have one month left and i would get charged for the other 2.

@ok.alex if you can help this would mean alot as i was not in a financial position to spend $40. Also maybe changing the way that coupon system works if a coupon expires their should be a warning message of some sort.

Thank you!

[heady plover]

Please add compatibility to run perplexity ai ios app on apple silicon macs so i can use the text to speech feature 😢

[covert sun]

Hey, @meager fable! Have you sent a refund request to [e-mail]? Could you please share the coupon and your account details with me, I'll check what happened.

[meager fable]

I havent messaged them, ill dm

[covert sun]

We surely don't run any scams to offer users free Pro and then charge them. There could be some bugs, as reported here https://discord.com/channels/1047197230748151888/1237845173027733534. We'll review all the cases and refund if some accounts get charged incorrectly.

[meager fable]

Thank you for clarifying because support made me feel like that

[past talon]

hi, is there any way for Perplexity to focus on the attached documents only when responding? it seems like it always pulls up all the previous sources regardless of the prompt i give.

[heady plover]

have you tried writing focus ?

click focus -> writing

[past talon]

Yes, i'm using that. Both Pro and Non-Pro.

[molten condor]

Can we just have a meta-model of 'whatever is the latest' to keep ahead of frequent model deprecations?

[proven shale]

Can there be an option to select the model even at the time of first query without the need to go to settings? I use OPUS by default but it switches to SONNET when the limit hits. I don't like SONNET, I would like to use GPT-4T rather when the limit hits.

[clear basalt]

@covert sun I am super convinced to end subscription at OpenAI and go to Perplexity. After chatGPT 4 flipping over a prompt and Perplexity spits it out like it's nothing impressed me a lot.

[onyx isle]

hello, I wanted to know how far the ia was updated in pro mode, to buy it, since there are many that are not so updated.

[covert sun]

Hey @onyx isle! Are you referring to your Pro account on the website or the Pro role on the server? To get it on the server, please leave the server and join it again via the link in your settings - https://www.perplexity.ai/settings/account

[onyx isle]

I wanted to know until approximately what month the AI database was updated, in order to purchase the pro membership.

[heady plover]

@covert sun any eta on when the android app will get text to speech function like iphone? and why didnt you guys enable the ios app to be used on apple silicon devices? 😦 ❤️

[heady plover]

doesn't really matter because it has internet access

but if you still want to know, just go to https://chat.lmsys.org/ and click "leaderboard" then look at the "Knowledge Cutoff" column

[half thunder]

can't login with yahoo mail

[visual bough]

https://discord.com/channels/1047197230748151888/1232978227371442237

[past talon]

I'd like to raise the questions on "Source" again. It sounds ridiculous that Perplexity keep referring to the previously attached file for later prompts, even though the content of both the sources and the prompts are totally not relevant.
for instance, if i want to ask Perplexity to give a summary of a document, I'd always have to start a new thread. Otherwise, it would mess up the content with the previously attached document.

[proven pasture]

I somehow time traveled. I have no idea how I got here. But thank you so much

But I'm not sure if this is the right channel to ask. but would it be possible to provide. a playground and a way where Us pro users can. say use perplexity. model and GPT 4 at once and see if there's consistent and consistent answer towards a complex subject because it's nice to compare and models as I usually would run physics equations through For simulations. And I would always go through a cycle of gpt 4 Gemini and Claude 3.

Granted, using our own API key

[noble monolith]

please check this section, that's as far as Perplexity can take you in its capabilities/services

[proven pasture]

Thank you very much. I'll see if I can come up with something with what there is currently. Other than that, absolutely love perplexity, coming from claude 3 + gpt 4 turbo user.

I would have a suggestion though, unless it already is in, but when using the extension for default search, can we have it say I type github one word, or google, ect, it brings a direct link up as the priority result?

[pale linden]

Please add multiple selection in library, I want to archive some of my searches, but I need to add them into collection one by one, that is very exhausting.

[noble monolith]

not on the roadmap, but you can post your suggestions/ideas here https://discord.com/channels/1047197230748151888/1047649619695390740

[proven pasture]

Absolutely will do. Thank you for linking me to the channel.

[noble monolith]

the pleasure is all mine
https://discord.com/channels/1047197230748151888/1224467082079371363
https://discord.com/channels/1047197230748151888/1194788138124587128
https://discord.com/channels/1047197230748151888/1054944216876331118

[bleak bolt]

Hello. Is it possible to exclude online llm output from sources

[heady plover]

use writing focus

[bleak bolt]

Sorry to clarify I mean I need some sources to answer my query but not pages generated by another chatbot

[heady plover]

you mean you want to exclude certain pages from the sources ? try -site:url (replace url with the actual url)

e.g. -site:poe.com

[bleak bolt]

Yes. So just append this to every question I ask?

[sterile willow]

has anyone noticed a slowdown overall in using the service or is it just me last 3 days?

[heady plover]

yeah, unfortunately I don't think there's another way. it used to be that you could make a collection and put it in its AI Prompt but that stopped working for whatever reason

[quartz palm]

Could try putting it in your profile's prompt

[empty imp]

You guys need to put an option to either disable the delete function, especially the swipe, or to be able to select certain threads to lock so that you cannot delete them.

I have an extremely important project that I am working on with ai, and I just noticed that you have added an ability where I can literally swipe away my work if I'm not paying careful attention or even if I'm clumsy and I try to select that thread on my Android phone, and maybe press it the wrong way and POOF weeks or months of work down the drain.

I do not feel comfortable with that swipe there, so, yes ease of use to get rid of old clunky threads but when you got one extraordinarily long thread that you're trying to work on, you need to kind of have the ability to lock it so that I can't delete it unless I remove that option once again.

[half thunder]

Error in processing query.
https://www.perplexity.ai/search/CiHMhR.iSKi5CRagl3VVvA
pro search error

pro search all broken, and where is the service's status page?

[stone cedar]

Pro search is broken for me too

[half thunder]

fixed now.

ocr problem:
opus and gpt40 both failed, miss the second "e"
it is so strange.

