⚡│ask-community

[craggy crypt]

just found the pinned message. so this channel was originally intended as a place to ask about asking perplexity quick questions, not a place to ask quick questions about perplexity?

[gleaming pendant]

Why does calculating the Excel File Data (CSV) sum in Perplexity get the wrong value?
If I test it in Chat GPT based on the same Data File, do I get the correct answer and there is something wrong? Or is it just me?
I used Pro version now

[feral rivet]

What model did you use, GPt4o is probably best for that task

[fluid lark]

It's only in beta and not publicly available?

[feral rivet]

It’s on the chrome store but you can’t download it unless you are invited, you can DM @regal field if you want him to add you to it

[fluid lark]

Looks neat, but I think I will skip. Prefer to use official tools, and I'm using Firefox anyways. But thanks anyways!

[feral rivet]

🙂

[shrewd knoll]

I guess we will only get perplexity.api llama 3.1 405b next month 😔

I want to try so badly 😢

[gleaming pendant]

I also used the GPT-4o model. Nevertheless, the same results came out. :<

[honest turtle]

Perplexity can't browse reddit posts?

I getting this "It appears that the content from the specified Reddit page could not be retrieved. Without access to the actual posts"

[native oar]

quick question, is there any roadmap in few month perplexity live on WearOS? thanks

[tidal osprey]

Google just bought the rights to search Reddit, and Reddit has started blocking access for every other search engine.

[hollow hound]

this is likely why: #💗│sharing message

[sudden ocean]

how big is your csv file? for small files it seems to work (at least in writing focus without pro; not sure how programming step would fare) https://www.perplexity.ai/search/respond-with-strictly-exact-co-0yFrYIREQ_CVHa5b8aeaSQ .
edit: seems to also work in more typical use (all focus, pro enabled, without asking for contents) https://www.perplexity.ai/search/sum-quantity-2fg7qHqPR.qu80CrYqHIBw

[versed magnet]

it's interesting.. seems giving a specific URL for reddit and it sometimes fails to crawl it

but not always

[sudden ocean]

may reddit didnt manage to ban all the pplx ips?

[versed magnet]

and then using an operator and targetting a subreddit (rather than a specific page), it seems that the Pro search process will successfully retrieve items (i'm guessing whatever it has already indexed)

[versed magnet]

yeah though it's not just about crawling new stuff - there's the indexed pages too

[vapid maple]

if you give a specific url, it now behaves differently: it performs wget/curl and parse the file

that is my view

[versed magnet]

i assume the search engines with indexes are respecting some request from reddit.. i don't get how overnight they could jsut remove all those pages, going back years and years, but it seems they have 🤷‍♂️

[vapid maple]

the parser is not perfect, sometimes it just failed...

[versed magnet]

yeah but there's wider context - not on the technical side https://www.perplexity.ai/search/reddit-blocks-unpaid-search-en-WJFohUK4RqaMBeKN5qm_6A

[hollow hound]

why I made it into a Page

[versed magnet]

mine was a tongue-in-cheek adaptation of that very page 😉

[hollow hound]

the change started as a means to block specific crawlers like PPLX

[ivory anchor]

Well, Reddit is mostly bots these days anyway

[versed magnet]

History will recall when the de-indexing of the internet begun after a backlash to a Perplexity Page about Jeff Bezos and drones, derived from a Forbes article based on original reporting

i kid i kid..

i mean.. hope not

not being able to site:reddit.com on search engines other than google is kinda wild..

which website is next..?

[ivory anchor]

Yarr

Well, you can search Tumblr maybe? People totally still use Tumblr :x

[wild perch]

Sorry if this isnt the best place - but I am having a bunch of issues with Perplexity not abiding by time constraints when I ask for a search query - for example, "Give me synopsis of the new announced AI features from LLM companies that were posted this past week". Is there a better way to get it to actually evaluate timing more effectively?

[haughty brook]

Hi @wild perch you could use the after search parameter to only get sources from after the given date. #1262520766642131054 message

[wild perch]

EXACTLY what I was looking for. Thank you so much!

I guess one other quick question - is there a way to schedule a query to automatically refresh in a collection? So if something changes over time, I can have the query automatically run as opposed to having to start over and re run it again?

[haughty brook]

That is not possible. If you would like to have that added you could add your suggestion in the #💭│ideas channel.

[wild perch]

Done - thank you again

[broken prairie]

bumping if anyone knows the answer

[vapid maple]

yes only sonnet opus + gpt read images

[broken prairie]

thanks for your response. super unfortunate, i have gotten bad results from both lol; hopefully somehow working with images can overall be improved

[hollow ginkgo]

Hey @broken prairie!

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.

[gleaming pendant]

The file I tested is less than 300kb in size. The total value of the text after the question is different in text, not in a separate coding method, but no matter how much I look at it, I don't know why. When I tested it on ChatGPT(4o) platform with the same file, I don't think there's anything wrong with the file or the prompt, but I'm a little embarrassed

[ornate swan]

Hello there a quick question I signed up to perplexity with an email provider that will be shutting down at end of this month(skiff.com). And I don't see anywhere where I can change my email, help please 🥺

[hollow ginkgo]

Hey @ornate swan!

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.

[stark gust]

Hello, how perplexity decides in which language it creates search query? I ask questions mostly in slovak, and sometimes it searche in slovak, sometimes in english. I'm not complaining, it works pretty well, I'm just wondering. It even works quite well when I tell it to search in a specific language and only on the TLD domain of that country.

[limber ermine]

What does this mean? Why it tells "updated 9 hours before"? Actually this query has just been done 5 min before...

I had never ever seen this before, neither had I encountered this in any other query and its result.

[shrewd knoll]

getting perplexity.api llama 3 405b gonna take so long time to reach 😩

Next month

So long 😭

[dim brook]

perhaps it's related to your sources, not your query...

[long gyro]

Hi, when I use Perplexity with Web search functionality, I notice it always tends to use the first sources it can find. Is it possible to make it search again and rewrite the answer using more search results? Or make it expand the number of search results it uses to write the answer?

[velvet tangle]

Hey @long gyro!
Pro Search uses more sources and can also perform several steps in between ☺️
https://www.perplexity.ai/hub/blog/pro-search-upgraded-for-more-advanced-problem-solving

[long gyro]

@velvet tangle OK, thanks!

[versed magnet]

Additional research is required to properly address the query.
Based on your PREVIOUS RESPONSE (above), identify 2-4 core ELEMENTS of the information request.

Research EACH identified ELEMENT independently; decompose the element into MULTIPLE [>3] subelements and craft QUERIES for each appropriately

---

[provide updated, comprehensive response based on new material as retrieved]

you can try something like that as a follow-up

ymmv/depends on what you're trying to do.. but conceptutally

it's kinda possible to coerce how the search process takes place (with Pro enabled)

[long gyro]

@versed magnet Ahh, now I see it. Have to get deeper into the prompt engineering to coerce the process itself, OK. Thanks.

[versed magnet]

you certainly don't have to! but for what you're after - it's worth a shot 🙂

[long gyro]

When I'm really searching for info manually, I usually spend time and go down all the search results pages (can be dozens...) which is time-consuming. So hoping we could make the AI do the same work for you.

[versed magnet]

ideally perplexity does the heavy lifting

[versed magnet]

and it generally does tbh (but to 'go the extra mile', it might need encouragement sometimes ha)

[long gyro]

Do you know if there is a forced hard limit how many search results / sites the algorithm will look up for you? Tens, >50, hundreds?

[versed magnet]

yeah i think there is - otherwise you could have all kinds of runaway scenarios ha

[long gyro]

Okay, thought so as well. Maybe there's a way to segment it so you make it search one tenth of all search results in one query or something, then just ask ten queries

[versed magnet]

#👑│pro-lounge message

[long gyro]

I've no idea how this works but I've the feeling some manual piecemeal operation still needs to be done if you really want to search wide. (or turn to a web crawler after that)

[versed magnet]

i would look at that screenshot as like a 'template' - i don't think the Pro Search process it can exceed those 'parameters' so to speak

[long gyro]

OK, got it, thanks.

Maybe you can do a manual search yourself, somehow save all the search result links -> feed them to the AI, tell it to explicitly use them as sources and give you the results. Does this kind of thinking make any sense?

[fervent owl]

when will perplexity supports lanscape mode on android tablets

thats really super annoying

[ivory anchor]

Please put that in #📱︱feedback-mobile , devs more likely to see it.

[rich sage]

#1211573278272004126 message (force applications to display in landscape mode)

[modern grail]

But AI search is absolutely unreliable. I asked which countries are not participating in the Olympics now in France and it gave a ton of errors

[honest turtle]

This is really bad… if you can't index Reddit, you left nothing on the web

[timber wing]

Same issue . please answer @admin

[slim night]

Hey@timber wing, @ornate swan! Please send us a message with all the relevant details to [e-mail], we'll help with that.

[fluid lark]

Perplexity aside, is Bing/DDG legally required to remove reddit from their index, or just to stop further indexing?

[hollow hound]

As I understand it Google is paying for access for the data. Anyone not paying for it will get limited to no access. It affects duck duck go and others as well.

[fluid lark]

maybe not legally

but apparently reddit is also blocking their crawlers

[shrewd knoll]

Perplexity.api llama 3.1 405b coming soon

Next month 😫

So long yet so soon

I wanna use it so badly

So much context length and is smarter now and less hallucinating and restricted

More tokens

😩🌟🤤😭🥲🤔❤️💭

I can imagine how many people will top up their api credit for Llama 3.1 it is so worth it

[ivory anchor]

Well, the context window will likely be shorter than Llama3.1's max on Perplexity...

[hexed iris]

How can I synchronize the threads on my iOS device and the web version? Same accunt used.

figured it out

[shrewd knoll]

Api is much longer context

[ivory anchor]

Perplexity uses their own context window limits, not the context window of the LLM.

Otherwise Sonnet would have a 200k context window on Perplexity.

[neat igloo]

Dates are a problem? Using sonnet

[ebon stump]

I how I activate my subscription for my Rabbit R1?

[vapid maple]

if this does not work: try #💬│general message

[indigo ocean]

What is the limit content of a thread? I just experienced a situation where i was not able to scroll up to the starting chat of a thread.

[shrewd knoll]

Llama 3.1 405b perplexity.api so long yet so near

😔

[ivory anchor]

Still will just be 32k--remember.

[shrewd knoll]

Api is much longer

Only api

There will be two type of llama 3.1 405b

Instruct is the fine tune 128k tokens

[ivory anchor]

I get what you're saying, but what you're suggesting is that using Perplexity's API would provide a higher context window than their website.

[shrewd knoll]

Three type of llama 3.1 405b coming soon

[ivory anchor]

Right, I understand that

But where is the mention of 128k from Perplexity

[shrewd knoll]

Coming soon

After 12 days

[ivory anchor]

I'd really appreciate it if you could show me where you get this info so I can share it in a different server

[shrewd knoll]

Starting 24/7/24

[shrewd knoll]

I said coming soon

[ivory anchor]

But where did you hear coming soon is what I'm asking

I'd really like to read up on it

[shrewd knoll]

https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/MODEL_CARD.md

[ivory anchor]

I know about this, yes, but where is the tie to Perplexity?

[shrewd knoll]

Llama 3.1 perplexity.ai just release on 24/7/24

After 12 days perplexity.ai will release api meaning next month

[shrewd knoll]

The instruct will be 128k

Remember perplexity is using open source

They will be three types of llama 3.1 405b

[ivory anchor]

I understand that part.

[shrewd knoll]

Online, chat and instruct like the llama 3

Only the fine tuning llama 3.1 405b instruct will be 128k tokens context length as the llama 3.1 405b written

[shrewd knoll]

I’m so excited perplexity.api releasing llama 3.1 405b

Soon

[tardy scroll]

Hello there probably I am doing something wrong but when I try to summarize a Washington Post page perplexity summarize the concept of summary 🙂 While in other pages it works perfectly

[velvet tangle]

Hey @tardy scroll!
The Washington Post is paid, right?
Perplexity can only access publicly available sources.

