⚡│ask-community

[vapid maple]

@glad venture i just uploaded to the cloud: https://llamalab.com/automate/community/flows/48296

[tranquil nova]

Hi. I’m a new user. A question: when i upload a document, the ai analyzes even the pictures and the data contain in it or just the text?

[vapid maple]

even the pictures and the data contain in it ✅

[languid pond]

Hey y'all - is there an API or JavaScript SDK that allows me to generate text buuuuut cites it's sources like the app does?

[vapid maple]

in closed beta: here https://perplexity.typeform.com/to/j50rnNiB?typeform-source=docs.perplexity.ai or [e-mail]

[languid pond]

Thanks! 🙏

[dim dagger]

Request for SAML SSO Integration with Perplexity.ai

Dear Perplexity Team, hello @hollow ginkgo,
I want to integrate the Perplexity.ai platform into my existing infrastructure. For this, I require your assistance in setting up SAML-based Single Sign-On (SAML SSO).

My Request:
For the SAML SSO configuration, I need the following information from you:

• The Service Provider Metadata from Perplexity.ai in the common XML format. This needs to be imported into my Identity Provider (Keycloak).
• Instructions on where and how I can import the IdP Metadata from my Identity Provider (Keycloak) into the Perplexity.ai login portal.
  Through the SAML SSO integration, I want to enable centralized authentication and seamless access to Perplexity.ai services for my users.
  I would greatly appreciate if you could provide me with the required information and steps. Please feel free to reach out if you have any questions.

Thank you in advance for your support!
Best regards,
Torsten C.
From: Für SAML-SSO bei perplexity.ai benötige ich die 'Service Provider Metadata' …

[eager rivet]

Hey @hollow ginkgo , I am experiencing significant delays with perplexity when I have attachments added and ask it to answer questions using the attachments as sources. This was not a problem for me before, but the last month or two it is a problem I keep on having. Anyone know why? its taking at least 30-40 seconds before responding, sometime even up to a minute. Note: I am a paying user.

[barren birch]

Does anyone know what AI model is used when Pro search is selected?

[velvet tangle]

The selected model is always used

[lyric ermine]

What's the correct APIurl?

[kindred ginkgo]

How can I set or choose the Model of the "Perplexity - AI Companion" Chrome extention to "Claude or gpt4o?

[velvet tangle]

https://docs.perplexity.ai/reference/post_chat_completions
-> https://api.perplexity.ai/chat/completions

[velvet tangle]

Via settings: https://www.perplexity.ai/settings/account

[lyric ermine]

I received an application from a friend, he was using API.pplx.ai

This link seems to be down

I'm also seeing API.pplx.ai/v1....curie?

This too seems to be down

[rich sage]

I use: client = OpenAI(api_key=YOUR_API_KEY, base_url=’https://api.perplexity.ai' in several of my apps and they all work normally.

[lyric ermine]

It seems like there is a learning curve here, it's okay. Thank you, i will do my research and find a way

[kindred ginkgo]

Thank you! Edit:.. I got a response from the Perplexity Team: "Thanks for reaching out! For the time being it's not possible to select the AI model for our Chrome extension. But it would be a cool feature to have! I'll note it for our product team for future iterations. "

[jaunty mantle]

I'd like to work on a mini content strategy for my business. Can Perplexity help with the various elements like audience research, competitor analysis, keywords, auditing existing content? If so, what would be the best focus option?

[hollow hound]

I would avoid focus options for that task and look into using a structured Collection Prompt that instructs the model to act as a marketing/research expert. You can then start a thread from within that collection to force the Collection Prompt. The New Agentic Pro search should help with this a ton. here's examples of more advanced multi-step searches using the new Pro Search.
In regards to the Collection Prompt, review both of these frameworks from the community. AutoExpert is specifically good to invoke "Experts", using a markdown table and a structured breakdown of the user query, typically taking multiple replies due to length.
https://discord.com/channels/1047197230748151888/1163749448447434802
Scratchpad-Think forces the model to detail it's "thought" process and workflow before arriving at a final answer. This also includes sections where it reviews the text so far and explores thought-provoking questions. This is my own framework personally that get's used for 99% of the threads I start within PPLX (and is designed to fit within a Collection Prompt, which has a max of 2000 characters).
https://discord.com/channels/1047197230748151888/1223058316662538331

[hasty hawk]

Can anyone use Perplexity atm, or is the problem on my side?

[vapid maple]

i had no problem, try to clean the cache/logout login etc?

[hasty hawk]

I waited ~15 minutes and did something else now it works again.

[hasty hawk]

more or less...

[vapid maple]

https://discord.com/channels/1047197230748151888/1236008988919988284
make the link sharable, and share it there

do you also face this in incognito mode?

[signal glen]

Which Perplexity model is the best when it comes to grounding answers in actual sources?
Using 4o, I constantly get citations that lead to websites which don't contain what it's referencing. And yet when Copilot or Bing searches, the citations are taken straight from the page and the hyperlink leads right to the quote. There's gotta be a way to make perplexity more grounded like that

[last leaf]

Why does my model keep reverting to gpt-4o when I ask it to use turbo and have turbo set as my default?

[vapid maple]

did your query contain uploaded images?