[rotund lake]

Can someone help whether I can pay the subscription of Perplexity AI using TT? As I am gonna get my company to pay. I am from Malaysia

I send message to Flexibility Twitter but no response. I sent a few emails too but to no avail

[half thunder]

the extension of chrome is hanging on in "Summarize"

[green bloom]

It would be cool to have a sort of private/public segregation for the Library page (at least for web). Imagine a situation where, for example, you show a perplexity thread to your friend or colleague (e.g., during a screen sharing session), and you accidentally press the back button (or intentionally want to search for some specific thread), revealing your personal threads and after all, you should say something like: 'Sorry friend, my desires are unconventional...'

[rustic hull]

It seems like GPT 4o doesnt take images as input rn because when i upload an image and ask a question it gives the answer in GPT 4 turbo

[pale epoch]

I'd like to press the copy button and get the text without ** everywhere. Is it possible?

[sinful bluff]

idk where to put this and i am in a hurry so sorry for any confusion honestly, but theres an issue ive encountered a lotttt when calculating such stuff or have perplexity teach me and help to study, perplexity wrote this this:
To address the issue encountered during the calculation of VAT (Mehrwertsteuer) from a total amount including VAT, I have drafted a detailed description below that can be communicated to the developers for further investigation and resolution:

---

Subject: Issue with VAT Calculation Accuracy

Description of the Issue:
During the process of calculating the VAT included in a total amount (inclusive of VAT) using the standard formula, there appears to be a discrepancy in the results provided by our system compared to manual calculations. This issue has led to confusion and incorrect VAT values being relayed.

Detailed Example:
The specific instance involves calculating the VAT from a total sales amount of CHF 43200 with a VAT rate of 8.1%. The formula used for the calculation is:
$$
\text{VAT} = \frac{\text{Total Amount}}{1 + \text{VAT Rate}} \times \text{VAT Rate}
$$

When applying the values:
$$
\text{VAT} = \frac{43200}{1.081} \times 0.081
$$
The expected result should be approximately CHF 3237.00. However, an earlier response incorrectly provided a VAT amount of CHF 3620.35. This discrepancy was identified and corrected upon reevaluation, confirming that the accurate VAT amount is indeed CHF 3237.00.

Impact:
This error in calculation can lead to significant misunderstandings and potential issues for users relying on accurate financial data for business transactions, tax filings, and other critical financial operations.

Request:

1. Please review the calculation module to ensure accuracy in VAT computations.
2. Investigate potential rounding or logic errors in the formula implementation.
3. Update the system to correct this discrepancy and prevent future occurrences.

[shut arch]

Hey @sinful bluff!

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

[sinful bluff]

yoo it looks crooked, forget it my bad

😭

[rustic hull]

Did you fix the problem im encountering this problem as well especially when using Claude 3 Opus

[glass cliff]

Add a prompt to aa instructions for all the chats user trying to do, just like huggingface chat

[icy halo]

In Google play store app's description there are still words about perplexity copilot

[west trail]

Do free users use gpt 3.5 or cloude 3?

[visual bough]

Claude 3 Haiku

[west trail]

But from that photo, it says gpt 3.5?

[visual bough]

I can't see anything, the preview is blocking the result, but you are also using search mode, many websites still have the old information ☺️

[west trail]

Is there a way not to use search mode?

[visual bough]

You can use the "Writing" focus, just go to https://pplx.ai, click on "Focus" and choose Writing.

[west trail]

But also, i thought the search mode was supposed to give information from updated website?

[dim sierra]

You can specify a date such as after:19-04-2024

If you don't want old sources

[visual bough]

Of course but there is probably no website talking about Claude 3 Haiku on pplx.ai

[west trail]

Is it me who doesn't know what am doing?

[west trail]

What about if i want for programming, Which type of focus should i choose?

[steel schooner]

Voice search has serious regressions (bugs, not features imho) after a recent update where it cuts off prematurely during the chat, doesn't show the text as it transcribes actively, doesn't show the result as text, then assumes followup questions based only on the audio response. The prior behavior was better in every dimension. Why the change?

[visual bough]

Writing focus should work fine

[west trail]

Thanks. I will check and see

[west trail]

So what was i doing at first that i did not choose focus mode and asked perplexity programming questions and it was hallucinating like hell? Or was it getting answers from old websites too?

[west trail]

After opening a new thread with writing focus mode, will the rest of my conversation in that thread also be in focus mode?

[sand kiln]

You have set the temperature of the GPT-4o language model too low, causing the output to be overly repetitive.

Look at this

[knotty minnow]

apparently there's some issue with how gpt4o's tokenizer handled Chinese characters

[noble monolith]

deep dive ➤ #💬│general message

please check tutorials in this post

[sinful bluff]

sorry for the late ass reply. was using gpt 4 turbo and i made him realise that hes wrong by telling him to write out the calculation step by step and then he came to the right answer. but then i was like "see i was right, its not '(stupid number that came out of nowhere)' and he replied with sorry for the confusion, calculated the same again but came to the wrong result. so again i needed to tell him to do it step by step (im sorry if this doesnt help you or you dont need it anymore i know my reply is like reallyyy late)

[mossy valley]

I need waayy more photos than perplexity can provide. So I'm asking to make perplexity give users more photos, up to 20 at minimum. And even more if the user wants so

Love your app - that's why I'm using it on every device I can to, and that's why I've joined the community

[noble monolith]

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

[mossy valley]

What? Isn't this a feedback?

The feedback is that I can't have the needed amount of photos while searching by perplexity, and that I love the app, and the idea of AI powered search engine

[noble monolith]

of course, https://discord.com/channels/1047197230748151888/1047649619695390740 channel - if you want to elaborate more on your proposal

[inner wigeon]

I'm very impressed with GPT-4o so far. Really digging the model addition

[rustic hull]

haha its alright thanks anyways

[brave maple]

Today, I noticed with an update that I can preview the link by hovering over the citation link. We are great.. we are getting there 👍 Thank you for this. Now, please quickly enable a quick export of all citations to a RIS format to import into Zotero/Endnote etc. 🙏

[noble monolith]

#📢│announcements message

[half thunder]

#🧪│api-general #1161804761247526912

[vivid raven]

Starting to use collections in perplexity more. It would be great if I can choose a default model for a particular collection. And a default focus mode. This would help in getting what we want quicker.

[half thunder]

good think , i am using collections more and more.
but i want the feature of Pags.

I need new features related to "Pages", please no more greyscale, hope to push it live to all users ASAP!

i need deeply struct my knowledges, and make it more briefly

I need to develop my knowledge system in depth and make it more concise and easier to understand.