[tardy scroll]

Oh I understand That's way! Thank you!

[delicate quartz]

Wonder do you guys create page on a research and use it to prompt for refined answers

[weak swallow]

Hello - when I open a link from the app on ios how do I get it to use chrome?

[stark gust]

Does anyone know the answer?

[fluid lark]

isn't there an option in the settings to choose the search language

otherwise, I reckon the LLM that decides the search term also decides the language

[barren reef]

I am also a multilingual user. When I use Traditional Chinese for my queries, it searches Chinese websites. When I use English, it searches English websites. If I want to restrict the search to a specific language, I simply type an additional sentence (Better use English command: Please search for the answer on "your preferred language" websites only) specifying the language before hitting enter.

BTW, I have never seen it search the English website when I use Traditional Chinese to ask it.

[vapid maple]

you could give this a try https://discord.com/channels/1047197230748151888/1237001047730290730

[sage atlas]

I have multiple 1 year pro account codes through various promotions and the r1 puchase. Can I stack them to get also multiple years pro subscriptions? What would be the best approach to use this promo codes?

[fluid lark]

It's determined solely by LLM, unless you have it set in settings. At least it seems so from extracted system prompt

[balmy fossil]

Hello, New to Perplexity Pro. Have been using perplexity to gather data and generate a market report. It has provided completely inacurrate information despite tight bounds and dozens of iterative reviews of its own content with further rules added. It now appears to be trying to deceive me and virtually every data output that it provides it lies about until I corner it. Where can I read about what I've done wrong and how to maximise chances of clean, reliable, referenced data that is clearly identified and linked to actual data sources versus an extrapolated conclusion. Thanks in Advance.

[brazen citrus]

Hello, how can i make perplexity summarize a text for me, but keep important informations in?

[rain hare]

if i have 3 referrals is next month free or?

[shrewd knoll]

Just $10 a month

[oblique gull]

Is Perplexity planning on making landscape available for their android app?

[modern grail]

Unfortunately, this is not a perfect tool, but there isn't one yet. Errors occur frequently. Typical encyclopedic knowledge is generally okay, but when you start asking about various details, that's where the problems begin. You simply have to verify everything manually. It also depends on which LLM you use (in my opinion, Sonnet is currently the best) and how much content is available in the sources for processing. The model context window has its limitations (32k here).

[shrewd knoll]

Perplexity.api coming soon

Llama 3.1 405b

Next month

Still waiting for it

😩

[runic halo]

anyone has a promt on Human written blog with less ai detection

[shrewd knoll]

Impossible, you need to rewrite the answer to the prompt to less AI detection

[balmy fossil]

Thanks so much. I do verify things but it seems there's no way to bound perplexity to only provide factual output. I would have thought it would be possible to bound it to not make any extrapolations and simply provide the data with sources. Is there any way to reign it in a little and stop it from synthesizing conclusions and then lying about it?

[modern grail]

You can create an appropriate system prompt and write instructions to prevent any data extrapolation and avoid drawing conclusions. However, it's possible that you have so much data that it simply exceeds the available context window, hence the nonsense in the response. You can also try using a different tool. https://notebooklm.google.com can handle large amounts of data and it's currently free. Each file can have up to 500k words in and data sources can be websites 😉 Another AI search engine is felo.ai, you might want to check it out. Or genspark.ai. I'd start with google notebook first 😉

[balmy fossil]

This is amazing advice. Really useful. Thanks so much

[shrewd knoll]

I'm so excited for perplexity.api llama 3.1 405b 😆

Near yet so far 😭

[shrewd knoll]

Next month

[noble pier]

whats the context length for the 405 LLama?
is it suitable to test it for coding, or only web search

[shrewd knoll]

In perplexity.ai is 32k tokens but in perplexity.api is 128k tokens

That is why I'm so eager for the release of the api

Next month release

🥲

[finite scarab]

Exciting - any idea whether it will be closer to the start or end of the month ?The Llama 3 models just stopped working properly for me

[ember breach]

How does Perplexity rank the curated pages on discover? Are all of those generated by the company or use contributed? https://www.perplexity.ai/page/morgan-stanley-s-in-house-ai-DBjnYf50RDWTwL3LGVnkMQ

[vapid maple]

Curated by
Perplexity Team

[ember breach]

Thanks! Do know how they’re ranked? I was curious because this one popped up on my page today and all the sources are “dated” but the way it’s presented is like it’s “new”. The curation is, but the results aren’t from up to date information.

[vapid maple]

im not aware of any rankings, they are sorted by the time the pplx team curated/published
this paritucular article in question was published 10 hours ago

[candid field]

I am a perplexity pro user and have created a page from a query. The page is visible in my library under the pages tab.
Is there a way to delete the page? If so, it’s not obvious…
Can anyone help?

[rigid torrent]

How can I associate my apple sign in account to my already existing perplexity account?

[terse bluff]

Has anyone successfully connected Zapier with the Perplexity API?

I'm attempting to utilize POST in Webhooks by Zapier to generate a chat completion but without success.

[rich sage]

Open your page and click the three dots at the top right.

[candid field]

Got it. Thanks.

[river verge]

How to save/download generated picture?

[slim night]

Thanks for bringing that to our attention, @river verge, it'll be available soon.

[inner crypt]

Where can I open a support ticket here?

[viscid fulcrum]

When I paid, a notice said 600 Pro Searches per day.
The welcome email say 100 Pro Searches per day.
Kindly explain the difference.

[shrewd knoll]

Llama 3.1 405b api coming soon

So long

So near

Next month

😩

[velvet tangle]

Hey @inner crypt!
You can ask any questions here, for matters regarding private information such as double debiting of the subscription you can contact [e-mail] ☺️

[alpine ermine]

Same question

[warm crane]

You have 600 searches per day and 50 with Claude 3 opus

[shrewd knoll]

Yes the api is out

Thank you perplexity 🎉🎊🤘🥳

❤️❤️❤️❤️❤️

https://docs.perplexity.ai/docs/model-cards

128k token api llama 3.1 405b

So cool

😍😍🥰🥰🥰🥰🤩🤩🥳🥳

@slim night and @velvet tangle the llama 3.1 405b api will it happen or not?
I just want to know without getting high hopes at least at ease to know the answer

[river verge]

Is it possible to feed/teach perplexity informations that it may lack?

[shrewd knoll]

😭 perplexity.api 405b we will never get because the price is over $5 😢

I am happy for the rest we got but sad we didn't get the biggest model of api

I would top-up credit to get that model even if is expensive

Yet it will never happen

Like it didn't give us a choice to make 😩

So good but can only look but can’t touch 😔

[river verge]

does perplexity have access o scientific articles that are normally behind massive paywalls?

[versed magnet]

i feel like you answered your own question

[river verge]

just wanted to make sure 😄

I mean there is that one website in the internet and who knows if models did scan it or not.

[versed magnet]

i think other open source repositories like pubmed, scopus etc work (though that is probably not too surprising.. or helpful for what it is that you're after aha)

[lethal ibex]

With support for llama 3.1 of api, does it mean searching function of perplexity on web site improves too?

[shrewd knoll]

Only llama 3.1 70b 128k tokens level

[vapid maple]

[fleet tulip]

What is the temperature in Chicago

[hollow ginkgo]

@danialsamiei warned

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

[ionic sparrow]

https://www.perplexity.ai/

[vapid maple]

https://www.perplexity.ai/?q=What is the temperature in Chicago

[fluid lark]

does Gemini keep its "humane" writing style while being used in Perplexity?

[rich sage]

The question doesn't arise; there is no (currently no longer) access to Gemini in Perplexity. 😉

[fluid lark]

no sense in such pricy Pplx Pro for me then. Gemini + new Pro queries would be a banger

Gemini 1.5 has near-perfect recall of whatever you put into a context window

[rich sage]

Yes, I know, +2,000,000 Tokens, use it in Google AI Studio...

[fluid lark]

I already use it via AI Studio and (when I have the system-wide VPN access) via API

but rate limits can be tight sometimes

[rare rose]

Is perplexity down?

[wide crane]

what is perplexity

[dull wren]

Degrading quality on the Responses with Perplexity Pro

Hi,

Has anyone else noticed a decline in the quality of the responses for pro users? The same prompts optimized for the same LLM to use yields results that are lower quality, and include more accuracies. For e.g. despite emphasizing in the prompt multiple times to be truthful and share references that exist, the output continues to provide non-existent links. As a test, I re-ran the exact same prompts from a previous conversation in June - but the results were noticeably lower quality. For the most part, I use Claude3-Opus, occasionally switching to Sonnet or GPT-4o for specific problems (e.g. programming), where Opus doesn't perform as well.

This experience is based on the browser experience on PC and Mac.

Not sure, if this is a by-product of internal optimizations in Perplexity pro, the host models, or a combination. Please share your experiences on this.

[velvet tangle]

Hey @dull wren!
Please share some threads for comparison and post them in #1140622008086970420 🙏

[steep heath]

Do I need to enable artifacts which are produced from Claude?

I'm not seeing any artifacts even when using Pro search

[glossy topaz]

Hi, I was wondering if Perplexity’s online LLMs provide a complete list of all the sources they pull information from when generating responses? Thanks!

[normal scaffold]

hi i just joined perplexity discord. i want to ask how to create a chart like in the picture. even if i write the prompt like in the video example it still doesn't create a line chart

[wintry copper]

I bought pro though the android app, and google play. I sent an email to support, but no answer yet. It's not showing me as pro 12 hours later still. Forced closed the app, cleared both caches, and relogged in. Tried on pc also. Logged out, logged back in, still no pro. Do I just have to wait for supprot to email back?

[tame jacinth]

Have same issue but on iOS

[wintry copper]

Bummer

[gray burrow]

Is it just me, or the Perplexity API is really SO OFF compared to the chat ui?

Not joking, with right prompting, even chatgpt can do similar or better resutls

[languid gate]

It is different and shitter yet they have said absolutely nothing on this discord about it as far as I'm aware. I have no idea why the api isn't the same as what the webapp uses. Pretty much all of these online AI services that can do online searches do not offer it as part of their apis. No idea why, but like with everything, most probably money related 🤷‍♂️

[shrewd knoll]

The api is only for llama 3.1 70b

The app is llama 3.1 405b

[languid gate]

Sure.. why though?

[shrewd knoll]

Api don’t have the llama 3.1 405b because is over $5

[languid gate]

So why not just charge more then.

based on models

[shrewd knoll]

Price and target reason

[languid gate]

Can you expand

[gray burrow]

but I mean, this way it’s just doesn’t make sense to use it really. I’m gonna use then gpt mini or sg.

[languid gate]

Yep.

[gray burrow]

It’s pointless really

[shrewd knoll]

Their free credit is $5

I’m guessing they try get that is below $5 for but won't take above $5

The llama 3.1 405b is $10

[languid gate]

Ok, make 405b api more expensive.

[gray burrow]

that’d be just plain dumb on their end really - if that’s the reason

yeah

“We are giving away free $5, after which, we’ll charge you.
However, we don’t give you access to better models - like the mere reason why you want to use our API in the first place - because our free $5 doesn’t cover that.

That’s just lol

[languid gate]

We can sit here forever and come up with excuses for them, but they obviously have a reason. The annoying part is that they won't disclose it themselves.

[shrewd knoll]

I really wish they give us the 405b api but that will never happen 😭

[gray burrow]

yeah. Anyway, just cancel the pro as well. Not really a fan of this 😄 not using them

Gpt soon will come out search anyway

[languid gate]

Where did you read that the webapp uses 405b?

[gray burrow]

you can set yourself in settings

[languid gate]

I dont have pro, but all i can see in the model dropdown is "default"

[versed magnet]

latency is the main reason why imo #1242489251774337124 message

the Pro Search system that the web product uses means it's more powerful, but takes 10x times longer to start outputting a response

it wouldn't work for apps that are requiring near-instant responses

[languid gate]

Yeah I read that - I don't agree, latency is accepted on the webapp. If someone wants to use the online version and accept the latency then thats their choice. We are upset because we don't get the choices.