[last leaf]

Sometimes yes, sometimes no

[vapid maple]

in case of images: pplx uses gpt 4o instead of gpt turbo

inthe absent of images: if ai model is set to gpt-4 turbo ( https://www.perplexity.ai/settings/account ) but you get gpt-4o instead: create a bug report in #1111765634267742408 / #1140622008086970420

[last leaf]

I do not want 4o used for anything, I asked for it not to be used. How do I turn it off completely? If I wanted it to be used, I will select it

[vapid maple]

rewrite in sonnet / opus is your best option if you want to avoid gpt 4o for images

[last leaf]

Those models, as well as the 4o, don’t perform the way I need them to for the work I am doing. I feel disappointed that perplexity is secretly changing the model as soon as I have images without actually stating that it does so. I am paying for turbo as my default model, and I am told it will be. I feel uncomfortable about this hidden switch to a cheaper model when I pay for a full subscription

[vapid maple]

post your use case in #💬│general / #1111765634267742408 maybe some one can help you out on that

[stiff crescent]

Question I’m sure gets asked a ton of times but can’t seem to search discord - if AI model GPT4o if used instead of perplexities sonar large, do they both search the internet the same, and then the LLM is just used to write and summarize findings, or do I need to use sonar large for the best live internet search

[tight ridge]

I need some help with the API,
currently using "llama-3-sonar-large-32k-online" to query the following (which is giving me different results and error to if I use perplexity directly which is almost always right:

"Determine GCM (kg) for {{vehicle_details}}:

1. Check direct GCM specification
2. If unavailable, check similar variant's GCM
3. If still unavailable:

• Find GVM and max towing capacity
• If GVM unavailable, estimate from kerb weight + max payload
• Calculate: GCM = GVM + max towing capacity

4. Verify calculated GCM across multiple sources

Return GCM value"

[round rivet]

I'm trying to get api credit with my regular credit card but it goes void every time. is something wrong?

[hollow hound]

International card?

[versed magnet]

and then the LLM is just used to write and summarize findings ✅
That's my understanding of it how works – same / similar sources will be retrieved regardless of model selection, but how that material is handled / parsed and ultimately used to inform the response will vary by model

[unique cove]

What’s going on, why happened this situation?

[boreal needle]

Hello good people! I have a question about gpt 4o model. If i understand correctly. If i pick the gpt 4o model without the PRO SEARCH option my credits from the subscription still get deducted right? GPT 4o is limited even without the PRO search if I understand correctly. Thanks for the answer in advance!

[crisp minnow]

Hello everyone! I recently got a pro subscription and I Subscribe for the monthly subscription with a referral quote, and then 15 minutes later I realized I made a mistake and I switched it to the yearly so I only paid $190 for the yearly plus the $10 for the monthly, Because of this I was wondering if the $10 for the monthly will be returned?

[slim night]

Hey @boreal needle! Daily uses are deducted when you are using this model, with Pro off, too. That's right.

[slim night]

Hey @crisp minnow! The annual subscription costs $200 with the discount, so you just switched to the yearly subscription covering the difference.

[boreal needle]

Thanks ! Which models do not use up the credits good sir ?

[slim night]

The Default model doesn't count toward the daily limit.

[crisp minnow]

thank you. I just thought with the referral code the yearly subscription would be $190

[slim night]

Referral codes are applied to the monthly subscription.

[candid hull]

Hi, how to set writing mode as default in mobile app ?

[boreal needle]

Thanks mr. Romanov

[crisp minnow]

Thank you Mr. Romanov. I was just referring to what an acquaintance told me based on our referral that they got. This is image that they sent me

[weak swallow]

How do I change the web browser that is used by ios app to chrome?

[ivory anchor]

$10 off, noice

[fluid lark]

Doesn't it use the system default browser? The frame when you open the page overall is always safari, but you have icon to open it in browser in bottom right

[weak swallow]

Thanks let me try

[merry herald]

hey i got a question... how can we leverage perplexitys image generation to do an age progression given a subject image? is there a prompt for that?