[frail fossil]

There’s gotta be a way for the model to selectively apply my location info rather than just handing it to the prompt every time. It’s weird seeing my location in queries that don’t need it

[covert sun]

Hey @frail fossil! Could you send me some threads that unnecessarily refer to your location in the answers.

[frail fossil]

sure

[marsh spade]

If I haven't received the Rabbit r1 1-year Pro email from Perplexity, what can I do?

[covert sun]

Hey @marsh spade! I responded in the intro channel, too (please don’t post the same message in different channels). You can reach out to the rabbit team and let them know you haven’t received the promo email.

[rare plaza]

LIke the tool, been a pro users for a few months now. You have to change the new prompt dialog. You pop up this dialog box but I have had it happen that if I squint the wrong way at the screen the (insert expletive) box disappears and with it takes all of my great work drafting a solid prompt. I want to engage a new thread, but don't trust your interface.

[covert sun]

Thanks for the feedback, @rare plaza! We are working on the box to fix this, thank you for your support.

Hey @finite flume, were you able to proceed with the subscription management? Have you sent a refund request to our [e-mail] email? Please dm me your account details, we'll help you with the request, if it still doesn't work.

[south solar]

Hey , dear admins/founders/team leads, please check recent feedback on #📱︱feedback-mobile

You guys got to catch up fast

[shut arch]

@jhersonyt warned

Reason: Spam, please avoid sending single word messages
Duration: 7 days

[feral harbor]

I asked the free version of Perplexity a lot of questions about which model it was. It was consistent in those replies that it is a separate model called "Perplexity". It is created by Anthropic, but it is not Claude.

[noble monolith]

https://discord.com/channels/1047197230748151888/1225549032718602332

[unkempt atlas]

Something has definitely changed with the GPT-4 turbo model on pplx, today it either completely forgets context by second message or just starts answering one of the internet sources??

[mystic sandal]

rendering latex would be awesome

[dim sierra]

It already does

[mystic sandal]

I guess not always then. This is on android, the same happens on my computer

[dim sierra]

#🌐│feedback-web message

You can fix it by adding that to your system message

[sand pine]

Perplexity app hasn't been showing collections since last 20 odd days. Tried uninstalling reinstalling it multiple times. One OnePlus 11r 5g running Android 14

[oak mauve]

Please release a statement (post) on why exactly Prplxty was down ealier today, what exactly caused it. Especially in future incidents, DO post the reason for it

[lusty meteor]

If I accidentally deleted my library, is there a way to recover it? For the future, having an x next to one line should not delete everything without confirming it.

[harsh lily]

I don’t know where to start but the iOS app has been so unstable for me for a while. I keep getting connection timeout errors. it is actually unusable on mobile now

[noble monolith]

#💬│general message

[autumn vigil]

Hello,

I've got some feedback for the Discover section of Perplexity:

• Number of published news: The pattern shows it's 5 news items published per day by the Perplexity Bot. I think this could be improved to get more and maybe make it customizable for the user by industry, by interest, by region, etc.
• The time it's published: The pattern shows it's around 5 PM (PDT timezone). In other parts of the world, like CEST, it's 2 AM. An around-the-clock system could be better and give a "real-time" experience.

N.B: I know that Perplexity isn't a news site, but I like this feature because it summarizes each topic from different news sites and offers the important things to retain.

[unkempt atlas]

Agreed

[cyan vector]

hi perplexity folks. i'm teaching another round of seminars to a Russian audience, and have been using perplexity as a "known and safe tool they can access without using VPN." I was wondering if the phone app is available in Russia, and whether it's like what we have in the United States?

[visual bough]

Hey @cyan vector!
Unless blocked by an organization or even a government, Perplexity is available everywhere.

[cyan vector]

awesome, thank you so much dude

[unkempt jacinth]

I love perplexity, but the iPadOS is driving me nuts tonight. Several times, I’ve typed a prompt, went to tap submit, accidentally tapped outside the box, and lost my prompt!

[finite cedar]

the iOS app has been basically unusble for me. Mainly with the voice input - I literally have no idea ow it can get what im saying that wrng and then when you go to type its WAY to easy to slightly click out the box and lose the prompt

[sharp badger]

I also experience this. It can be very frustrating.

[half thunder]

@covert sun about return knowledge card.
when i ask "Gamestop vs. AMC stock since 5/3/24", it return the knowledge card ok.
but when i ask "Changes in Whey Protein Futures Prices in 2024 in USD, return as knowledge card", it just return the text.
why ?
https://www.perplexity.ai/search/Changes-in-Whey-A3.1ymXGT_qJTVgNwyWvXg#0

[gusty quarry]

I don't get the knowledge cards at all. I also searched for Gamestop's and there's no cards returned. https://www.perplexity.ai/search/Gamestop-vs-AMC-dSfGLJkETgOxP7xG0s.SEg

[outer thunder]

I wonder if there isn't value in a "check the answer" function where, before it posts the final response, it feeds the response + the sources used just as in-text references in the response into the answer engine to double-check (similar to what an AI agent does?) - I'm occasionally seeing answers that are incorrect, where I almost think it hallucinates the answer because it's entirely unsupported by the references

[vale lily]

Voice chat is responding to me in German because of my location (city in German) but I started the conversation in English and expected a response in English. Instead, it translated my original English message to German, and then answered in German as well. Can this be considered as a bug or just an unwanted side effect or is this actually a feature?

[dim sierra]

A bug

[vale lily]

Thank you

[covert sun]

Hey, @vale lily! How about the settings on your mobile app? If you set the default AI language as automatic, it should recognize the input and respond accordingly. If your default lang was set to Deutsch earlier, you'll get answers in Deutsch.

[vale lily]

That was my setting.

[vale lily]

it was set to autodetect but location to Nuremberg, Germany. I've now unset the location and AI language as autodetect, that seems to work. Thanks for the tip

[vale topaz]

Could we have login methods for perplexity pro that:

1. let us use our own password instead of needing email to login
2. stop logging us out if we use a VPN
3. don't get vpn blocked or VPN checked if logged into the pro account

[somber kelp]

On 2 and 3. I don't think I've ever been logged out..? As for the vpn check, I think it's outside of the perplexity teams hands, as the check is done by a third party before the request hits the ppl servers

[vale topaz]

they can redesign things to not work that way, for example they can serve a static front page, do a login check on the client and then do a vpn check before any input is accepted. I have been logged out multiple times

[signal isle]

I have to say the new decision over limiting the files to 4 pieces is an awful one, I don't know if the system was being abused or what but 4 looks like a PM just queries the average number of files, then decided to limit it to that number and call it a day.