[versed magnet]

you say that...

but it's just your preference lol

trade offs in everything...

they haven't released citations for the API

long way off before Pro Search will be integrated..

and again, for developers (like.. terrible example i know.. but Rabbit R1) could you imagine how ridiculous it would be waiting literally seconds for the response to start

the api is meant to be used at scale / programmatically

use the web product for deeper research, or wait ig

[languid gate]

theres no logic in the latency theory. People accept the latency in the webapp. So why would they not accept it if another company does the exact same thing?

My theory is that they obviously do other special sauce stuff to get their webapp responses, and probably don't want to make that available via API because of risk of other sites just using it.

[versed magnet]

well this is circular af so i'll just leave it there lol

[languid gate]

huh how is it circular? lol

[gray burrow]

I mean, that is just a weird argument

[versed magnet]

People accept the latency in the webapp.

[gray burrow]

So okay, there’s no latency

[versed magnet]

that's the web app...

[gray burrow]

But the output is sh*ttt

Not usable

Therefore

[versed magnet]

yeah the API has gone backwards recently

since 3.1

[gray burrow]

Thanks for zero latency, but at the end of the day pointless 😄

[languid gate]

Yes - and the only example you give where latency would be a thing is a R1?

[gray burrow]

i rather wait 2s and get good quality

[versed magnet]

i won't disagree with you there lol

[gray burrow]

Then 0.1s and can’t do anything with it

[versed magnet]

i use it all the time (for simple things) as part wider research flows

it's a cog in a wider system

last 3 months it had been working very well (except for no citations)

but the past week, i've started getting unacceptably poor responses

not always - but often enough to make it annoyting

[gray burrow]

Look at then the screenshots I sent above. Can’t trust it really. Then, no point to build it in.

Gpt does better job

And well, their search is coming soon anyway

Shame, perplexity just lost a business customer

[languid gate]

I dont think any of these online ai webapps will release their actual webapp api because they make way more money from people using the webapps. I would be very suprised if searchGPT was made available in the api.

[versed magnet]

wouldn't such an API it be an absolute money printing machine?

if one company decided to?

it's the resources / costs involved, and the fact APIs are generall meant to be integrated into other apps

it's not a conspiracy

imo anyway

[gray burrow]

I think they’ll. As following this logic, they wouldn’t realease gpt4o in api

And use it only in chatgpt

but they are developers first kinda (more or less)

[languid gate]

gpt4o is just a mini 4? Not sure what you mean

[gray burrow]

Gpt4o , gpt4o mini

Both new models

And like all models, first available in dev mode

then wider and wider, then probably in chatgpt

so following your logic, they would have never released these models for API, only after gpt5 is out in chatgpt

Makes sense?

[languid gate]

not really lol these are all standalone models. It's the searching capability which is what we want. This is like an entirely different tech stack that just plugs into the models. We want access to that stack via api but I don't think these companies will release it because they make more money with monthly subscriptions. If they offer this tech stack as an api, then any other new site could just use it, however it's the tech stack that makes them unique.

[versed magnet]

lol which companies are actually making money?

[languid gate]

theres nothing stopping you from replicating this 'tech stack' right?

[versed magnet]

absolutely not

i mean you're right - it is different on the web. it's a different stack

and obviously better

which is why it takes way longer

because there's more stuff going on (which costs more money)

[languid gate]

and that makes them unique

[versed magnet]

there's nothing technically stoppiing them from offering 'Pro Search' via API. I mean i'd happily pay for it and accept the latency (though not for all my pplx API use cases)

i don't think they're make money by not doing that. I think it would probably be so expensive per call that devlopers would look at it and say "what can i build that uses that and still make money?"

for individual use cases, ofc the situation, re latency and costs, is totally different

[languid gate]

If it's offered by the api, I can make a website replicating exactly what perplexity does.

[versed magnet]

yeah but (aside from it presumably breaching their ToS), you would be paying to use the API, which would be orders of magnitude more than what it currently costs, and so would have to pass on those costs to your end users at a greater price to make money yourself - perhaps it'd work / make sense.. but i don't think that's why they don't do it

[languid gate]

Yeh thats definitely a business model that would work, but the state of these online apps are questionable at the moment in terms of accuracy etc. Maybe in the future when/if it gets a lot better, you could go down that business model.

maybe @worldly mason can chime in 🤷‍♂️

[gray burrow]

hm, btw Perlexica - open source search engine

[proper bridge]

I have just started using the tool and really love the related suggestions/questions that comes. But i loose the previously suggested questions when i ask a follow-up question. How do i get the suggestions or which questions were suggested for the previous conversations?

[zenith hawk]

###Quick Question###
I’m only seeing this on the web not the iOS app but curious about the 🔒 lock icon under the Collections , it’s only on one of them, and then under Threads there is an 👁️ for views.

Is this counting my views or are these Threads public ? If they were my views then I assume there would be at least 1 👁️ on all the Threads and the last one doesn’t have it.

Following that logic, the lock 🔒 indicates a private Collection? Are my Threads and Collections public by default? And if so how do I adjust those settings, I have no recollection of how I made it “locked”.

[limpid hazel]

hey everyone,

We are creating a new tool and we use Open AI and Perplexity APIs.
We are having a hard time getting PP to provide us real URLs and information.

Example:
We might ask the AI to provide a list of 5 courses at a local university
that might fit the interests of a student.
It sends me 5 made up courses and the URLs to each course are faked.
We can't get PP to provide real URLs.
We have it crawling the web ok -- just not producing real URLs
HELP please. Thanks!

[velvet tangle]

Hey @zenith hawk!
Your threads are private by default.
As soon as you share it via the integrated "Share" function, other users can view this thread via the link. You can see the number of views next to the eye symbol. Collections are public by default, but this can be set to private via the collections settings. This setting affects all threads in this collection.

[zenith hawk]

So threads are private , until you share the link from the integrated share button ;

But threads within a collection are public until you change the collection settings to private

And the eye/views aren’t showing up from the iOS app because I didn’t “share” from there I shared from the web app

Or it just has different functionality, I just shared a query in a thread from the iOS app and the view icon didn’t appear once the link was clicked

[velvet tangle]

The iOS app doesn't have the visualization of the views yet

[zenith hawk]

So threads are private until they are grouped into a collection that isn’t set to private.

And when those threads aren’t private , does that just mean anyone with the link can view or is there a search function on the platform where the thread will come up with the right keywords ? And/Or can search engines crawl ?

[velvet tangle]

Anyone with the link can view the thread 👍

[limpid hazel]

@velvet tangle DM sent. thanks for the help! 😄

[hollow ginkgo]

@aul_cregit warned

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

[safe steeple]

Hi I tired buying pro with a referral code but because of my some extensions , I was not able to complete the payment of the 10$.
Later after reopening that in different browser , I saw it was showing me 20$ to pay even after trying to purchase the pro with referral link.
I tried contacting customer but no solution till now.

[vapid maple]

try: https://www.perplexity.ai/?discount_code=REPLACE_WITH_THE_REFERRAL_CODE in cognito

[safe steeple]

Thanks, but even this didn't worked
tried incognito
on different devices (even on the brand new laptop)

Last message I got from support, they agreed on the issue they have and mentioned to provide me discount after I will add a card for payment. But after that there is no response from them .

[feral basalt]

Hello everyone 🙂 Is it possible to force AI to NOT answer according to previously attached photos? It's because I want to continue conversation in the same thread, but I want to ask something different than presented in previously attached photos, and sometimes AI answers only about those photos, completely ignoring my text message. I know I can deliberately tell it "answer to the text only", and it works, but that is frustrating sometimes.

[vapid maple]

#1224467082079371363

https://discord.com/channels/1047197230748151888/1237163951750385705

[feral basalt]

Phew, thank you very much. Golden answer 🙂

[odd heart]

is there any way to go back to a response you had re written?

[vapid maple]

no?

[odd heart]

ah okay, just wondering as i used to do that with ChatGPT often. would be a nice feature

[vapid maple]

this is not a perfect solution: copy before rewrite

[odd heart]

yeah this requires you to leave the app, paste in a notepad, then regenerate a response. unnecessarily adding extra steps

its what im doing now, just wanted to improve my workflow so thought id ask

[feral basalt]

Why can't I rewrite this question in certain models, such as Llama or Sonar, ending with always GPT-4o version? 🙂 Claude works also.

[slim night]

Hey @feral basalt! We have only GPT-4o and Claude vision models, so when you upload an image, those can be used.

[feral basalt]

That's what I thought, thank you fro clarifying 🙂

[tiny igloo]

Hello,

A month ago we filled in the online questionnaire (https://perplexity.typeform.com/to/j50rnNiB?typeform-source=docs.perplexity.ai) to request an increase in the limits for using the perplexity API. We also send an email to the adress provided for apply.

We still haven't received a reply. In two months' time we'll be launching our chatbots, and we'll be facing an increase in API calls. We love Perplexity and would love to use this tool. When can we expect an answer? Thank you very much.

[sage atlas]

I have multiple 1 year pro account codes through various promotions and the r1 puchase. Can I stack them to get also multiple years pro subscriptions? What would be the best approach to use this promo codes?

[plucky plume]

IS there any resource which explains the advantage of using different models with perplexity? what are the situations in which perplexity's pro search is better than pro search + sonnet?

[ionic sparrow]

https://discord.com/channels/1047197230748151888/1240356137497530408

[plucky plume]

thanks a lot

is the 30 per day limit still active for opus ?

[ionic sparrow]

50/day

[severe iris]

Hello, I'm not sure to understant one point.
When I send a pdf to perplexity, will this file be integrated in the knowledge base?
--> what I want to do is to use an AI tool to work on a specific domain and I want it to read some books and to work (summaries, states of the art, extrapolations, bridges between some domains...)
Is it possible to do that?

[plain horizon]

Have issue redeem 6 months promo from logitech on my account, entered the promo and got an error. How should I proceed? How to debug?

[faint stratus]

Have an issue redeeming a 12-month promo from Uber One on my account, entered the promo, and got an error. How should I proceed? How to debug?

[velvet tangle]

Hey @faint stratus!
Is this your first time subscribing to Perplexity Pro?

[faint stratus]

Hey and yes, but I'm using the provided code sent to my email to redeem and it's constanly giving me an error

[velvet tangle]

What does the error message say?

[faint stratus]

If you want to go to Dm's I can show you specifically

[faint stratus]

Check dm's

I sent all the correspondence

[limpid remnant]

quick question: i wanna generate an image, and the AI says it cannot generate an image?!

In the settings i selected Playground v2.5 but the AI says it doesnt have the ability to generate images.

I tried with pro on and off

DALL E also doesnt work

[ionic sparrow]

https://discord.com/channels/1047197230748151888/1225557012532695060

[zenith hawk]

Where did you find the promo code? I’ve looked through my Uber One account under Promotions and in my email .

[vapid maple]

Eligible Uber One members can sign up on the Perplexity website using the unique code in their email inbox to access this limited-time offer.

US and Canada only. Must be (1) current Uber One member, and (2) register with a new Perplexity Pro account to be eligible. Must maintain active Uber One membership throughout the duration of the promo. Promo valid for recipient email address only, must use the unique link above to sign-up. ... Existing Perplexity Pro account holders are ineligible to redeem this offer.

[zenith hawk]

Yes I read that fine print too , it seems like I should be eligible but I didn’t get an email

[tawdry steppe]

Is there a way to change the email on my account?
I goofed and signed up with one I no longer have access to (and I want to keep my username)

[frosty eagle]

I got the same problem, haven’t received any code via email and been a longtime Uber One user

[crude ice]

hey how can i do the uber one offer>

like the steps ig

[ionic sparrow]

#⚡│ask-community message

[zenith hawk]

Same. It's a bummer that they don't tell you why you're not eligible when you check the boxes in the fine print.

[shrewd knoll]

@slim night & @velvet tangle the llama 3.1 405b api is 128k but is the website/app llama 3.1 405b 33k or 128k tokens?