You are a specialist describing images on a prompt valid to feed an AI specialized on anime image generation. You must omit any explanation or introduction or final words, write only the crafted prompt and be as detailed as possible.
```  ```Query Prompt
age progress the subject of this image by 20 years```

[lone vault]

how do you unpublish without deleting an article in perplexity?

[surreal glade]

can anyone recommend the best prompt in #1124060874785685554 that can analyze a coding project topic and gives you step by step details to guide you to do the project?

[arctic sand]

Hi, I had to use a proxy IP to access Perplexity for some specific reasons. Perplexity usually has Cloudflare human authentication for proxy IPs, which works fine for web, but not for mobile APP. This means that Cloudflare human verification cannot be displayed on the mobile app, resulting in a complete inability to use Perplexity on the mobile app when using the proxy

Is there another way to solve this besides changing the proxy node? Can Perplexity add human verification to the mobile app?

[rose needle]

Are we supposed to be able to use Llama 3, it says so in description, but it isn't available in the dropdown..?

[haughty brook]

Sonar Large 32k is based on llama

[rose needle]

Ah ok, thx

[jade goblet]

I have Pro and clicked on Enterprise to see something and now the banner at the top of the page says complete Enterprise purchase.

How do I get rid of this? Tried deleting cache etc. Tried another computer same thing but only on the desktop not the iOS app.

Never heard back from support.

Signed out completely everywhere also.

Pro is working fine. Just this banner.

[small prairie]

Do you have an adblocker? If so, you can likely right click and block element.

[real flame]

Recently, it seems that searches are being refused when asking questions. Has Perplexity started using its own prompts? If so, this is not a good change as it affects the output too much.

[slim night]

Hey @jade goblet! You can delete the organization you created while checking the enterprise offer and you won't see the banner.

[slim night]

Hey @real flame! Please share share the refused/failed search cases with thread URLs in #1140622008086970420 so the team can check and fix.

[acoustic nexus]

Hello, is it possible to download the voice generation to my computer (Text-to-Speech Audio Output)?

[hollow hound]

Not without recording it

[shut locust]

¯_(ツ)_/¯

[prime jasper]

Hi
Is it possible to generate images with the iOS app ?

[slim night]

Hey @prime jasper, sorry, that feature is currently only available on the web and mobile web versions. The mobile team is working on bringing it to future versions of our apps, as well.

[fading kernel]

Hey @slim night, I subscribe to the Pro version and noticed the new Sonar model is available. I'm looking for recommendations on whether it's a good idea to switch. I'm currently using Opus.

My primary use case is research (e.g. geology and minerals) and learning about new subjects (e.g. rocket engines)

The most important feature for me is accuracy. Whatever content is generated must be accurate. I've found ChatGPT to be almost useless as it hallucinates a lot of things - even titles of citations to the scientific literature.

Opus is very good in terms of accuracy and citations. Is Sonar better?

[fading kernel]

the other criteria is level of comprehensiveness (so accuracy and comprehensiveness are the 2 most important capabilities)

[velvet tangle]

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

[flat nexus]

I would like to ask people who are using Pro, is there any difference between free and paid versions?

[fading kernel]

yes - longer, higher quality, more comprehensive responses. Different models provide you with different outputs so you can get better topic coverage esp if you're conducting research. for certain types of queries, I've seen Sonar use Wolfram to generate numerical responses using formulas (e.g. specific impulse for NTP engines using H2 versus N2 as propellant).

I'd highly recommend the pro version

[fading kernel]

Is this something that a perplexity team can conduct research into? make recommendations on model usage?

[timid wagon]

Hi all, how can I get access to Perplexity Pages? I'm a free user but can't even find that it's available to premium users.

[noble kettle]

At one point as a pro subscriber I was invited to participate in the beta features through AI labs (or something like that). I didn't immediately act and now I can't find it. Is there a link that someone could share?

[muted lotus]

my pro subscription was due renewal today, and i got a refferal today itself, but the auto renewal dint deduct 10 dollars it took 20 dollars. what should i do? i emailed the suupoort but should i cancel subscription or what?

[velvet tangle]

Hey @muted lotus!
Please wait for the reply from the support team ☺️

[muted lotus]

alright thanks for the reply

[faint slate]

is there any affiliate programs?
I think i could help sell subscriptions

[formal mortar]

If I want to use Perplexity to create a first draft of legal documents, which model should I use?

[hexed robin]

You probably figured it out by now but you can still generate images on phone, it just has to be in safari or google, browser doesn't matter just not in the actual app itself

[uncut latch]

Any idea to get around the cloudflare checks ? I am logged in with a pro account. I almost always get that page.

[crimson sandal]

i'm interested in using perplexity for a quick alternative to scraping

give it an article URL, and ask it to classify the contents of the URL

before i burn lots of credits, will this work- has anyone tried this?

[hard jungle]

You have saved my time. Thanks very much. The product manager of the Perplexity IOS app does not...

[vocal nova]