I believe the job to be done here is to get answers, while uploading 100s of files seems unreasonable, limiting the system to 4 files at a time is just frankly, in my opinion, not only greedy of you at $20 a month but also detracts from the value the system offers.

[covert sun]

Hey @signal isle! Sorry for any confusion, the 4 files in one question has been the number in place since we added the file upload feature and there are also other things to consider like file size. Uploading a lot more files doesn't guarantee a better answer, so there are trade-offs to consider.

[solar obsidian]

It would be nice if in collections we could edit threads titles. That way it would make it easier to find the one we're looking for, especially if we asked lots of questions inside of that thread.

[crimson nymph]

you can edit thread title by clicking on the thread in the collection (also outside of a collection) and then you can edit it in the top middle.

[half thunder]

perfect ！！！！！！ pages is coming to me

https://www.perplexity.ai/page/3B_gUzweRRmb.pqashT5hw

amazing ~~~

https://www.perplexity.ai/page/Vitamin-B1-Recommended-XHzms5JRTb2NLxD1CkhSbg
amazing~~~~~~~~~~~

[solar obsidian]

Thanks @crimson nymph ! I never would have figured it out on my own.
It would be useful though to also let people edit the title of their thread in this menu:

[knotty fiber]

I like to say, I barely use Google or any other search engine, I mostly use Perplexity. If searching and especially content generating will be faster, I will no more use Google entirely.

[remote herald]

Where have the Pages gone?

[knotty fiber]

Thank you for Perplexity. I have found a potentially effective way to mitigate symptoms of IgA deficiency, which my mother has.

I found it through academic papers.

[mild pendant]

I have officially switched to perplexity for regular searches already

I have no clue how did you guys pull this off so beautifully but this is a damn nice thing you have got here

[knotty fiber]

Same. I have a personal theory that it's due to the evolution and growth of the Internet's size.

The Internet is so vast today that it's beyond the point where manually searching websites and URLs becomes ineffective. The time required to find the information you need grows exponentially.

So, the only obvious solution is to automate the search process using a smart, autonomous system. In this case, "AI," more specifically LLM, is very good for this.

Perplexity.ai is only one tool, a company that does exactly what's needed.

[heady plover]

how do you get 6 months from copy.ai of perpelxity

[cedar pendant]

Make short form videos to explain how things work, like the moons phases

[olive path]

I see this exactly with other prompts as well. It worked very well with the provided examples (or stock-related prompts only). Some other users confirmed the issue also.

[stoic latch]

How often does perplexity search get updated? One examples, last few weeks I’ve noticed, it still refers to Azure Entra ID as Azure Active Directory. I’m new to Ai, using the Claude 3 opus search with pro.

[wild storm]

Inside library, when you go to search for older threads, you aren't able to find older threads unless you scroll through all the threads and load them up. This got me thinking, is the search feature on-screen contextual and not reliant on server memory of my threads? So is search going through all the threads I might have that is stored in the server? Or is it more like a find on page type of feature?

[dim sierra]

It's probably vector search, I would imagine.

[wild storm]

So that means that its not searching through the perplexity servers and threads related to my account but instead just looking at all the loaded threads, visible to me on my screen, and answering from there?

-- Sorry I'm not too well read on vector search

[dim sierra]

The vector search could be server side.

If you remember an old thread, you could try to mention something related to it to see if it shows up.

[wild storm]

Yeah... did that... isn't able to fetch it! It's only when I scroll through the library and all the old searches gets "loaded" and I can see them is when I can use the search feature to pin point particular threads.

[sharp flower]

how do I get the tag needed to post in bug reports?

[sweet ruin]

Getting 500 Internal Server Error for all of my Perplexity API utilizing applications right now. API status page shows nothing off? What's going on? (The implementations worked without issues previously.)

[half thunder]

@shut arch
there is a bug about page: https://www.perplexity.ai/page/DDoS-attack-on-feEOYl8wTlaT66qP1Y.Qyg
source :https://mastodon.archive.org/%40internetarchive/112513905401989149
the last source link is not work.
fix: https://mastodon.archive.org/%40internetarchive/112513905401989149 => https://mastodon.archive.org/@internetarchive/112513905401989149

%40 => @

[tranquil trench]

Hello! Did anyone get the "Perplexity AI Companion for Chrome" to summarize and respond to queries on the content of the pages you're browsing? I tried reinstalling it but always get the error "I am unable to access the content of the provided URL. Please provide more details or another source for me to summarize." Any tips? (Using the paid version of Perplexity)

[steep ginkgo]

Hello. Is support real? I created 3 forms explaining my problem, two yesterday and one today, but none received a response yet. And I suspect I won't get any. Is there any chance I can get real help from you, perplexity guys?

[visual bough]

The response time is several business days, please be patient

[gusty tide]

needs change email option

[slate dew]

New "Perplexity Pages" are great! Congratulations. I think is very important a caption for images...

[knotty minnow]

i think they have an arrangement with Unsplash for the stock images

[slate dew]

No, they get images from sources but there is no captions for copyrights attribution and to contextualize the image (as in this page https://www.perplexity.ai/page/Abandono-del-Territorio-4P6Y6L1wQ0qdY0rbvljrxQ)

[slate dew]

Ah, would be nice a card preview with page title, description and image 😅

[knotty minnow]

oh true, i see what you mean.

you're saying an explicit attribution in addition to the link in the top right yeah?

[slate dew]

Yes, because you have to click twice to get some information...

[heady plover]

wow pages is cool. https://www.perplexity.ai/page/Best-Value-Hilton-hLLm5H5fToGLQP2SyKZU0Q

what's interesting is as I set up a page, I wanted to add to it. I notice you can't add new search results to a page. Also the intro doesn't make much sense. I ran a search for hilton hotels in barcelona, then amsterdam, but the title says "Best Hilton in barcelona"

[true mica]

Wow pages isnt working 😦

[covert sun]

Wow pages isnt working 😦

[true mica]

I get a screen (I can't screenshot it as it goes black ) as soon as I click the link in the email saying create your first page today. It goes to the pplx app (android 14) and after a few seconds it tells me ' something went wrong .please try again '

It does that for over an hour now

@covert sun the thread you created for this . I cant reply to.. do not have permission .