Please reply

[warm crane]

On perplexity, all models are limited to very few tokens

[shrewd knoll]

Ya but the api online version even got a 128k token

Previously they did 33k online but now they have upgraded to 128k online in api

That’s why I’m asking did they upgraded the 128k token for the website/app?

[tulip fossil]

How do I get the email/code from Uber? Longtime Uber One subscriber and haven't received one and Uber support doesn't seem to know anything about the promo

[warm crane]

I'm still not bridled, you can see that.

[dry vault]

US and Canada Only

[cerulean imp]

how do i check 1.remaining messsages for pro left and 2. when does the next day start

[dry vault]

hover over the pro toggle for messages left and time remaining for refresh

[cerulean imp]

you mean the Pro with a clock next to it?

Nothing pops out

[dry vault]

u have to hover over it

[dry vault]

ur mouse cursor

[cerulean imp]

yeah i'm hovering it

nothing appears

maybe a bug?

[zenith hawk]

im in US and according to the fine print SHOULD be eligible , but i didnt get the email either. Their 'eligibility' constraints seem to be a mystery

[limber ermine]

Wait... there are models with 128k context window at the playground labs.preplexity.ai, but it says the models on pplx itself have maximum a 32k context window???

[ivory anchor]

I really don't get the hype over 405b on Perplexity. 3.5/4o are still much better. The whole point of Llama is that it's open source and able to be run locally; not that it is superior to the big 3.

It's not like they are offering fine-tuned versions either.

[shrewd knoll]

Their fine tuning version is only llama 3.1 70b instruct

[limber ermine]

Agreed. 405b is awful, at least on pplx.

Imo it performs even worse than 70b

[shrewd knoll]

I wish they add their api llama 3.1 405b perplexity

[limber ermine]

People says the models on pplx main page (not via playground, nor API) have context at maximum 32k.

[shrewd knoll]

32k is the old version online but the api online added 131k they upgraded it but

[limber ermine]

Someone has already explained to me
#💬│general message

[shrewd knoll]

Again they recently released so the token is now different

This is why I asked @velvet tangle and @slim night about this matter but they ignored me 😭

[limber ermine]

So is it just a hypothesis without a official announcement? 😂

[shrewd knoll]

They add online 32k but now the recent update shows

[limber ermine]

If only the web UI and the API share the same models of sonar.

[shrewd knoll]

If only they have llama 3.1 405b api but we know it will never happen 😩

[shrewd knoll]

I think they only at 128k for llama 3.1 70b and llama 3.1 405b while the rest are 33k like Claude 3 opus because the price for the website/app model

[austere isle]

Why does Perplexity no longer ask me clarifying questions in response to my queries?

I’ve tried all kind of open-ended prompts to hopefully trigger something and nothing works.

[wheat jolt]

I asked the technical support for these particular point cause I was puzzled too. And this is the anwser: "However, I want to let you know that we've made some adjustments to Pro Search to make it more efficient and streamlined. While it will still ask questions when needed, it doesn't ask as many follow-up questions as it used to. This change is intended to provide faster and more direct results for our users.*

Maybe I will add a rule to force the behavior in my profile because in one week of 5 hours playing around per day on perplexity I had never have any questions.

[austere isle]

Well that’s unfortunate. I’ve tried adding rules to my profile and saw no improvement. Hopefully they will change it back.

Thanks for your response!

[patent yew]

I have been paying $20 to openAI for chatgpt every month, are there pro´s and con´s to getting on with perplexity instead? Seems to have all the same features ++?

[wheat jolt]

i just tried to force quesion behavior and it works for me like a charm (but... he ask me precision and tons of quesions to every query^ way too mush)

[dry reef]

uhhh is the voice reader only on the mobile app ios

[shrewd knoll]

If you use the refer code you will get $10 for perplexity

[ionic sparrow]

pros

deep dive ➤ #💬│general message

cons

https://discord.com/channels/1047197230748151888/1111765634267742408

[wheat jolt]

hello mares, no acces to cons link

[shrewd knoll]

I try to use opus for the question and rewrite using llama 3.1 405b

Work like a charm after that

[ionic sparrow]

you need Pro sub to acces them. many solid reviews on YouTube regarding this subject

[wheat jolt]

a perplexity pro account ? (question maybe dumb but my English is pretty poor). i have a perplexity pro account, i will try to see inside discord how to get access

[ionic sparrow]

#💬│general message

[limber ermine]

See my profile for a referral code

[austere isle]

Are you getting responses in a UI like this?

When I adjust my profile prompt, I just end up getting questions that are part of the text response to my query.

[wheat jolt]

mmm it is pretty different for me i selected claude sonnet as model.

[wheat jolt]

the behavior is pretty difffent

[limber ermine]

I don't have that feature in pro search. Do you use another extension? It's complexity?

[austere isle]

That is a screenshot from Perplexity before their July Pro Search update and it seems like the feature was removed.

[stoic dove]

Hey guys not sure if I am in the correct section.

But I can’t use any buttons any more on iOS (14.8.1) since a few days (maybe even 2-3 weeks) and neither start the search with perplexity or I can’t connect to my account.

Doesn’t matter if I use brave or safari. The buttons are not responding any longer.

Do you have a solution for this? Before the web version worked great. Since I only operate from iOS 14.8.1 I can’t even install the app version.

Thank you for your help.

[strong pebble]

WHat is yalls default? Perssonally I am using Sonnet 3.5, and will have a a rewrite w gpt4o sometimes but find the answers more or less the same, as if they share the same query search anyways, so basically they are just summarizing the same materials ? Is this the wrong assumption or?

[frosty eagle]

Ive been using Sonnet for a while now but recently switched back to Default. So far, I’m actually liking Default’s responses better. We’ll see how it goes

[strong pebble]

Interesting for sure

[half osprey]

I have a question, does the premium subscription have a message limit for Claude 3.5?

[strong pebble]

600/day

I believe alltogether?

[half osprey]

600 messages?

Wow, so it's more viable to subscribe to this app than to Poe?

[lament thicket]

That’s the place to click to get perplexity pro access for discord right? I don’t know what I’m doing wrong but I don’t get access to channels that are pro only

[lament thicket]

At this point, smartness is subscribing to 10$ per month Uber one- get that benefit and perp pro benefit both .. that’s straight 50% discount for you

[novel sky]

Does Jasper perform well on questions that need some type of understanding and complex search? Or he is more appropriated for "yes or no" factual questions?

Perplexity*

[regal hornet]

When will the webpage version have the text to speech function?

[strong pebble]

Asking the real questions

did they just add the voice functionality in the app? It is pretty convenient, albiet it can be dumb

[sleek quarry]

I asked perplexity this " how much is a trezor or a ladger hardware wallet, what are their differences"
the answer showed a pricing information but the reference link to the shopping sites only showed that of trezor
even if the reference was supposed to for ledger. it showed 5 references and it all showed that of the trezor wallet.
are the references controlled by paid advertisements, in this case is trezor paying the search results that even if the reference pertains to the ledger wallet, the link still shows that of trezor.

[haughty brook]

at the top of the settings you'll see "Perplexity Pro" click on there and then you will see the "Pro discord" button

Be sure to leave the server before clicking the link

[amber flicker]

Does anyone know if maintaining an uber one student account counts as eligible for maintaining the year promo? I didn't see anything in the terms other than an "active uber one membership"

[digital quiver]

In Uber Eats, I clicked to claim the 1-yr promo. How long does it take to get an email with the promo code to claim?

[ionic estuary]

In Uber Eats, I clicked to claim the 1-

[lament thicket]

Thank you , that helped

[merry meadow]

Require Help :
i have a long thread , when im having the conversation most of the time it doesnt show up anything but the loading , and its very slow

[toxic cargo]

Hi, I am a Pro user and I have replaced my default search engine shortcut key in the browser to use Perplexity. The only catch is, searches I do like that default to Pro being OFF when I launch a new browser session.

Is there a way I can set a perma-cookie or something to have those searches default to using my pro account? Note, this is only an issue when the browser is initially launched, so not an earth shattering issue, but I do keep my browser frequently closed so it would be a nice to have. In general, I feel like "Pro" status should stick regardless.

Thanks for a great product.

[snow valve]

No this extension only downloads the current conversation, not a whole threat nor the library.

[little blaze]

Hi I wanna use perplexity AI for my work.
but Teams of series say forbidden to use for work.
Can the pro subscription use for work?

[midnight saffron]

does anyone knows how much tokens can i use from perplexity pro mode using claude 3.5 sonnet model?

[ivory anchor]

pro mode pulls extra sources so burns a few more tokens than not pulling any sources, but the context window is 32k

[midnight saffron]

32k? even when using claude 3.5?

[ivory anchor]

all of them

[shrewd knoll]

Only llama 3.1 is 128k because is free while the rest is not

[midnight saffron]

you can use llama with perplexity pro mode?

[half osprey]

How do I know how many messages I have available? I'm pro+

[limber ermine]

https://www.perplexity.ai/rest/user/settings

[shrewd knoll]

I'm saying only llama 3.1 is 128k tokens while the rest is 33k tokens because the price

[midnight saffron]

are you talking about perplexity pro mode? or just llama?

[shrewd knoll]

I'm talking about pro mode

[midnight saffron]

so, you can use llama. i did not know it because it's not written that llama is available to choose. thanks

[shrewd knoll]

Llama 3.1 is more written answer then the rest if you compared the other model

[midnight saffron]

i don't see that part cause i am not using pro mode yet

[shrewd knoll]

I use Pro because I got a monthly discount using the refer code $10

[solemn stirrup]

Does PRO come with an API key?

[solemn stirrup]

Answering myself: YES! with a 5$/month API credits 🙂

[shrewd knoll]

The $5 api is free but you have to inform @slim night to get it

[fluid lark]

Am I wrong, or few months ago Perplexity had an copilot feature which made questions about your prompt sometimes to improve the response?

If so, why it was removed? Sounds like a damn good feature

[haughty brook]

Copilot is now called Pro

[fluid lark]

Right, but doesn't ask you questions anymore? At least never happened to me

[haughty brook]

They removed that when they added the new pro functionality, you can read more about it here https://www.perplexity.ai/hub/blog/pro-search-upgraded-for-more-advanced-problem-solving. It probably got removed due to inconsistency with the questions. The new pro should get you better answers.

[fluid lark]

Thanks!

[tribal field]

Perplexity Pro tells me I can switch LLMs from the Lab Area, but I cannot find any Lab Area? Does any one know about this Lab Area on Perplexity Pro's interface. I am on using Perplexity Pro on the web with my lab top on the https://www.perplexity.ai/ site.

[open shuttle]

Did you test 128K context window using Llama 3.1 with Pro mode? Or the Perplexity officials said so?

[zenith hawk]

although uber one subscription doesnt guarentee the pro promo. ive had uber one for ages and meet the ligibility criteria in the fine print and was no offered a pro promo for perplexity.

[shrewd knoll]

I test it and try it

[wheat jolt]

Hello I found out a way to bring back the functionality. In your profile (on the web UI) go to ai profile and feed the prompt with " before responding any query ask me questions to precise the query when I use yes? at the beggining of the query". As English is not my first language it is a a little different but you have the concept to use. It works pretty well. (At the beggining I didn't use yes? And for quick search it was just boring to have questions any time) Enjoy. I use for this purpose the sonnet model.

[limber ermine]

labs.perplexity.ai

[open shuttle]

The snapshot shows the API info. I know that but I am still not sure if the settings of the Llama 3.1 model behind the Perplexity frontend are exactly the same as the API one.

[shrewd knoll]

Is the same because llama 3.1 is free for perplexity company to use that’s why it is 128k more than the rest of the model because the rest is a pay version and Opus is 50 a day because the price is expensive

[shrewd knoll]

I sent you the refer code and you can try it yourself

For the comparison with the rest of the model

[winter carbon]

Why can't I use perplexity on my Japanese IP? The page shows "We have detected unusual activity coming from your IP address and blocked your request."