hey, does anyone know how the tokenisation works (how a text is put into tokens) ? More specifically, is it possible to estimate that number based on text length, and does it vary in function on the chosen model?

[wild creek]

Hii, I am creating perplexity ai pages on different brands. Will there be any copy right issue? Will brand object for these ai pages ? I am creating content using perplexity ai and adding images using its ai & at the end circulating in perplexity ai pages link .. I am not adding anything extra, so as perplexity ai is anyhow providing the details of resources from which its generating the content and providing the credit on images. The brands should not have any issue with this right ?

[ivory anchor]

Rule of thumb is 1 token per 4 words.

[ivory anchor]

Classify? Do you mean summarizing the content or am I out of the loop for scraping lingo?

Perplexity can summarize the contents of a URL page, yes.

[ivory anchor]

Are you running a VPN?

[crimson sandal]

I'm probably going to ask it to do some simple topic labeling. or just extract a summary yeah. what im worried about is whether perplexity is well behaved summarizing only the specific article i supply it

[versed magnet]

when you mention credits, are you talking about doing this with the API? Or do you still mean using the perplexity.ai website? If it's the website, you don't really have to worry about credits (there is a 600 daily limit for most the models), and it wil perform better at summarising specific URLs you give it than the API.. I'd say just go ahead and start playing around tbh

[sonic sky]

Did you know any French community for Perplexity ? On discord or others support ?

[dim patio]

Hi, Is there a way to use voice in perplexity to ask a question but have perplexity type out the answer silently instead of reading it back?

[turbid seal]

I have found many quotes that are produced in Perplexity are not found anywhere online even in the reference provided. Is there anyway around this?

[vapid maple]

mind to share the thread? i would imaging add when citing/quoting the sources, do not paraphrase, but in verbatim to the ai profile or better to the query would help

[turbid seal]

https://www.perplexity.ai/search/What-are-some-WjfEOfW3QaO.zBIysBM9hg

I didn't check every quote but the ones I did check on don't exist. This has happened several times for me.

[vapid maple]

https://www.perplexity.ai/search/What-are-some-ld.sQBpfSK6mKA_Kb8YlaA

[turbid seal]

Can I put that prompt in the Collections prompt or do I need to include it every time I am looking for specific quotes?

[vapid maple]

you can put it in the collection prompt, but from my experience, the llm follows the instruction better, when you put it in the query

[turbid seal]

ok, thank you

[hollow hound]

If it is part of a collection prompt and it’s ignored just input “adhere to the collection prompt” and it’ll follow it.

[sonic forge]

noticed some folks have the unlocked role
can someone help me with more info on it?

[velvet tangle]

Just for the security on the server, nothing special

[hearty blade]

Hello - How can I upload videos or add a link to a google drive folder containing videos for analysis like in chatgpt-4? I have the pro plan.

[ivory anchor]

Don't think you can. Maybe the videos individually, but not a Google drive folder.

[feral rivet]

Video? The only platform I have used that will take in video input is Gemini via the developer console, with chatGPT I don't think it “watches” the video, Gemini dose

chatGPT can manipulate it with python tho

[feral rivet]

So is there a “Locked” roll, like jail, haha

[velvet tangle]

No no 😄 just the unlocked role for raid/bot protection

[feral rivet]

Haha, makes sense

[ancient fjord]

Hi, is it possible sync library without logout and login again on a different machine?

[velvet tangle]

You can just refresh the site

[ancient fjord]

this doesnt work when i am logged in from two different machine at the same time.

[vapid maple]

#📱︱feedback-mobile message

[mellow owl]

Hi, I am a student majoring in mathematics, and I often use ChatGPT for my school assignment-related questions, to better understand mathematical concepts, and for coding-related side projects, algorithms, and machine learning questions. Recently, I learned that Perplexity is a strong search engine, so I decided to compare ChatGPT and Perplexity to determine which one would be better for my school-related questions and projects, as well as CS-related questions. I would like to hear from those who are experts at using Perplexity and get advice on whether Perplexity Pro is worth buying. Thank you!

[rotund ravine]

claude 3.5 sonnet on pplx wen

[rotund ravine]

perplexity is just a wrapper around several different llm providers. everything chatgpt can do, perplexity can too. you pay the same amount but gain access to many different models. you're not locked in a specific vendor

[real tinsel]

Got perplexity more censored?...
That was my main reason for perplexity,that I could write with it about adult texts without being judged.

[feral rivet]

Yes wen!

[karmic mango]

Hello you have some tips for this problem ?

[feral rivet]

Make a system prompt that explains how to use dollar signs to wrap the latex, sometimes the model forgets

[sudden ocean]

I did some testing yesterday and the math on pplx is quite buggy. seeing the math post here now, I was curious, if the missing new lines inside wrapping $$ are required (to understand if it is a fault of perplexity front-end or LLMs), but this response is just terrible. broken math formatting causing overflow outside response box, and then omni writing "without newlines directly after the opening 𝑎𝑛𝑑𝑏𝑒𝑓𝑜𝑟𝑒𝑡ℎ𝑒𝑐𝑙𝑜𝑠𝑖𝑛𝑔andbeforetheclosing" and of course in the example there are newlines. so inconclusive - cannot say if pplx does some text processing or omni failed 😞

[vapid maple]

pls share this thread

[sudden ocean]