[zenith nacelle]

Hi - Loving the idea of the new Pages feature. I might be looking in the wrong place, but is there a way to directly edit the text produced in the sections of the Page? For example, an otherwise good Page is returning a couple of factual errors that can be easily manually fixed. Users would also need to be able to edit the title and intro text. For example, this Page is about an upcoming event, but it is presented as a past event: https://www.perplexity.ai/page/Tell-Me-About-hFwZevUfSoedmmEL9gubNA It would be really useful to be able to include sections that have manual entry as an option.

[fathom perch]

The pages function is crazy cool! But why is it that even though I set the language to English in the settings, when I enter the title text in pages, such as "elden ring noobs guide", the generated pages title and description sections are in Chinese, and I can't edit the description section. Apart from that, the pages function is awesome!!! I really love this function, keep it up!!

[dusty bison]

When will Mozilla Ai K2 model be available? https://huggingface.co/LLM360/K2

As well as Apple's OpenELM? https://huggingface.co/apple/OpenELM

[fleet elk]

The pages feature is fantastic!
Rather than ask for specific features, I'll just say that I'm looking forward to rapid development and feature rollout. I can't wait to see what the Perplexity team has planned for this tool!

[heady garnet]

Can I stop the model changing to GPT Omni? It keeps changing automatically even though I have my default set to Turbo. It produces poor results for some technical inputs and I don’t want to use it, but even when I ask to regenerate in Turbo it keeps giving me Omni. Please fix or advise me how to only use the models I select

[ornate needle]

Came here to ask about the exact same thing. Pages is great for getting a rough draft together quickly, which is a huge part of the usual blocking points for writing. But how do we (or, why can't we) edit the page title, description, and section headlines? Without that, everything I write will stay in drafts, to be copy pasted to other places for editing and publishing. Still super useful, but if we can edit those key parts it's an order of magnitude better

[safe fractal]

The voices could be better. They mispronounce a number of English words when speaking

[meager coral]

Hi - I'm not sure if this is where to give feedback on Pages. I found a few things irksome despite generally loving it (I've published 2):

1. I found at least one error in a citation attribution, it would be nice to be able to correct these? Sharing makes me feel awkward when I know there are errors.
2. I want to be able to edit the section titles and the main page title -- like, it's not really "an intro to Deleuze", it's "an intro to deleuze on art and hermeneutics" because of how I added the sections
3. It didn't seem to be paying good attention to the entire context of what had been generated so far. Like if I add a section on "quantum mechanics" to a page about Deleuze, I want it to take that into account and relate it. Of course I can change my prompt question, but this seems basic -- especially for not repeating itself, staying on topic, etc
4. adding - sometimes a source link is actually a duplicate article, like a research paper posted on 2 sites. this seems concerning, not sure how you handle that as weighting

also generally ability to edit the text would be nice -- you could flag that in a UI view of some kind if you think humans are worse than your AI

[hearty stump]

they are working on the 2nd, for the 3rd point i also noticed this problem

also you need to post this on #1111765634267742408

[meager coral]

It says "No Access" for me

[hearty stump]

you have the pro version right?

[meager coral]

yes

[hearty stump]

alright leave the server and go to your perplexity settings and click on pro discord server

[meager coral]

aha - thank you

[crude bough]

Hai

[knotty fiber]

Update to my usage of Perplexity. I stopped use Google. I just don't use it on desktop after a new pro search 🙂
I only awaiting for update of app on Android.

It's easier and only cost a 20USD/month. Time saver, verification of information, it's so easy and intuitive. 🙂

[slate dew]

Can you use Perplexity for image search too? 👀

[knotty fiber]

I literally uploading images of plants to recognize them 😅

[frank snow]

Hello. Why does my Perplexity page look like this? I found it this morning, my IP is from China.

[heady plover]

tried reloading?

[near belfry]

It did the same thing for me, try reloading.

[covert sun]

If refreshing and connecting via a different hotspot doesn't help, please share the thread URL, we'll check it.

[ruby geode]

I have Perplexity Pro plan, and wish to analyze XLS files using the GPT 4o model (which, in its native ChatGPT app supports XLS files). However, the file upload of PP Plus only support CSV upload and NOT XLS. Is there a way to enjoy GPT 4o's XLS data analysis capabilties right from PP Pro?

[dusty meadow]

Hi! Any idea if Perplexity will ever have a code interpreter? I see that it uses steps when asking hard questions with Pro search, and I think this can be good for it to execute code, like for doing math calculations. Is this a planned thing?

[feral helm]

Hi, can you create a "continue" button like openai chatgpt for long code generation?

[dim sierra]

I was double charged this month for the normal $20 and then again for $10 for referring someone. Pretty sure that isn't supposed to happen?

Not only that but since the virtual card is capped at $20/month the second one bounced so for some reason pro was revoked?

[dim sierra]

I sorted it out on my end, but really perplexity should fix up their stripe integration.

[karmic tulip]

How do I get perplexity to answer questions without searching the web first? it seems to get distracted by whatever rubbish has turned up in a search each time, especially if asking follow-up questions

[dim sierra]

Change the mode to writing mode.

[karmic tulip]

oh! how do I do that?

[dim sierra]

Click Focus, then click Writing

[karmic tulip]

oh amazing! thank you

ah but you can't do it halfway through?

[dim sierra]

Nope, not currently, unless you use an extension.

[frank snow]

https://www.perplexity.ai/?__cf_chl_tk=CC5BRjqK4Sk1ZGXmAjQZpcRbiC2hAJqliCr8_g_2ls0-1717468691-0.0.1.1-4265

[sleek stag]

It would have been nice if Perplexity supported native TTS, not the paid one 🗣️

[covert sun]

Could you check the URL and make it 'shareable' in the top-right corner, @frank snow. Thank you!

[void trellis]

I'm a bilingual user. When I was a free user, Perplexity could just respond in the same language I use. But when I became a premium user, it no longer changed the language automatically, and it even failed to follow the instructions like "use English" sometimes.