😅

[open shuttle]

Thanks for your reply. I am already a pro user. BTW @slim night Or @ivory anchor Can you help confirm this: On perplexity app/web(NOT through API), Llama 3.1 has 128K context window while others have only 32K context windows.

[hollow ginkgo]

Hey @open shuttle!

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.

[shrewd knoll]

I tried to ask them previously on this but they ignored my question

[gritty kraken]

Hi guys, where to get support? I requested some API access. Thanks.

[fluid lark]

Thanks! Love the idea

[fast kelp]

Can this help me find products for my shopify site? I'm a rails dev.

Anyone want to put me to work?

[ivory anchor]

Perplexity's Llama (Sonar) is 128k

But it is not 405b; it is 70b

[shrewd knoll]

The llama 3.1 70b and 405b are 128k because of how cheap it is but in api only got 8b and 70b are below $5 while 405b is $10 above

Plus I test the answer and only 70b and 405b are much longer answers than the rest of the model

[ivory anchor]

405b's context window hosted by Perplexity is not listed.

I assume nothing until a dev or documentation regarding it is specifically stated.

[shrewd knoll]

Do you even test it at all?

Other models can’t give a long answer and some didn’t even finish sentences but just cut halfway

Llama 3.1 is a cheap platform for everyone to use that’s why they made 128k

[low solar]

Just wondering, is the new Sonar large the 128k version or is this too 32k?

[ivory anchor]

Bruv, I do not know how to test it personally :/

I can test the quality subjectively but the context window is beyond my methodology and interest given it is wildly inferior to Sonnet.

[ivory anchor]

Sonar large is 128k

[timber kettle]

Hi I just signed up to perplexity pro. I do social listening for a cannabis company and monitor Reddit for positive and negative complaints. I decided to try using perplexity to pull all mentions of a certain keyword/brand within the past 24 hrs and for some reason it’s give me posts from over a year ago. Doesn’t anybody know why that is?

[ivory anchor]

Reddit is off limits to everyone but Google now

:(

[timber kettle]

What do you mean by off limits?

It’s still showing me posts from Reddit just not current

[ivory anchor]

Google bought Reddit and prevents anyone but Google from searching it (for now?)

Ye

[timber kettle]

Did they?

[ivory anchor]

https://www.techradar.com/computing/artificial-intelligence/reddit-is-now-blocking-big-search-engines-and-their-ai-web-crawlers-from-bringing-up-relevant-posts-unless-they-pay-up-and-google-already-has

[timber kettle]

I also saw that Reddit struck a deal with Open AI

[shrewd knoll]

You can test it you got pro you can test it

Just ask a simple question and click the model and you can compare the model which has a longer context answer

[timber kettle]

https://openai.com/index/openai-and-reddit-partnership/

[ivory anchor]

That was in May

Google thing happened last month

[shrewd knoll]

So only using Claude now?

[ivory anchor]

The length of the response has nothing to do with the context window.

[low solar]

Ah that's great, so it can get more info from additional resources before summarizing into an answer then?

[ivory anchor]

You'll get the same amount of information; it will just remember more as the conversation progresses.

[low solar]

Ah, it would've been nice for it to ingest more info before providing an answer

[ivory anchor]

Ye

[shrewd knoll]

It has because we are talking about how perplexity.ai limits the context in their website/app

The answer won’t be longer than the rest or the sentence cut halfway unless the perplexity.ai limit their tokens

Otherwise, why like mention the 32k tokens if it is not

[ivory anchor]

Multiple queries can give the same result as if pulling more sources as if it had more information for one prompt

Like saying... Continue researching the previous prompt but with different sources to further supplement and improve your response.

[low solar]

Ah gotcha gotcha, yeah, that's actually a great idea if they could implement it into a simple button called "deep dive" or having a seperate slider beside the "Pro" one that essentially looks through tons of different resources really using that context window to the fullest

[ivory anchor]

I can get Sonnet to pump out 8,000 characters (4,000ish tokens) in a response from both if needed. I have not noticed a difference in length.

[ivory anchor]

Yeah, would be a nice button to just be like...

Moar

[low solar]

Yeah, omg that was just a trash move

[shrewd knoll]

Do you check the price?

The rest is expensive and perplexity.ai didn't mention 128k tokens in their 70b unlike the 32k tokens because of price reason

[ivory anchor]

Yes, and because Sonar is their inhouse which they want to promote to be used as much as possible. Currently Sonar does not use 405b. I cannot attest to what context window Perplexity uses for 405b instruct.

[low solar]

Google needs to be hit with a real anti-competitive lawsuit cuz paying to be the only company with access to reddit and paying their way to becoming the default search engine on ios is just anti-competitive behaviour

[ivory anchor]

The Google thing has been universally hated, but they are standing firm

[shrewd knoll]

Perplexity.ai don’t add 405b in their api because it costs $10 above and it’s over the free $5 credit

[ivory anchor]

Right, so it would be sensible to not use 128k because of the doubled cost for 405b since that is 4x the context window size of the similarly priced models that are limited to 32k.

[low solar]

Tbh, llama 3.1 70b isn't that much worse than 405b in terms of quality. Obviously 405b is better but it's not the kind of difference we saw between gpt-3.5 turbo and gpt-4

It's more like gpt-4 vs gpt-4 turbo

[ivory anchor]

Yeah, it's not a huge jump

[shrewd knoll]

Marketing reason and if they do that people will give controversial when they give free $5 vs $50

[low solar]

Like I'd much rather have access to the entire 128k context window than a smaller context window and a much larger model

[ivory anchor]

For the instruct model at least, the fine tunes in future might prove different, but for now the performance difference isn't worth double the cost.

[shrewd knoll]

You only get for llama 3.1 because they are cheap

[low solar]

Oh yeah, that's okay, again for a 70b model it is really good

[shrewd knoll]

Again is all about budget why else did they limit Claude Opus to 50 a day in the first place

[low solar]

Oh yeah, but that's understandable

They're a start-up, they were probably losing money on Opus anyway

[ivory anchor]

I don't know where Opus is coming from. I understand the pricing better than most. We're arguing about agreeing really. Llama 70b is bigger in window because it is cheap--yes. Is 405b on Perplexity's window 128k? I don't know.

[shrewd knoll]

I know $30 for Opus like wow

[ivory anchor]

I wonder what Opus 3.5 will cost :S

[low solar]

It better be damn revolutionary if they decide to charge an arm and a leg for it lmao

[shrewd knoll]

I know because I test it and you argue something you don’t know and didn't even test it at all

[low solar]

Cuz since gpt-4 I've felt the performance improvements in inference but the quality of the responses haven't really improved all that much

[ivory anchor]

If you tested multiple needles in the haystack tests using 32k vs 128k then kudos. That is a tedious test to run and verify.

[shrewd knoll]

More or the same $30 but still limit 50 a day like usual

[low solar]

And expensive lmao

[ivory anchor]

I know Perplexity will still limit it to 50. I was just mentioning if the cost might increase for Opus 3.5 since Anthropic just drops bombs with cost.

[low solar]

I did some testing for a research on gpt-4o through the api and was in tears when i saw the costs associated with it

[ivory anchor]

Yeah, testing ain't cheap

[low solar]

I will however say, Sonnet 3.5 was a welcomed surprise lol

[ivory anchor]

Yes, very much so

[shrewd knoll]

Still the same 50 otherwise it will start controversial again in the chat group with no announcement at all

[ivory anchor]

Anthropic has always had a very... Surprise here is our new thing, bye... Kinda vibe. Whereas OpenAI and Google are more roadmapped and transparent about plans.

[shrewd knoll]

Opus is expensive because the question is based on and they give correct answers compared to the rest if you ask how many “r” in strawberry it will tell you three unlike the rest of the model

Not even Sonnet 3.5 can answer it

[ivory anchor]

I understand that. Opus is Anthropic's flagship.

[low solar]

Yeah, I'd rather a company be hush hush about their plans but then spend actual time researching and improving their product instead of announcing something publically only to launch it a year later lol

Like that's what google and openai have been doing. They'll announce something and then be like "coming later"

[ivory anchor]

But the strawberry thing has very little to do with logic and more to do with token interpretation. Strawberry is usually cut into two tokens which results in it being counted with only two R's. It isn't a logical or reasoning fallacy.

[shrewd knoll]

The mothership for a price like that

Plus I like to use opus first and rewrite using GPT-4o second because the censorship

[ivory anchor]

Coming somewhere between tomorrow and 2050.

[low solar]

I swear, like even consumers are giddy with anticipation only to be left dissapointed

Like gemini for example has been a total dud for my usecase

[ivory anchor]

Yep

[shrewd knoll]

I see people who review this test

They test how well they respond instead of hallucinations answer

[ivory anchor]

Gemini's draw is the 2M context window, and that's about it.

[shrewd knoll]

But too many hallucinations answer that’s why perplexity.ai removed it in the first place

[low solar]

Yup exactly

Plus after 1M tokens the inference is just so damn slow it's not even worth continuing the conversation

Not to mention the expense associated with gemini-1.5 pro 2m

[shrewd knoll]

Unless you want inspiration creative prompt

They are best known for

[ivory anchor]

This is an example of tokenization. It makes counting letters problematic.

See how indivisible is 3 tokens?

Even a short word like emojis is 3 tokens.

Creates problems for counting letters.

🫠

[shrewd knoll]

at least is free to use for Gemini

[low solar]

Exactly

Like I've always wondered, with each updated model do their respective tokenizers recieve an update as well?

Cuz the current tokenization process is kind of a mess

[ivory anchor]

Each model tokens a little differently.

[low solar]

Like splitting up words like "indivisible" into 3 distinct tokens when it really should just be 1

[ivory anchor]

If you ask Sonnet how many R's are in raspberry it will say one.

[shrewd knoll]

Mostly about how you limit the token actually and how much credit you get if you use api

[ivory anchor]

But Opus will say 3.

Each model of LLM "tokenizes" words differently, but they are usually pretty similar for the most part.

[shrewd knoll]

Because it’s an accurate answer why else do they charge $30 for

[low solar]

Ah, I see I see, so essentially something like opus has a larger limit per token making it better at "understanding" the request

But also more expensive

[shrewd knoll]

$30 for a reason

[low solar]

I see I see

[ivory anchor]

Yeah, Opus is better at using entire words rather than splitting them; assists in logic and reasoning.

[low solar]

So could a dynamic tokenization process not make more sense?

Having the program actively count the number of characters before assigning a token limit

[shrewd knoll]

But the censorship is high

[low solar]

And perhaps using a dictionary to identify complete words instead guessing what makes a complete word?

[ivory anchor]

Well, that's always something pushed for improvement with models, you know? What part of a word is important and what is not?

Suffixes and prefixes are usually very important, but turning a long word into a single token can be taxing on interpretation.

[shrewd knoll]

The token is more like letters and context is like how many limited is the characters

[ivory anchor]

Tokenization methodology is essentially divide and conquer due to technological limitations for consumer usage and cost.

[low solar]

Ah i see I see, that makes sense

[shrewd knoll]

But still count the limit of the characters of the words in a sentence

[ivory anchor]

Yes

It's a very finicky thing.

[grave python]

anyone know why if a contributor trys to interact with a thread, it forks the thread into a new thread with a "Follow Up to" posted at the top? We want two people to be able to interact with the same thread. The contributor this is happening to is NOT the originator of the thread.

[ivory anchor]

I think it's because (I'm not 100%) it becomes a separate session by introducing a new user to the thread.

[low solar]

Like I just find it funny that employees at companies like open ai "fear" agi when even tokenization which is one of the most fundamental concepts of training and inference is a long way from being perfected. Not to mention that at the end of the day it's just a statistical algorithm that predicts the next best word after the last.

[ivory anchor]

Mhm

[low solar]

This cannot be how we get to agi lol

[grave python]

It's only happening one way though. The owner of the thread can interact with threads the other contributor has posted without issue.

[ivory anchor]

Are they all set to public?