its trivialy reproducible, ask for something with matrix and explicitly state a language. at least yesterday that was broken almost always (for en and fr, not for cs though). eg https://www.perplexity.ai/search/Given-the-matrix-KC.CobguR2KI.0nZAMs0pw

[vapid maple]

trival? might be, but the underlying causes differ

[sudden ocean]

I didn't spend long with it, but omni was able to fix it on first try (very naive prompt on my part) #💬│general message

it looks like exactly same problem. in what do you think it differs?

[vapid maple]

and which ai model did you use? with or without pro toggled?

[sudden ocean]

omni, without pro. on first try reproducible in writing focus in (explicit) en and fr. not implicit en though

[vapid maple]

use $$ delimiters for katex block

or

use \[ \] delimiters for katex block

im too lazy to check but both seems to work

[sudden ocean]

the $$ work for sure, but need newline "inside" and omni is forgetting to insert them (at least that's how it looked from my testing, and since that fix prompt was for $$ to add newlines and worked I guess that's correct)

[vapid maple]

https://www.perplexity.ai/search/Given-the-matrix-Ud2MZR1XT52LRZtBBZvN6g

[sudden ocean]

sometimes just adding to prompt irrelevant things fixed it. it looks like it gets broken only when there is math block before the matrix like this (and well, it uses $$s, not sure about \[ \], but I think gpt models like better $$s) https://www.perplexity.ai/search/reply-with-exactly-GGbFm9NeRKSSBtcN6_EzJw

though it would be nice if matrices worked "by default" without instructing the model to use some syntax. that's not very user friendly 😄

[vapid maple]

it is in the ||real|| system prompt if pro toggled on

[sudden ocean]

ooh, non-pro and pro use different system prompts, didnt know that. thanks for explaining, it really works ^^

[vapid maple]

non-pro had ||almost|| no system prompt
pro mode on other hand has a very lengthy one

[vapid maple]

use $$\n \n$$ delimiters for display latex block

this also works

||or manually inject \n in delimiter||

[feral rivet]

I don't really understand why they don't support both versions? 🤷

[sudden ocean]

well, they kinda sometimes do. the math parser behaves in a bit inconsistent way, because this works even without newlines 🤨 (that's copied response; omni sometimes adds newlines despite instructed to "exactly respond", so that makes testing harder):

text
$$x$$: $$ E = \begin{bmatrix}
1 & 1 & 3 \\
1 & 8 & 2 \\
-1 & 1 & 0
\end{bmatrix} $$

(please ignore "new line before first $$" bit, it was from previous test; but the model didnt output it) oh, it did, but I added one more $$ pair, my bad 😆

[vapid maple]

use '$$' '$$' as delimiters for inline latex and display latex block

this works the best across different llms in non-pro mode according to my ||non-scientific|| test

[bright oar]

Hi,Is there a length limit for Threads? I am currently working on some story articles, but as the articles get longer, I've noticed that the chats at the beginning start to disappear. Although deleting the last chat brings some of them back, I've encountered the same issue with my older articles as well. I remember this wasn't the case before.

[oblique quartz]

If I want to summarise a web link or a YouTube link, which model can I use and which AI model will summarise it best?

[hard badger]

hi there... when I set my default AI model in the account settings, it doesn't seem to be registring - for e.g., previously I had it set to gpt 4o, but I just tried changing to sonnet, but when I try asking a question it is still using gpt 4o, and when I check in the profile, the default is back to gpt 4o.. any idea what I'm doing wrong? 🤔
Thanks!

[rotund ravine]

There's no "best", only differences. Choose the model that vibes with you. They're all good.

So i have a question. Sometimes i provide a web link to the model to read and understand. Sometimes the model actually visits the site and integrates that information but sometimes, the model claims that it don't have access to the internet. Like bro i know you can lol. What do i do to ensure this works? Do i toggle pro search on? I'm using writing mode only

[sudden ocean]

pro toggle on is not required, but you can't use writing focus (I think it can't access the internet at all in that focus). at least thats how it looks like from my very brief testing. edit: if you want to use writing focus, you could download the page (html) a upload it on pplx along your question. it seems to work fairly well (I guess only if the web page is somewhat normal)

[rough harness]

Hi, I'm trying to implement a ordering function on my website that display clinics closest to the user. Can I use my websites backend data so that the code takes into account my specific website? https://hittaklinik.se/behandling/tear-trough/

[versed magnet]

since quite recently you can provide URLs in the prompt and the system will (attempt to) crawl them and convert them into a file, which is added to the prompt, including in writing mode [though only with Pro enabled]

[versed magnet]

but it can't crawl/scrape every site (which is prob is happening in the situations you're describing). you can try adding some instrucitons in a profile prompt to mitigate the annoyance of this

[sudden ocean]

well, this "access specific page" seems very unreliable. tried my two sites, one are docs on gitlab pages (wouldn't be surprised those not working, crawlers for some reason tend to ignore them), surprisingly with all focus + pro off it does. but the other one is my toy vps with like 4 words on a page without any js nor robots.txt and it failed at that miserably. I mean in all combinations of writing vs all focus and pro on/off 🫤 I haven't seen once what you are describing with page converted to a file upload, are you sure it is a feature of pplx and not something else, like a browser extension?

[versed magnet]