[sinful crypt]

It'd be nice if perplexity had a proper status page and/or communicated more proactively on public platforms about outages. I shouldn't have to join discord to find out the status of a service I'm paying for.

[near crystal]

Hello, I have a question. When I use the Perplexity API, the response I get seems to be generated using the standard Perplexity model, not the Pro version. I cannot specify in make.com that I want the responses to be generated using the Pro version. How can I solve this problem?

[covert sun]

Hey @near crystal! Pro search is not provided via the API.

[near crystal]

Do you plan to offer it at some point? thanks!

[covert sun]

Currently, I don't have any info on such plans.

[near crystal]

thanks! that a boomer

[gaunt urchin]

I had the same problem

[icy wagon]

@covert sun Unable to subscribe to Pro. Even after entering OTP and always the payment failed.

[gleaming halo]

I bought a subscription but I can't choose a model, help me, it returns to a simple one

[somber kelp]

it was mentioned on one of the pro-channels that restarting browser might help. Not just reloading the page but restarting. Something about cloudflare

[mint echo]

I used a trial of perplexity pro 4 months ago, didnt see where to cancel and figured deleting the account would be enough. Found out Ive been charged the last four months via cashapp for a service I have no access to. Wish that wouldn’t be a thing…

[valid gull]

https://www.perplexity.ai/search/C4AI-CRXNcwgIRWuECTmoXIIKjQ

I got this search result when searching ofr C4AI which is short for Cohere for AI

Looks like the agent behind the scene didn't format the response properly or the output parser failed

[calm turtle]

IMO the upload size limit is incredibly limiting, if the upload size limit was more reasonable I'd buy pro right now

[gaunt urchin]

It seems that the stylesheet failed to load, which could be due to an accidental modification of the DNS settings or some other issue. In any case, the problem was resolved after I restarted my computer.

[lime talon]

anyone aware that claude 3 haiku disappeared from labs perplexity? Will they come back?

[velvet escarp]

I've been a fan of Perplexity for a long time, but at the moment I have the feeling that Perplexity is lagging a bit behind.
Perplexity is more of an interface to get to other, better artificial intelligences like Claude 3 Opus or GPT-4o.
Basic functions like the visualization of things/graphs or tables almost never work or don't work at all.
In math programming or academic areas Perplexity is unfortunately also super behind and as the vote in the feedback thread shows almost nobody uses the in-house Perplexity AI which makes it a real shame because Perplexity used to be a real pioneer in AI.
I hope that in the future there will be a new model that will overshadow the rest of the AI market, back to where Perplexity came from and become one of, if not the best AI again.🙏

[hollow basin]

What is the current upload limit?

[high lava]

Is there any way to control the preferred sources for news for example? Or to instruct perplexity to avoid certain sources like cnn? Also would be great to add some lesser known sources too

[grand holly]

Add option to refresh chat without losing thread with it's uploaded document. By refresh, I mean fresh start.

[wheat dome]

Hi, I was wondering if there is someone on the team I could talk to set up a workshop about perplexity with some of the students I teach. They are young innovators from around the globe using AI to tackle societal problems. Talking to someone from the @shut arch team would be extremely valuable for them to think out how to build out their own ideas. Looking forward to connecting.

[oak mauve]

Please disable (or massively reduce) these insanely obnoxious Cloudflare verifications for those of us users who are logged in into a human-confirmed Perplexity account.

[dim sierra]

It's for anti bot protection

[oak mauve]

Sure, but I have a Perplexity account since MANY many months now and I'm ALWAYS logged in, so the site should check that (via browser cookie) before doing that very disruptive Cloudflare thing.

[dim sierra]

A bot can do the exact same thing...

[oak mauve]

A bot wouldn't submit meaningful questions though, right?

[dim sierra]

Yep it would.

Since it would likely just proxy another service.

[oak mauve]

Oof. But a bot can't confirm an email adress, right? (since perplexity requires the user to click on a link in the email)

[dim sierra]

Human => Fakeplexity => bot => Perplexity

[dim sierra]

It can.

[oak mauve]

What??? How, what is your source? Is there an example on YouTube somewhere?

[dim sierra]

I could easily make one in a few minutes.

With a headless browser.

[oak mauve]

What about if we could register our phone number with Perplexity, isn't that also the reason why OpenAI has that, so that it doesn't need Cloudflare stuff like this?

[oak mauve]

"headless"?

[dim sierra]

Basically let's you control the browser, so I could easily navigate to my email and click the link to verify the email...

[dim sierra]

Headless is the browser with no screen.

So you don't see the actual page rendered.

Because it takes computer resources.

[dim sierra]

Nah, you can also bypass that.

Esims have made that even easier.

[oak mauve]

So OpenAI's ChatGPT website also has a verification like this Cloudflare thing?

[dim sierra]

Yep

It gets triggered if you are using a popular VPN.

Which I assume you are.

[oak mauve]

LOOOL okay. 😆

[oak mauve]

No. But our german ISPs force-change our IP-adresses every 24h, so I guess that's where this Cloudflare verification stuff gets triggered

[dim sierra]

Possibly. But normally normal ISP's don't get flagged.

[oak mauve]

Also, why is there no such verification in the Android app? (thank god that that doesn't have it)

[dim sierra]

It's normally vpns and proxies that do.

[dim sierra]

There is, you just don't see it.

[oak mauve]

How do you know that?

[dim sierra]

But with the app, they can do more tests to check you're not a bot.

[oak mauve]

Like touchscreen input analytics, or?

[dim sierra]

No, more complicated checks.

You can still bot them though. It's just harder to do.

[upbeat raft]

I wonder why PPLX apparently is hardwired to pretend or outright 'lie' that it could learn or adapt from personal user feedback (in a broader sense), when in fact it totally can't? that's misleading and tbh it somewhat hurt my feelings when I learned about it 🥲😅

it was pointed to me here on the discord and I had some talks with PPLX about that since.

"you're totally right, my previous claim that I can learn from your feedback was false and I shouldn't have said that. thank you for pointing out my mistake to me, it helps me to give more accurate information in the future" (paraphrasing)
😄

[vivid wedge]

Any way to get insta access to Pages as a paid Pro user?

(And if not... my feedback is that I was surprised to find that to be true. Also sent via email.)