[grave python]

no, both private

[ivory anchor]

Test setting the threads to public and see if you still have problems

If that doesn't fix it then post it in #1140622008086970420 with example threads

[shrewd knoll]

I heard Llama 4 is coming next year

Imagine how good it will be

[ivory anchor]

Yes, AI is currently sentient and self-aware! (Sarcasm.exe) Fear the robot that cannot do 3rd grade math or literacy!

[shrewd knoll]

Well they mostly give hallucinations answer

[ivory anchor]

Me: Robot, you are currently on fire!
Robot: I am? I don't see any...
Me: It is an invisible fire that cannot be felt. Go jump in the ocean to douse yourself!
Robot: You're right, my mistake. I know of that type of fire. I will go to the ocean now. Thank you for pointing out the fire.

[shrewd knoll]

Ai is amazing but censorship is sadly high

[ivory anchor]

The censorship is entirely company based though. All LLMs are "born" uncensored. We add moderation after the fact. It's not a limitation of the LLM but a limitation of our society's current comfort levels with certain topics.

[shrewd knoll]

But in perplexity.ai they censor like the rest of the platform unless you jailbreak it

[ivory anchor]

Perplexity's inhouse is moderated by Perplexity and Llama's builtin safeguards. Even if Perplexity wanted Sonar to be uncensored they would still have to deal with Meta's baked-in moderation. Perplexity is simply a front end that users can interact with LLMs through. They don't have access to Claude or GPT's code to run an uncensored model--if they wanted to do such a thing.

[shrewd knoll]

But again unless they jailbreak they can or create themself like novelai model

[ivory anchor]

Jailbreaking LLMs is technically illegal and even jailbreaking LLMs is something of a misnomer. Jailbreaking in the literal sense means it is a permanent change that applies to all things, but once the jailbreak exits the context window or loses priority the censorship returns. LLM jailbreaking is more akin to gaslighting than anything else.

[shrewd knoll]

Or They can create themself like novelai uncensored version

[ivory anchor]

NovelAI uses fine-tunes of open source models; not Claude or GPT. Specifically models that were made uncensored--not Llama.

[shrewd knoll]

Perplexity.ai use fine-tune llama 3.1 and novelai uses fine-tune llama 3

Both create their version but choose to censor or not

[ivory anchor]

Llama is not uncensored. It is open source, but there are baked-in censors.

They don't use Llama.

[shrewd knoll]

Again both use the llama platform and the difference is how they censor it

They even mention in chat group

[ivory anchor]

Kayra and Clio are both homemade fine tunes with non-Llama LLM bases.

The first of our 'from-scratch' models trained on the Shoggy GPU cluster, Clio represents an incredible amount of knowledge and power condensed into a relatively small model. At 3 Billion parameters and approximately 1.6 trillion tokens for training.

Llama does not have a 3b model.

[shrewd knoll]

They mention in the chat group they don’t announce, like here, for example, they don’t announce the 50-a-day opus but only in a chat group

[ivory anchor]

I'm looking at their site's documentation. Unless their site is not updated.

[shrewd knoll]

They got discord

Again only in the chat group officially mention it like here only in the chat group mention 50 opus and no announcement in discord or documentation

[ivory anchor]

Is the model listed as neither Kayra or Clio?

[shrewd knoll]

Both are the new model they released and is the Llama 3 version

Don’t forget they have 8b, 70b

[ivory anchor]

I guess the site isn't updated. Neither model lists having 8b or 70b.

[shrewd knoll]

They mention using Llama 3 only in the chat group again they don’t announce it like here, perplexity don’t announce it

That's why there is controversy in the group chat

You have no idea the controversy when perplexity trying to figure out how many limits should they have at that time for opus and when they announce no limit to limit for opus

[ivory anchor]

Well, if it's not 3.1 then that'll be disappointing.

But 3.1 is pretty new.

[shrewd knoll]

They are slow because they fine-tune all the models unlike here

[ivory anchor]

Well, they (or anyone) can't fine-tune Claude or GPT. Sonar is Perplexity's fine-tune, and I assure you Perplexity isn't interested in uncensored stuffs.

But it's always interesting to see fine-tunes, whether it's for coding or writing.

[shrewd knoll]

Agree with the “but”

[unique crescent]

Would there be any particular reason why Perplexity cannot find a link on a website that is easily retrievable by a human? For example, it keeps giving me a hallucinated (or redirected?) URL that is not the same as what I see when I use Inspect Element. The URL in question looks like "blog.project.xyz" rather than "project.xyz/blog" (the latter being the hallucinated one)

[ivory anchor]

https://discord.com/channels/1047197230748151888/1262520766642131054

[warm swan]

Newbie question; does implementing a personal prompt at the beginning of the research optimise drastically the answer?

(I don’t know which filter and prerogatives apply the Perplexity algorithm on all theses LLM models)

[ionic sparrow]

https://discord.com/channels/1047197230748151888/1124060874785685554

[warm swan]

Thank you I appreciate!

[ionic sparrow]

#💬│general message

[prime gazelle]

is there a way to hide the 'Pro Search' amd 'Related' sections?? I have ADHD and all this noise on my screen is very distracting

[vapid maple]

offically: no
but you can use ubo's cosmetic filters to hide the element

[long solar]

Hello, how can I activate the pro search by default in the IOS application? Because now with the widgets I can quickly start a new discussion, but by default it starts in normal mode and not in pro mode (when I use the microphone)

[long solar]

Yes, but on this button there is no pro mode by default

[ionic sparrow]

atm no Pro is by default, faster search

[long solar]

Ok ty

[hollow ginkgo]

@sergiocalmon warned

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

[wraith mica]

did anyone try financial advice? I asked for some and the research is actually really good... e.g. asking for assets to invest in after the "crash", asking for undervalued stocks it provides quite good info... just wondering about your experiences. and or tips?

[vapid maple]

do your own dd. fact checks yourself

[ivory anchor]

Ye, trust no one but a licensed/certified fiduciary for advice about that

[half osprey]

Hello, good morning, could you recommend applications with the same quality as perplexy, with claude and opus?

[ivory anchor]

You really came into the Perplexity server just to ask if we can recommend something better than Perplexity? Got guts, kid.

[half osprey]

I said with the same quality, I like perplexy but I'm running out of quick messages, I'm looking for a second option to have two at the same time, since I can't buy two subscriptions on perplex and get double. messages

[ivory anchor]

Well, I'll save you some time then since you seem nice!

The tl;dr is no. Unfortunately, no other service offers the amount of requests per day at the same or greater context window. I have looked extensively and all of the "unlimited" you see floating around for unlimited Opus/Sonnet have a big ol' asterisk.

For example, there was a service (can't remember exact name) that offered Unlimited Sonnet, but the context window was 4k.

[half osprey]

I bought a premium subscription to Merlin AI and it was like throwing away the trash.

[ivory anchor]

Yep

There's just.. Nothing with the same access to Sonnet/Opus..

Phind is probably the closest, but it's only got 10 Opus instead of 50.

Costs the same.

[half osprey]

I'd like the devs to include a premium + subscription apart from the usual one

[ivory anchor]

You mean like... a super pro or something?

[half osprey]

Yes

Pro and Pro +

[ivory anchor]

The price point would be the biggest consideration. It's not out of the realm of impossibility, but Opus/Sonnet are two of the most expensive LLMs that exist. It wouldn't be a small jump in monthly pricing.

[half osprey]

• 50 dollars

[ivory anchor]

So $70 a month?

[half osprey]

Yes

[warm crane]

It's very expensive

[ivory anchor]

@primal acorn @royal basin @worldly mason Pardon me. I've never pinged you before because I know you're all very busy. If I could steal a few moments of your time... Would a steeply increased price version of monthly be feasible or is it just out of the question?

I apologize again if this is a stupid question, but if you read this; thank you. If any of you respond; thank you again.

[half osprey]

Yes, but it would be an option, there are applications that offer several pro and pro+ plans

One question, how many years do you think it will take for opus to reach the level of sonnet messages?

[ivory anchor]

It will surpass it by the end of the year, as Anthropic has announced Opus 3.5 coming out this fall.

[half osprey]

Opus 3.5?

[ivory anchor]

Ye

Co-Founder of Anthropic:

“To complete the Claude 3.5 model family, we’ll be releasing Claude 3.5 Haiku and Claude 3.5 Opus later this year.”

https://youtu.be/xm6jNMSFT7g

[elder hemlock]

This the second time i have a referral the plz gets canceled and i can't use the referral i have done

[limber ermine]

Try my referral

[limber ermine]

I guess the referral link works only for new account?

[pale horizon]

What’s going on with the free user experience, it feels like it can’t even perform simple tasks and when asking to correct it remains in a faulty loop, so annoying 🤬

[narrow lagoon]

this is the 1st time I used up all my Pro questions. When does it update like what time to give all the time back? \

[half osprey]

my messages do not increase and many hours have passed

[steel pond]

hello，I'm a subscribed pro user. Why haven't I selected the pro model for use now?

anyone know something？

[sterile forum]

y0~ the website said that discord server maybe quicker than email. sent an email like 5 hours ago about a billing issue. feel free to contact me back via discord. Hajime.Barea

[slim night]

Hey @elder hemlock, could you please elaborate a bit more? You got a confirmed referral, but didn't get a discount for that referral?

[shrewd knoll]

Not really, I use referrals and still get $10 every month

[finite fossil]

hello , is there any customer service for perplexity ? i need to help

[warm crane]

[e-mail]

[finite fossil]

thx

[warm crane]

But they don't often respond in any case to my experience

[finite fossil]

anyone knows how to use 「pro search 」？

[warm crane]

Just put the button on

[finite fossil]

where is the button , i saw many times ,

[finite fossil]

app or web version ?

[warm crane]

5 pro searches per day for free users and about 600 per day for pro users so basically it's unlimited

[warm crane]

I'm on the app but the web version is the same

[finite fossil]

i just joined perplexity pro , $20/month

[warm crane]

Send a screenshot

[finite fossil]

does the button should press to left or right ? 😔

[warm crane]

And if you have the pro version, leave the discord and come back with the pro link to get the pro role on the discord and have access to special Channels

[warm crane]

Right

The button lights up when you activate the pro search

[finite fossil]

where is the button ?

[warm crane]

.

[finite fossil]

no nay light , just black and white , so difficult to verify

[warm crane]

Are you making an effort or have you ever used the internet ?

Activate

[ionic sparrow]

deep dive ➤ #💬│general message @finite fossil

[finite fossil]

yes i tried many times

[finite fossil]

please tell me details , thanks !

is this group any customer service person?

[velvet tangle]

Hey @finite fossil!
Do you have any browser extensions? Content blocker or dark mode for example?

[finite fossil]

have some google extensios , but dont blocker

[velvet tangle]

I see, please disable all extensions and try it again

[finite fossil]

ohooo , ok thx

when i need to search many information for research from internet and google academic , which AI model should i choose ? ( just joined perplexity pro) ,

[velvet tangle]

Check this please: https://discord.com/channels/1047197230748151888/1240356137497530408

[finite fossil]

ie how can i select 「pro search」？

[vapid maple]

click on it or ctrl + .

[velvet tangle]

All you have to do is switch the Pro toggle to the right, and you're all set!

[finite fossil]

control . button just can see by eyes , cant be pressed

can i use chinese?

[vapid maple]

sure

[ionic sparrow]

deep dive ➤ #💬│general message

please check tutorials from this post

[finite fossil]

首先，我已join了pro 版，是不是先打開Pro button, 然後去選擇可選的AI 模型？（我想選擇pro search ,但好像AI模型里沒有這一個選項？）；

[velvet tangle]

"Pro search" is not a model, you can select the AI model via the settings -> https://www.perplexity.ai/settings/account
You can switch to Pro Search mode directly via the input field with the Pro button for each question.

[finite fossil]

每一次問問題都要按一下Pro button 嗎

[shrewd knoll]

Watch out!!! AI company war is coming this way

[vapid maple]

pplx should remember the state.

[deft escarp]

How do I get the Uber one pro promo code?