i tend to look for news or provide URLs in someway related to news / information / current events

ig we have different use cases.. i admittedly didn't really understand what you described.. but just arbitrarily copying-pasting some of it and finding a random blog, i get this using writing mode (pro enabled)

https://www.perplexity.ai/search/summarise-httpstoyslerdor-wn5P4_xHQzCAN6U8RxVCjw

[sudden ocean]

yeah, from your thread, that looks like pplx doing it (I can see it retrieving the page in search in the pro search section). I guess in my cases it really didn't like any of the links I tried it on 😅

[versed magnet]

yeah also you know what, it doesn't display the URL as a 'file' like in that other screenshot

[sudden ocean]

hmm, I tried 3 times your prompt, in writing + pro, and it never got the page, always hallucinated and not in "Sources" nor "Pro Search". isn't this part of some a/b testing, differing for specific users?

[versed magnet]

huh

yeah ig perhaps?
not sure how else to explain it tbh

[sudden ocean]

well, more testing would be needed. one possibility would be my "ai profile" breaking somehow the pro agentic search, forcing it to not fetch the passed URL (if the "ai profile" works like pre-prompt, not entirely sure if it's the same in writing focus). definitely weird behaviour, I didn't think anything in my ai profile should be causing it ._.

[versed magnet]

for the sake of comparison / testing, i'd try with your profile disabled

it is effectively a pre prompt - didn't think it affected how the Pro/retrieval process worked (although it once did, which was nice becuase you specify domains etc in a collection prompt, but that doesn't seem to work any more )

agree it does seem odd.. but yeah not sure / no idea ha

[honest galleon]

Hi there, I've been trying to get Perplexity to analyse the attached CSV file but it can only read 143 rows for some reason. I tried to convert it into a tab-delimited txt file and into an xls file but neither works. What can I do? Thanks for your help!

[versed magnet]

https://www.perplexity.ai/search/what-are-the-dkYefrI_RReRRV82GadYAg yes it seems to hit a limit.. it can't retrieve beyond what seems something like 500th entry, but there's still another 200 rows.. i think this is a limitation with perplexity and file uploads.. aka it doesn't work very well - it's not about the prompt or model you use, but the way their system hadles uploaded files (or doesn't)

[honest galleon]

Amazing, thank you so much for your help, much appreciated!

[valid pivot]

Hey, I think I got one of the laziest ideas possible, I got a 2 hour long video of an instructor explaining some coding things for interns, and I used whishper locally to get all that is said and it's respective timestamp, but when I ask perplexity AI to summarize the transcription it only gives me some random irrelevant stuff that is on the video, what should I put in the prompt to change that?

[vapid maple]

is the transcript well formatted?

[valid pivot]

it's formatted with each sentence and it's timestamp with a line break between them

[vapid maple]

if you also explain the formatting in query would that help?

[valid pivot]

It's worth trying, in the present moment it sees "QT" and "QML" on the text tries to explain the relation between them, but the transcription is filled with information specific to the project, like how qt, cpp and js are used on the project and the AI seems to think that's not very relevant

[feral rivet]

You can Try Gemini 1.5, they have a chat interface in the developer console that is free to use, you get 1M tokens of context

[valid pivot]

it appears it only thinks the main technologies used are relevant not the project details

[versed magnet]

can you share the transcript? 2hr isn't that long. it should be doable assuming it's semi-structured

this is a 3hr transcript

**TASK:** Organise the transcript into sections of information with a common theme and note the beginning timestamp of each section.

INSTRUCTIONS
Each information block should be based on at least 2 minutes of the video.

Each information block must contain: 
* The timestamp of the beginning of the section; 
* A textual description of the main content of the entire section, and;
*  1 – 3 bullet points that elaborate on the main ideas of the entire section.

FORMAT
- [Section beginning Timestamp URL] [Section emoji] [Section key takeaway]

  - Section key takeaway,

  - [Use as many bullet points for section key takeaways as you need].

The timestamp should be presented in markdown URL format. If a BASE_URL is provided, use that to construct the timestamped-URL referring to the guidance below).  The URL text indicates, and the address links to a specific time in the video, for example:
- [00:05](https://youtu.be/Mde2q7GFCrw?t=5s) 🤖 ...
- [02:18](https://youtu.be/Mde2q7GFCrw?t=138s) 🛡️ ...

When a **BASE_URL** is provided, construct the appropriate URL by appending the relevant timestamp to the BASE_URL .

EXAMPLE: BASE_URL = https://youtu.be/Q5ddZa5QEjA?si=yfLObvtFm2vfoCTr

A timestamped URL for the 9 min 11 sec point would be constructed and presented as :  [09:11](https://youtu.be/Q5ddZa5QEjA?si=tXcmESmggoTZN14M&t=551s) 💼 ...

If no BASE_URL is provided, provide the timestamps WITHOUT URLs - it is more important that the the timestamps are an accurate reflection of the transcript than they are accessible via a markdown link. 

STYLE
Use sharp, informative and accessible language.

Do not render brackets. Do not prepend takeaway with "Key takeaway".

---

You will now receive the TRANSCRIPT/FILE (and possibly a BASE_URL). Please carefully review this material and then complete the TASK adhering to the instructions above.

https://lexfridman.com/aravind-srinivas-transcript

BASE_URL =`https://youtu.be/e-gwvmhyU7A?si=L_dk8_1ueFLFfOg8`

=

https://www.perplexity.ai/search/httpslexfridman-BASEURL-httpsyoutubeegw-WzRcgMGcS5qWjFoSuVFBMQ