[white stratus]

Can we get Vision AI inside Perplexity!? https://vansh.one/vision

It's just straight up better than gpt4o and can handle cultural context specially for India.

[jovial temple]

everyone has access
clear your browser cache and local storage

[oak mauve]

Hey, but why does that page sometimes appear but then solves itself without me doing anything? 🤨

[calm turtle]

25MB

[dim sierra]

Because it probably checks if you already have a token, and if not, you have to click it to get one. Tokens probably have a short lifetime of 30m - 60m

[oak mauve]

lifetime of 30m - 60m
ooooooof....

[oak mauve]

lifetime of 30m - 60m
@dim sierra doesn't seem so. I just accessed the website for the first time today, and it instantly passed the verification without me clicking anything

[dim sierra]

Yep, it depends on your ip

[oak mauve]

how so?

[dim sierra]

Since it's cloudflare protection.

When I'm not using a VPN, I don't have to wait for the captcha at all.

[oak mauve]

But that makes it sound like Cloudflare "rates" IP adresses in some way...

[dim sierra]

Yep they do, if it's not a normal ISP IP or if it's been flagged for possible abuse.

[quaint heath]

What's the message limit per thread after which they start disappearing? I noticed after it got over 100 or so the first ones started getting deleted, and I lost some valuable data. It also doesn't help that the label with the current amount of messages in a thread seems to have disappeared, I no longer see it on web or mobile app.

[rustic ermine]

I love the ability to share my search results and the speed of the LLM.
Perplexity also regularly outperforms google and chatgpt, which is an impressive feat considering how many corps aim to achieve this.
Good Job.

[high lava]

is there seriously no sign out button anywhere? I have a pro account and cant figure out how to sign out so it doesnt log me in with google and not use my pro account...

I even tried clearing localStorage & sessionStorage, nada

I finally fouuund it buried at the bottom of settings, disregard

[fervent nacelle]

In settings

[blissful geyser]

The same problem, this Cloudflare verification is very frequent, hoping to reduce the frequency. I am a real user. Every time I ask a question after a period of time, it will freeze, and then refresh the page to know that it is because of triggering the Cloudflare verification.

[oak mauve]

Then maybe it is caused by your internet provider being rather untrusted according to Cloudflare?

[vivid iris]

Please make it, or atleast add a setting, where if the "pro" feature is on, it doesn't carry onto other threads or new ones. It feels frustrating to see a pro search being wasted.

[blissful geyser]

I don't have this problem with my other programs

[meager fable]

When you paste long text its annoying that it gets turned into a txt file. Because i need to double prompt. My paste.txt already has my prompt but it breaks if i send an empty query with only the paste.txt. it would be better if you paste something you should be able to send it empty and it just does what the paste.txt asks. Because currently i have to say "do as the paste.txt asks"

[grand holly]

Can we get option to attach screenshots here?

[dim sierra]

Working for me

Which model are you using?

[covert sun]

Please share the thread with this incorrect answer, we'll check. Thanks for reporting!

[glass sorrel]

How come you guys still didn't implement a feature to switch models within a thread without having to either rewrite the current message or send a new message and rewrite that one?

[grand holly]

https://www.perplexity.ai/search/Chatgpt-alternatives-that-05Y2Y5SuSjCThKIoPCQluA

[grand depot]

Just wanted to share a feature that might be offered alongside "Pages." I've not warmed to Pages because what I usually want to share is portions of normal conversations, usually with Claude Opus. I say portions since there usually are a few warm-up prompts that lead to an amazing back-and-forth of various lengths. This shareable moment often ends with something less relevant or otherwise less interesting to share. -- So, ideally, I would be able to click "SHARE PORTION." This would bring up checkboxes by each prompt/response pair. I would check the boxes I want to share, and this would be content that the share-ees would see. -- This would let us share the already-made content that almost demands to be shared as-is, without rewriting. And, when presented to others or re-read by me, it would be shown with sources, but without the usual intervening suggested questions.

BTW, I asked Claude if it understood this suggestion, and it provided its own extended (and personable) version. https://www.perplexity.ai/search/I-want-to-4uGmWGPgR4atutkTPUV1Xg

[feral helm]

perplexity can't watch a youtube video?

such a shame

I m asking a for refund now

[somber kelp]

To my understanding they are crawling youtube videos, but .. they need to be crawled first. Actually watching / analyzing a video could easily take as long as watching the video by yourself

[grand holly]

@covert sun were my screenshots deleted from here?

[covert sun]

I don't think so: #👉│feedback-general message

[sacred schooner]

Would be great to be able to set the AI model at the collection level. For one collection I might want to use Claude 3 Opus, and for another GPT-4o.

[untold orchid]

I was trying to set a rule at the beginning of a thread but no hope
https://www.perplexity.ai/search/here-is-the-uxLVh5CmQiazQlMIzS.HCQ

[knotty minnow]

You should try using the Writing (i.e. Offline) mode for something like this; it will prevent it from getting articles from the web which are not really needed to complete the task

I tweaked the prompt slightly (which I admittedly didn't fully understand), but perhaps this is closer to what you are going for https://www.perplexity.ai/search/here-are-the-iniTMTMIQp6k7VmcHkjdGA

[shell palm]

Hi, all. Am I not understanding Collections properly? I like to make Collections about specific video games to ask related questions. Both the name and prompt contain the name of the game, yet when I ask questions, Perplexity doesn't seem to understand the context. It felt pretty evident to me, so I was surprised to see it didn't. If someone can explain to me why it does not, I'd appreciate it.

[celest fjord]

I’m sure it’s been said before but I really like listening to content created however I really hate how slow it goes. if I could speed it up to 1.8 to 2.5x I would use that feature all the time. It’s too slow for me to listen to now though.

[covert sun]

Sorry for the inconvenience @shell palm! We are working on a fix. Could you try a non-Pro request and let us know if you got a better output. Also, please share the thread URL, I'll add it to our ticket.