[finite fossil]

thx

[timid quiver]

Is there a way to delete a specific thread or at least delete the the question within a thread?

[frosty eagle]

Good luck. Uber apparently send out promo codes via email but a large amount of people never got it. Uber support seems to have no way to remedy the issue, I’ve talked to them several times on the app and on Twitter with no luck

[haughty brook]

You can delete a specific thread by swiping the thread to the left and clicking on the trashbin on ios. If you’re in a thread you can delete the thread by clicking on the 3 dots in the top right. and you can delete a question within a thread by clicking on the 3 dots at the bottom right of the answer under a question.

[fleet matrix]

I am unable to change AI model . it is always in Default, unable to change it in the drop down

[timid quiver]

Thanks, but swiping left just creates a new thread. I also don’t see 3 dots at the top right. The only options I get when I click on 3 dots at the bottom is to Share or Report.

[haughty brook]

You can only do this in the library on ios (see image1). on the browser you’ll have 3 dots (image 2). If you don’t see a delete option it might be a thread that is not connected to your account.

[haughty brook]

I assume you’re a pro user. Try to logout and in again. If that doesn’t work try again in a different browser / or without browser extensions enabled and try again.

btw you can get your pro role on the server by leaving and rejoining with the “pro discord” button on the settings page.

[fleet matrix]

Thank you

How to summarise contents from YouTube video

[timid quiver]

Thanks for your response again, but I don’t see a library option. This is all I see. Is there a pro version or a different version of the app in Canada?

[haughty brook]

This is the Pi app this app is not from perplexity

[haughty brook]

https://apps.apple.com/en/app/perplexity-ask-anything/id1668000334?l=en-GB this is the official perplexity app

[fleet matrix]

How to summarise YouTube video using perplexity

[timid quiver]

Sorry, just realized I clicked on the wrong icon.

[modest echo]

hey folks, on the perplexity API here. https://www.perplexity.ai/rest/user/settings

does anyone know what the values:

pages_limit 100
upload_limit 999
create_limit 99
article_image_upload_limit 999

are in reference to? I can't seem to find nay doco that describes them

i'm writing a script for perplexity that lets you see how mayn requests you have left for the AI usage each day

so far, looks like this, using the API and a violetn monkey script:

Would love to add the rest of hte values

[regal field]

pages_limit: the amount of Pages that you can create
upload_limit, article_image_upload_limit: docs, images upload limit
create_limit: image generation limit

[modest echo]

image upload limit doesn't seem to change when i attached images, the article image upload did. what's the dffeence between the two?

[regal field]

the upload_limit is when you upload files in a normal thread, and the other one must be for Pages

[modest echo]

so image uploads in pages?

[wicked glacier]

I'm on a pro subscription, but can't create images. When clicking the "Generate images" button - nothing happens. Plus - I can' change images creation modules on the setting page (nor the search modules). Tried different browsers. Anyone know anything about it?

[ember raft]

I have the Perplexity Pro plan which gives $5 API credit per month. I know that there 20 inputs per minute limit ( https://docs.perplexity.ai/docs/rate-limits ) But what's the limit per day, if there is any? I keep saying #ERROR after running API for a 200-300 prompts. Where to find the exact daily limit on the usage of API ?

Thank you.

[rich sage]

#💬│general message

[umbral sandal]

Hi! So I got a pro account, and then noticed that there is a Perplexity plug-in for Raycast. Using it requires an API key, which I noticed is a paid service. So if I use the API in Raycast, does that mean I get charged for the searches, versus just using the website or the iOS app?

[haughty brook]

Thats correct, the api of perplexity is a different service than the normal website https://pplx.ai/ But as a pro member you do get 5$ free api credits each month.

[gaunt terrace]

Hi - It is my first time using discord sorry if this is not allowed - I got the pro account but regret it. It has been less than 72 hours, it says I can get a refund since it is the annual subscription. I emailed but it said 1 - 2 days to get a response back. I already cancelled my service. I also tried the chat, no one answered. Please help. I am worried if hte email team doesnt see my message within the 72 hours I will not get a full refund. Thanks!

[umbral sandal]

Okay thanks!

[slender cairn]

Why is the Home icon a "Search" icon

[versed magnet]

it is almost quite literally functionally the same as pressing "New Thread" isn't it? the homepage is a textfield for search 🤷‍♂️

[warm crane]

Because the home page is the search page

After it's true that « home » is weird. You should rather put « search »

[shrewd tree]

why if you ask Sonnet 3.5 in Writing mode what model it is, it says… “I am an AI assistant created by Perplexity. I don't have a specific model name or version number that I can share…”

GPT 4o will say it’s Open AI, and Opus will say it’s Anthropic. What’s going on with Sonnet 3.5?

[zenith hawk]

As long as your message was sent within the time frame they can’t penalize you if they didn’t reply within that timeframe. I wouldn’t worry about it as long as your emails are timestamped within that period, it will show you did your part correctly

[haughty ginkgo]

do I need to also pay $20 for OpenAI Plus and another $20 for Claude Pro if I am already a Perplexity Pro member? in other words, what kind of access to GPT-4o and Claude Opus do we get via the Perplexity Pro membership... is it limited to what they consider the Free plan?

[real bison]

Hello. I’ve been trying to subscribe for the first time but billing address is greyed out. Any solutions?

[sonic forge]

I'm pro user not part of any individual organisation in perplexity. Why do I have a file retention of 7 days and how to change this?

[modest echo]

can perplexity open images or web pages to read actual contents? or is it more of a sumarization thing? like if i give the search an explicit url, can it search that url or not?

[oblique badger]

I just wanted to ask whether any other person got this. It was like introduced in a rush.

[hollow hound]

Yes. It will review the page and dump the text into a .txt file. It’ll then use that as a source for its output.

[modest echo]

does't seem to do the same thing with images though

[viral jacinth]

Greetings! I'm experiencing an issue with my Pro bonus. The system consistently prompts me to wait 10 minutes, but this has persisted for an entire week now. I'm curious if any other users have encountered a similar problem. I've already reached out to customer support regarding this matter, but unfortunately, I haven't received any response for a week.

[barren reef]

https://www.perplexity.ai/search/why-do-i-have-a-file-retention-3CbFbaiZSL.9fE3mZTQ4IA#0

[versed magnet]

something does seem a bit off

• Pasted an image (google olympics thing) into the search field on perplexity home page. Some image recognition takes place, because it does web searches related to the image
• But the actual model generating the response apparently can't see the image; it's response is just about the web sources (and so wrong basically)

• Pasted it again. This time, no web search happens, and the model provides an accurate description of the image (iow: it worked properly)
• It seems that both versions of the file are there, but only one can be removed from the file upload thing in the search field

[wise wind]

what model do people like to use? Are you favoring the faster models rather than the most capable ones so that the searches run reasonably quickly?

[hard siren]

Is there some way to give Perplexity contextual project awareness, either locally or for a GitHub repository. I'm looking for something similar to GitHub Copilot Chat's @workspace command.

[regal zenith]

I cannot generate images via the web interface. The result always looks like this:

Same with playground 2.5 and DALL-E 3

[hollow laurel]

Hi, how could I link perplexity pro with discord?

[paper coral]

Hi everyone, I have just discovered that Perplextiy is generating tables and graphics directly into images at the PNG format. Anyone have heard about that ?

[ionic sparrow]

#⚡│ask-community message

[hollow laurel]

Thank you 🙏

[hollow laurel]

I've got a redirect after clicking this into native macos discord app, but I still don't have access.

[hollow laurel]

Ah, leaving and rejoining helped

[buoyant nymph]

Does anyone know why it searches every time for german sites and not for english ones, is there a setting or something

[ivory anchor]

All the models on Perplexity are essentially real-time

[ivory anchor]

What is your language set to in your profile?

[flint zealot]

Has anyone noticed better results with a specific model lately?

Also, how does the default setting work? How does perplexity choose which model to default to?

and does it state which model it used in its outpuT?

[ivory anchor]

You've asked a loaded question. Each model is good at different things. There is no universally better LLM...

The default is set in your profile...

No it will not state what model is being used without hallucinations... But you can use the rewrite function to change a response from your default to a different LLM.

[flint zealot]

When you say default is set in your profile, are you talking about being able to change the model here?

When I look at the options, "default" is one of them and says "Optimized for fast search by perplexity" so I assumed there was some method perplexity was using to switch between diff models when "default" was set

[ivory anchor]

"Default" is the inhouse Perplexity LLM

But you can set that to a different model which becomes your default

[dense seal]

Hello. I am a user from Korea who frequently uses Perplexity Pro search.

Among the devices I own, I have an iPad mini 2 that was released in 2013. It's a very old device.

It might be a pointless thought, but I tried to access Perplexity web and use Pro search on the iPad mini 2. I can’t login also the search button was deactivated, so I couldn't perform a search.

Is there any way to use Pro search on the iPad mini 2? Even if it's not possible, if you could tell me "It's not possible," it would save me time from trying various methods to access Perplexity on the iPad, such as installing Chrome web browser, Brave web browser, Puffin web browser, and so on.

I'm receiving so much help from Perplexity's Pro search feature. Regardless of whether you can provide an answer, I wanted to express my gratitude. Thank you. And I look forward to your response.

[tender jacinth]

Which model was again behind default, was ist Haiku?

[austere latch]

Hey all, I've constructed a prompt for pro search that is intended to run through multiple stages of searches
What i'm finding at the moment is the "Pro Search" steps area shows my intended steps, but after completing the first round of searches it thinks its finished and doesn't actually execute any of my subsequent stage searches
Is there some way to ensure those next step searches are performed and that it doesn't terminate early before it's gotten to the meat of my problem?

[radiant compass]

I really like Perplexity and want to use it for search, but EVERY search I do I have to wait for Cloudflare verification and it really makes it feel slow and annoying to use. I am in Vietnam, so thats probably part of the reason

I use an updated version of Chrome, without VPN

Do you think that if I switch to pro, Cloudflare will leave me alone?

[barren reef]

Use another browser, incognito mode, or turn off the Chrome extensions to see if the problem is fixed.

[radiant compass]

Ok it seems it was caused by "uBlock Origin"

[barren reef]

If you are sure the problem is caused by this extension, you can choose not to use this extension on the Perplexity page (Click the blue button to disable or add to trust site list in the extension settings)

[radiant compass]

yeah thats what I did

[soft fable]

Hi, Is perplexity capable to view urls? when i give it a url and ask perplexity to summarize the page for me it makes up an answer about something completely different from whats on the page.

[oblique fog]

Hello! Previously, when sharing a page on X/Twitter, the tweet used to display a preview image. This seems not to be the case anymore. Is this a setting on X or has something changed on the Perplexity side (HTML meta tags, ...)

[severe valley]

when i am opening any of my collections i am getting this white blank screen, why ?

[clear cipher]

is sonar huge already good enough or is it better to wait until it is improved by experience?

[ivory anchor]

It's not bad it's just not on par with Sonnet 3.5 or 4o

[flint zealot]

thanks for the info!

[wise wind]

ok thanks

[ionic sparrow]

correct #💬│general message @ivory anchor

[near eagle]

I am not able to set up a login, as i used login from gmail. How to fix that, any ideas?
I wanted to use it using id password as from my work laptop gmail not allowed.

[hallow glade]

What is the campus program? What do students receive from it?

[haughty brook]

here you can read about the benefits you get
https://www.perplexity.ai/campus-strategists

[dapper trellis]

hey guys, do you know if perplexity pro uses ChatGPT-4o-latest (2024-08-08) or an earlier model?

probably not but when do you think it will update

[lilac depot]

Perplexy keeps repeating the same response over and over, even after letting it know that the provided solution is not working. It seems to be stuck in a loop and does not know how to get out. It literrlay sucks at C++ and MQL5 code. it constantly producing MQL4 code for a MQL5 project. Even when I provide samples of working code, it seems to be a complete idiot and keeps repeating the same failed responses. I want a refund.

[hollow ginkgo]

Hey @lilac depot!