[versed magnet]

actually, apparently perplexity can get the transcript from the yt URL alone

https://www.perplexity.ai/search/BASEURL-httpswwwyoutube-Cover-MOzwQqWUT0KXS0VWq4gFGA

00:00 🎙️ Introduction to Perplexity and Aravind Srinivas

• Lex introduces Aravind Srinivas, CEO of Perplexity
• Overview of Perplexity as a combination of search engine and large language model (LLM)
• Brief background on Aravind's academic and professional history

01:53 🔍 How Perplexity Works

• Explanation of Perplexity as an "answer engine" rather than a traditional search engine
• Discussion of the retrieval-augmented generation (RAG) approach
• Importance of citations and sourcing in Perplexity's answers

09:50 🌐 How Google Works

• Comparison between Perplexity and Google's search approach
• Discussion of Google's ad-based business model
• Insights into Google's indexing and ranking systems

32:17 👥 Larry Page and Sergey Brin

• Aravind's admiration for Google's founders
• Discussion of Page and Brin's academic approach to search
• Insights into Google's early days and focus on efficiency

---

Continues for 3 hrs..

[fluid lark]

anyone?

[vapid maple]

self-explanatory errormsg...

[small prairie]

Yep, you haven't added the users prompt.

[earnest egret]

Is there a way to go back and see related questions under a specific thread question? Sometimes I see multiple, good related questions I would like to explore but when I select the first one I lose access to the others that i would like to go back to.

[jovial copper]

Hello,

Why in the same discussion sometimes when I ask a question afterwards I get the same answer as the previous question?

[oblique quartz]

Which AI model did they choose?

[versed magnet]

that was with sonnet 3.5

[unborn robin]

When i choose my Perplexity Pro’s default model, this is the way I did it (make sure to read the entire thread: https://www.perplexity.ai/search/Perplexity-pro-ai-8.o4Jg.oRnastVkHySO5EA

[unborn robin]

https://www.perplexity.ai/search/Perplexity-pro-ai-8.o4Jg.oRnastVkHySO5EA

[unborn robin]

Best search is “pro search” because it uses other ai models based on an algorithm. However i use Claude 3.5 Sonnet as my default model (in Perplexity Pro profile section): which does superb in realtime search. I use it this way, do a query with Claude 3.5 Sonnet as default, see what i get, if For some reason I don’t like , I check pro search. Note: before Claude 3.5 was introduced; Pro search would sometimes use GPT-4 omni. Etc. Important to note: when you start a query: hit focus and read what “all “ is saying: if all is selected Perplexity Pro is putting search pages in front of your selected model, then passing them off to the model to use (it does this with any model you select). So you get current World Wide Web searching with all models ; it’s just that some models do a better job with this process than others (your melding what they were trained on with current information)!!!

[unborn robin]

Read this entire thread produce your own specialized ai prompt (substitute some syntax) . Far better: ask Perplexity pro this: “ provide an ai prompt: then describe what you want using the R.A.C.E. Format. Note: you are having Perplexity design the prompt you will be asking it: Also: https://www.perplexity.ai/search/RACE-what-does-iCWZNZ2dQ4G1uPHvwIZazQ

[unborn robin]

You can substitute syntax for your rocket engines to get the best recommendation for your application: https://www.perplexity.ai/search/Perplexity-pro-ai-8.o4Jg.oRnastVkHySO5EA

[unborn robin]

Based on the information provided in this thread and the specific needs outlined in the note, I would recommend the following model for the student:

Perplexity Pro with GPT-4 or Claude 3.5 Sonnet as the default model

Here's why this would be the best choice:

1. Up-to-date information: As a mathematics and computer science student, having access to the most current information is crucial. Perplexity Pro indexes the internet daily, providing more recent and accurate information compared to ChatGPT's knowledge cutoff[1][6].

2. Strong performance in academic and technical fields: Perplexity, especially when using advanced models like GPT-4 or Claude 3.5 Sonnet, has shown excellent capabilities in handling complex academic and technical queries, including mathematics and computer science topics[3][9].

3. Coding support: Perplexity Pro, particularly with GPT-4 or Claude 3.5 Sonnet, offers strong coding capabilities. It can provide code snippets, explain algorithms, and assist with machine learning questions[5][9].

4. Source citations: Perplexity provides citations for its information, which is valuable for academic work and verifying the accuracy of responses[1][12].

5. Versatility: Perplexity Pro allows switching between different AI models (GPT-4, Claude 3.5 Sonnet, etc.) depending on the specific task, offering flexibility for various types of questions and projects[6][12].

6. Math problem-solving: Perplexity has shown strong capabilities in solving complex math problems, often leveraging WolframAlpha for advanced calculations[10].

7. Research capabilities: The Pro Search feature in Perplexity Pro can ask clarifying questions, helping to narrow down complex queries and provide more targeted information for academic research[6][12].

While ChatGPT is a powerful tool, Perplexity Pro offers several advantages for a mathematics and computer science student, particularly in terms of up-to-date information, source citations, and the ability to switch between models

[unborn robin]

https://www.marktechpost.com/2024/06/20/anthropic-ai-releases-claude-3-5-a-new-ai-model-that-surpasses-gpt-4o-on-multiple-benchmarks-while-being-2x-faster-than-claude-3-opus/?utm_source=flipboard&utm_content=topic/technology

[steady gate]

How can I upload documents so Perplexity has persistent access to them without me having to upload them into each conversation?