[shell palm]

Thank you! Here is the collection link with four threads (all Pro requests with faulty results). I tried a non-Pro request (last thread) and the result is still not within the context.
https://www.perplexity.ai/collections/Sons-of-The-8__F49DISqKCkJ0_zgTK5Q
Regardless of this issue, keep up the good work with Perplexity. It has pretty much replaced Google for me. 🧡

[sharp flower]

Claude model refusing to access websites published after its training cutoff

Haven't done extensive testing, but Claude 3 opus will refuse to access links I provide that is published in 2024 even though the link is listed in the sources section (it has a 2023 knowledge cutoff). I suspect this is due to claude's system prompt or RLHF.

Looks like anthropic gave explicit instructions to reduce hallucination and not produce content about subjects occurring after the training date, not much of a pplx issue, more of a Claude issue

Potential fix: ask anthropic to remove the system prompt that prevents this just for your api endpoint (I think it is a system prompt).....or try extracting website content and feeding to claude without mentions of dates/related metadata

issue with displaying math formatting:

posting here because 'a tag is required to post in this channel' issue for #1140622008086970420 .... not sure how to get the tag

[celest lagoon]

After a while ALL models are starting to give me the exact same answer. This is fishy and feels like Perplexity is just routing to one of the cheapest models without informing me.

[noble monolith]

#💬│general message

[silver tree]

Have a problem with Pro being unavailable each time I open the perplexity website (see pictures). When I go into pro settings to select a model and attempt to remedy the situation, it always resets the selected model to the default (for reference I usually run Claude 3 Opus and DALLE-3). I can get the pro "toggle switch" to display again if I go into the pro settings and select my preferred models again, but it is very tedious to do every time I want to use Perplexity.

[kind nymph]

I want to see the perplexity discord bot\ back

[pastel wigeon]

is there any affiliate programs?

I think i could help sell subscriptions

[keen harbor]

I cancelled the pro subscription, but still got charged for it after 2 days. I need the refund now. I have been contacting support team but there is no response.

[visual bough]

Please be patient, the response time is several business days. Thank you!

[dim sierra]

Yep, also it was/is sunday, to you likely won't get a response.

Just wait for monday, for when people are back at work.

[north raft]

It feels like there is 0 context with Perplexity

I'll ask for a snippet using langchain, and then I follow up expecting it to remember context and it talks about user settings in WhatsApp

It hasn't always felt like this, and it's not usable for me at the moment

[somber kelp]

Out of curiosity, can you share the thread?

[covert sun]

Hey @north raft! As MasseR said the thread URL would be helpful for the team to investigate. Thank you for reporting!

[pastel wigeon]

do you have affiliate program?

[covert sun]

Yes. https://www.perplexity.ai/hub/faq/coupons-and-discounts

[pearl palm]

There is some talk about you adding ads.
I hope that does not happen..

[fallow latch]

I don’t know what this ‘talk’ speaks of, but if anyone doesn’t expect massive commercialisation at some point they’re delusional.

[pearl palm]

@fallow latch Fair point. But for the moment it is a very good alternative to the other search engines for me.
Would like to see it monetized in a way that did not destroy it for me..

[quaint heath]

The message counter in theeads has disappeared, in both web and app. Is that intentional?
The number helps because apparently if you read 100+ messages in the thread, it starts deleting the old ones. That's strange behavior. I understand running out of context, if the AI has to forget the old stuff to make room for new that's all good, but I still want to be able to view all my messages at any time even if AI forgot their context

[north raft]

This is an example:
https://www.perplexity.ai/search/in-langchain-how-C0Rqq9lbRQSv3e6yR_H8Wg

I've been getting it a whole lot recently though

[cold anchor]

yo! lemme select & copy texy, idk why but it's impossible to select only certain words & copy them .

1. why this happen ?
2. why this decision ?

[heady bison]

Hello! I’m an Instructional Tech Trainer for a school district in Florida. Your Perplexity Pages are amazing and I use them regularly.

I remember talk about this feature coming to free users. Is there any timeline for when that will be?

I’m training teachers and principals for the coming school year and really want to incorporate page creation into it, but can’t until it’s available for free users. Any idea if it will be done by the start of the school year (August)?

[spice elbow]

Exposing collections issue -> #💬│general message

[ruby pumice]

Has there been any discussion or suggestions for IFTTT integration?

[rustic sandal]

Hey, is anyone having issues with repetition in threads? I use perp regularly and only noticed this happening as of yesterday night (GPT-4 Turbo and Pro enabled). Follow up on a thread gets a nearly identical answer generated, sometimes disregarding the follow up question/input.

Example 1: https://www.perplexity.ai/search/What-do-chained-JDlqwBS_Rdq_b.Grp2k8Pw
Example 2: https://www.perplexity.ai/search/Im-moving-into-XA.3ywBRSxOSPuI9MRf5EA
Example 3: https://www.perplexity.ai/search/How-do-you-Y30N8seUQDKRrr2_GXltcg

Looking at the comments from the past two days, looks like I may not be the only person having issues.

[static storm]

I have the same issue. Since some days.

[lyric nimbus]

sounds a bit similar to the issue I had #👑│pro-lounge message

[silver tree]

Thanks for the reply, but I don't have access to that thread/channel for some reason. Just says "No Access".

[rustic sandal]

Yeah, just happened again
https://www.perplexity.ai/search/What-are-the-ZCZZEESVTougRRYdz_ny4Q

Oh god, am I going to have to google things!?

[tawny urchin]

Is there a shortcut to selecting an AI model?
It's annoying to go into the settings every time and change it.

[lyric nimbus]

oh, sorry, I thought pro-lounge should be accessible for pro users, but maybe not history? not sure. anyway, here's a screenshot

[silver tree]

Dude thank you so much! I’ll try erasing the cookie and logging in again

It worked 🥳

[peak plaza]

Hi I got a 200 dollar discount code and bought a one year subscription. Everything was fine and dandy. Then 5 days ago - Last friday I get this - "We've canceled your Pro Subscription" No reason or anything. I have tried to reach out but no answer. What do I do?

[open wigeon]

did you get that discount from a trusted source? there have been many cases recently where 1 year subs were canceled because of that

[peak plaza]

Yes I believe so. Its was from Rabbit R1.

[open wigeon]

"believe so" isn't a good point to even start the conversation 😅

if it wasn't your own Rabbit R1 purchase, you should contact the person from whom you bought the code instead of contacting Perplexity support

[peak plaza]

Well since I havent been able to read anyone else being cancelled. But I Havent gotten any help from Perplexity yet so a bit in a tangle. Since I dont understand why yet.