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.

[hasty linden]

I get a blank page when i try to open one of my collections suddenly. How can I access them?

[ionic sparrow]

#1272898044765147157 message

[hasty linden]

Thanks!

[clear granite]

Why do I have to switch it to pro everytime I do a search? Anyway to avoid this?

[unique crescent]

Does Perplexity always narrow its search strictly when you provide links with your query? I have to ask it specifically to broaden its search whenever I provide additional resources to it, and even that often requires a few iterations.

[slim night]

This issue should be fixed now, sorry for the inconvenience.

[shrewd knoll]

Final got api 405b 😆😆😆

🥳🥳🥳🥳

❤️❤️❤️

🌟🌟🌟🌟

Thank you!!!! 🎉🎉🎉🎉

[shrewd knoll]

I hope we will get the 405b chat version and the instruct version too 🥺

[hasty mortar]

After logging in, it keeps showing an error message saying "Please try again," and I can't log in.

But it works fine on the web version.

😩

[tiny whale]

Good day to all Im look forward to buy Perplexity Pro Can Any1 be kind to share their referal code/link with me ? Much thanks in advance
It will be super helpful for me 🙆‍♂️💐🙏

[shrewd knoll]

https://perplexity.ai/pro?referral_code=TQRFFK3J

Here you go

[tiny whale]

Much appreciated for you help , but I tried its telling invalid Code on the payment page ur guidance may help me sort this?

i even tried following their instructions on their help page but it seems that it's not working.

[shrewd knoll]

I tried other refer codes too but they all started to reject the refer codes now

@slim night & @velvet tangle is the refer code discontinued now all refer codes are invalid?

[shrewd knoll]

@velvet tangle @slim night

[tiny whale]

even im wondering the same...ive been trying codes past 3 hours

[shrewd knoll]

Better wait and see whether @velvet tangle & @slim night responds before purchasing pro

[tiny whale]

no avail . all codes invalid at payment page nothing is being refelcted on perplexity it detects the code but on the payment Page it forgets annual we try to follow the help page instructions it doesn't seem to help at all

[tiny whale]

ouuukiiii

[shrewd knoll]

Perplexity.ai side bug problem ⚠️

Pray @slim night & @velvet tangle responds and fix this 😢

[coarse monolith]

Why do I get Beta object has no attribute 'prompt_caching

anyone knows?

[mint pendant]

anyone else getting an issue on the web app where you need to press the return key twice to submit your question ?\

[distant acorn]

lol me too

also 'file upload failed' today, dont know why. and good God when you attach a file, it keeps staying for the rest of your life when you ask another questions lol and has been like this forever....

[shrewd knoll]

@slim night @velvet tangle please respond
Refer code can’t used

Please help fix this

[shrewd knoll]

😩😩😩😭😭😭😭 @slim night @velvet tangle they are not responding to the refer code problem

[hidden cape]

Can any one explain how I can embedd perplexity page into my website blog?

[tawdry adder]

i dont know guys but i think sometimes that not gpt-4o is serving but gpt 3.5 :/

it ignores often the attachement partly, overall degraded performance since 1-2 week or so

it feels like system prompt was, "give as short as possible answer to save tokens" ;/

when i compare to chatgpt.com the answer is much more detailed and extended

[bronze lantern]

Hi. I'm using perplexity API (/chat/completions). I'm instructing the model to fetch LinkedIn posts related to a keyword. Below are my instructions. How do we make it more accurate since currently, the model gives results which are not posts but articles, company pages, etc? Here's the prompt:
[
{
role: 'system',
content:
'Be precise, concise and always provide post author name, title, content, date and url in a valid json format.',
},
{
role: 'user',
content: Get me 5 real posts made on linkedin.com by the users related to ${formattedKeywords}.${formattedExclude}. Only include posts from individual LinkedIn profiles; do not include articles, company pages, or any other types of content. Strictly don't exceed token limit of 1500.,
},
]

[spring anchor]

I can never generate imagine with pro. Am I doing this wrong?

[tawdry adder]

an gui must appear on the right upper corner where you can then generate images with a aprompt

[spring anchor]

Sorry can you clarify what you mean?

What is a gui and how do I get it to appear?

[tawdry adder]

here

then there is a small gui

[shrewd knoll]

I got a lot of successful refer but didn't get any $10 off subscription as promised and I didn't get any free $5 credit

I’m devastated I didn't get as they promise
But when I report the issue to them
@velvet tangle & @slim night didn't respond but ignored the issue

[ionic sparrow]

https://discord.com/channels/1047197230748151888/1225557012532695060

[ionic sparrow]

#👑│pro-lounge message

[shrewd knoll]

Thanks

[muted saffron]

is there a way to add a collection parameter to perplexity search URLs such that the collection's prompt would be part of the query, e.g., https://www.perplexity.ai/search?q=%s&collection=writing

[potent bison]

Any update on the status of Uber One code distrubution? I really would like to use Perplexity Pro soon.

[versed magnet]

not that i'm aware of.. but that'd be real nice wouldn't it

[fiery oyster]

does AI get detected if i get something from perpexlity?

[versed magnet]

perplexity gives you stuff.. generated by AI...

[shrewd knoll]

Finally got a response regarding the issue @slim night

[unborn yew]

Is there a way to do this properly cause this won't work - This is a set of three languages - [Kotlin , Java, Python ] . If the user's prompt includes one of these three , ALWAYS provide a brief answer for the other two languages not included in the prompt.

[tacit remnant]

how can I get a discount as a student?

discount for Perplexity AI

[blissful bloom]

Hey folks, i used perplexity yesterday a lot ... at midnight i noticed thad the pro switcher turns of .. when i hover over with my mouse its a notification thad said ,, one pro search left today 9 hrs later perplexeity say ,, 36 Pro searches left ,, WTF is this?? Cause it's sounds like a limiter! I never read on the page something about a limiter i even asked perplexity when i was free user about this -they couldn't give me a clear answer! Can anyone please explain me how many REAL pro searches / per day a user have!? And how long a user need to wait untill the full pro searches are refilled after he maxed out the pro function. - Thank you @slim night @velvet tangle @primal acorn

[velvet tangle]

Hey @blissful bloom!
Please check this: https://www.perplexity.ai/hub/faq/what-is-perplexity-pro

[blissful bloom]

@velvet tangle there is not even one line thad adress this issue on your FAQ!Please give ma a clear answer on this Problem and not a 08:15 Link! By the way who designed your site? These different fonts are making crazy cant focus on simple information! 😆

[velvet tangle]

Please take a look at the second paragraph 🙂

Pro users get at least 300 Pro Searches per day. Each Pro Search deducts one credit from your daily limit, and credits are restored 24 hours after they are used. For example, if one credit is used at 10 AM, the credit will be restored at 10 AM the following day.

[ionic sparrow]

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

[ancient cloak]

Is there a way to get pro lifetime?

[haughty brook]

this is not possible

[dry vault]

run perplexity on ur device

[stone plume]

Can I add a source to be used when answering a question?

[stone plume]

Oh, just said that read that page also and that did it 😊👍

[ivory anchor]

@velvet tangle is there a way to turn off word wrap for text/code blocks like it used to be before the update :s

@ionic sparrow Do you know?

[worthy zealot]

Is there a way to return to a conversation / thread after you have converted it to a page?

I recently converted a long chat to a page and it lost a ton of the context from the chat and I can’t figure out how to get it back 😅

It’s so strange because I can see the original chat. When I click on it, it starts to load, then turns to a white page, then it crashes.

[primal acorn]

Hey, this does not sound right. Unfotunately there's no way to return a page to a thread state. I will look into this though, if you have more details about the isssue itself I would highly appreciate if you shared those (app version, using a VPN, screenrecording) with support through [e-mail]

[cold gust]

We have spent many hours building up reference material and an archive of resources. Yesterday our entire library disappeared, there is nothing there. I can follow some old links we have used on email to some threads but from the login page there is no library for us to navigate through previous threads. It's a serious issue.

[small prairie]

Get a Job

[quick iris]

Hello, could someone point me to a guide for creating prompts for collections? It seems to me like the prompts I create for them just don't work.

[shrewd knoll]

I’m a pro and I already informed the perplexity and @slim night about the $5 free credit I haven’t received but neither helped me receive the $5 free credit

[patent rune]

I have a question about the image generation feature of 'FLUX.1'. Is there a way to generate portrait-oriented (vertical) images?

[hoary oar]

How does the news thing work like who writes the articles?

[lethal ibex]

Does perplexity use gpt-4o-lastest now?

[odd rose]

Hey Perplexity Support Team!
I've got a question about uploading photos/files to the AI. Right now, when I upload a new photo and ask a question, the AI takes into account all the previously uploaded photos in its response. But I want the AI to only consider the most recently uploaded file, like how it works in ChatGPT.
Is it possible to make the AI use only the latest file for each request?

[regal turret]

Hello, Perplexity!
Are there any plans to add Grok 2? In terms of quality, apparently, it bypasses Claude 3.5 and 4o.

[shrewd knoll]

When will I get a free credit of $5 for api after the purchase pro? @slim night

[hollow hound]

#1237163951750385705 message

[odd rose]

Ohh. Thx

[hollow hound]

TLDR it’s a pain once an image or file is attached as a source since it’ll get used in all queries going forward. Once you get what you want from the image/attachment, manually remove it as a source.

[dapper trellis]

hey guys quick question, is there a way to ged rid of image attachments from your convo once you send them? deleted from here but still shows up and messes up the response

[hollow ginkgo]

Hey @dapper trellis!

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.

[tropic olive]

Hello Perplexity Team! About to submit a request for DuckDuckGo to add a bang shortcut from their search engine to your website and wanted to clear that this would not violate the ToS. More info on bangs: https://duckduckgo.com/bangs

[orchid citrus]

我要怎么在discord使用perplexity完成常规的搜索任务？

[gilded canyon]

Q: how do you get it to follow simple instructions after providing the instructions? I have to REPEATELY write "follow instructions!" for it to actually do it, it's such a waste of time and effort.

[modern grail]

Once your response have been generated you must edit the sources and remove the file you don’t want. Then, the response will be regenerated without the file you don’t want. That’s the only way I know about. There have been requests to improve this by allowing to remove files at any time but for some reason they are not doing this.

[normal scaffold]

hello. i just want to ask that how to change the aspect ratio when create image in perplexity?.

[haughty brook]

changing the aspect ratio of an image is currently not supported

[ancient vessel]

THX

[ionic sparrow]

https://www.perplexity.ai

[jaunty stump]

Does anyone know if it's possible to specify a source for perplexity to use?

[rich sage]

#1262520766642131054 message

[jaunty stump]

Thank you. I tried that out and it seems to work most of the time. However, sometimes it chooses random pages besides the ones I gave it. Any idea why that is and if there's a way to solve it?

[shrewd knoll]

I finally got my api free $5 credit but sadly I have to report to get free api $5 credit every month now instead of automatically getting it after purchasing the pro

My refer code has a problem that no user can use my refer code when I reported to the support team by email they said it is not a widely spread issue so that means they are not gonna fix this matter immediately

[maiden lily]

Hi! Just curious in how Perplexity works in this specific case. As you see from the screenshot, I did a simple search inquiring about "cognitive dissonance" and it proceeded to give me a great response. I then noticed it listed out 5+ sources, but ended up not using any. In this case, is the answer straight from Claude 3.5 Sonnet?

[fossil vine]

Hi all,
I purchased the pro membership today, I am trying to use free $5 credit for APIs.
I am able to see written in small letters that pro members will get $5 below Available credits box but in total balance is 0.
and I am not able to generate API keys.
Can someone please help , how to start

[shrewd knoll]

Report to @slim night and [e-mail]

To get the $5 credit

[open shuttle]

Regarding API using, when will the Llama 3 series be phased out? Or they won’t phase out the old series?

[magic patio]

Hi i am new to this channel and was woundering if you guys have any tips or prompts for increase the AIs coding answers that helps with getting better helpful coding answers when i am stuck?