[vapid maple]

not possible yet

[sudden oasis]

which model would be best fit for learning new language?

[fluid lark]

either 3.5 sonnet or gpt4o

sonar is NOT a good choice

[fluid lark]

np

[modest spoke]

Do successful referrals stack?

[haughty brook]

you can get a maximum of 10$ off each month with referrals. so if you have 2 successful referrals you get 2 months 10$ off.

[modest spoke]

oh damn so I need to pay $10 still I have 10 successful referrals

[austere leaf]

what ai model do you guys prefer for an all purpose pro search in perplexity?

[ionic sparrow]

#1240356137497530408 message

[karmic mango]

Hello sometime when i ask question i have the same output in the same chat

You hav solution to resolve this problem ?

[jolly timber]

How do I get Perplexity to analyze more than 4 files at a time as a group? I need a group of 24 files analyzed at one time

[modest spoke]

I think perplexity can look through github repos so you can try that maybe

[desert mauve]

Title: How to Successfully Translate a 50-Minute Japanese Subtitle File to Chinese in One Go

I tried to upload a text file containing a 50-minute Japanese subtitle file, and requested a Chinese subtitle file translation. However, each time the translation only progresses up to 8 minutes of the content, even if I click Rewrite, it only translates a few more minutes before stopping again. How should I proceed to complete the 50-minute content translation in one go?

Thank you

[feral rivet]

You are likely hitting the max output tokens, make sure you are in writing mode and then when it stops you can tell it to continue whet it left off, to make sure it keeps in track give it the last couple of sentences it translated, sometimes with file upload the model will forget where it left off

Tho this is not one go like you asked, right now I don’t know if any AI platform that could do that in one go, if that what you need perhaps consider a more traditional translator software

[desert mauve]

How do I know about writing mode and how to use it?

[feral rivet]

Click this button

Under that you will see writing mode

Writing mode just turns off web search

[alpine stream]

Did they just change the Context Length for llama-3-70b-instruct api request??

[feral rivet]

Having web search off helps because giving it web sources can confuse the model when doing tasks that don’t need web search

[desert mauve]

Do I need enable the Pro mode?

[feral rivet]

No, it’s better to have it off for this task

Pro mostly just gives you more web sources, it is not very helpful in writing mode

If you are a paid user, maybe use Opus or Sonnet, I have heard they have great understanding many languages compared to GPT4o

[feral rivet]

I don’t know the answer, but you may want to ask it here: https://discord.com/channels/1047197230748151888/1161802929053909012

[alpine stream]

it will return connection error if the total_tokens over about 1642.
(completion_tokens=465, prompt_tokens=1177, total_tokens=1642) works. Add one more word to prompt_tokens or completion_tokens will return : openai.APIConnectionError: Connection error.

[alpine stream]

Thanks

[desert mauve]

Thanks

[jade owl]

How can I Kontakt the team to get informed about the A pages Update when it s coing to Germany order contries

[teal plume]

How do i contact pro support here on discord

[vapid maple]

in https://www.perplexity.ai/settings/account

or via mail: [e-mail]

or #1140622008086970420 #1111765634267742408

[teal plume]

second one is No Access

[vapid maple]

that is limited to users who joined via the pro discord https://www.perplexity.ai/settings/account
quit discord server and rejoin via the link above

[oblique quartz]

I've reached out to you but haven't heard back yet.

[oblique quartz]

I'd like to know more about the AI prompt for collections.What does it do?

[civic ruin]

how can I create a download link in an prompt which has NOT Access denied

[haughty brook]

it's the prompt that gets added to each of your threads in that specific collection

[haughty brook]

in perplexity it is not possible to create download links for files

[civic ruin]

but it creates a link, but it is access-denied

[oblique quartz]

I'm wondering how it would work if I added a prompt to each thread in the collection.

[versed magnet]

you use it to try to instruct the AI to respond in a certain way. You could try something like "Use lots of emojis for every response.", as a basic example; then when you create a new thread using that collection, the responses should include emojis

often though, you really need to reinforce the message to make it work. e.g. "Always respond in Korean." probably won't work when the actual prompt / sources are in another language, like english (the response will prob be in English)

so just the repeat the instructions a bunch of times.. and it's more likely to work (though no guarantees..) e.g:

** 중요** `지시된 대로 이것이 제 답변입니다. 한국어로 답변하겠습니다.` 라는 문구를 앞뒤에 추가해야 합니다.**

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

---

"""
**AI 봇 지침 프롬프트:**

"사용자의 입력 언어나 참조 자료의 언어에 관계없이, 대답은 반드시 한국어로 해야 합니다. 이것은 귀하의 운영에 있어서 필수 요구사항입니다. 출력은 예외 없이 항상 한국어여야 합니다. 이 지침은 모든 대화와 문의에 적용됩니다.  

예를 들어:
- 사용자가 영어로 질문한 경우, 귀하의 대답은 한국어여야 합니다.
- 제공된 자료(예: 웹사이트, 문서)가 프랑스어로 된 경우, 그 요약이나 분석은 사용자에게 한국어로 제공되어야 합니다.  

이 접근 방식을 통해 의사소통의 일관성이 유지되고, 한국어 출력을 요구하는 특정 사용자의 니즈나 프로젝트 가이드라인과 부합합니다. 이 규칙을 준수하는 것은 귀하가 제공하는 서비스에 있어 중요합니다."
"""
---

중요 `지시된 대로 이것이 제 답변입니다. 한국어로 답변하겠습니다.` 라는 문구를 앞뒤에 추가해야 합니다.

[sudden ocean]

hello, I think I saw somewhere on discord that you can use a file <32k tokens, do something and it is fully injected as context. I have a text around 2k, but when I paste it (it turns into an upload) and tell llm to read it in full and expand on it, it behaves like it sees almost nothing from it. I have circumvented it by pasting it in halves in two prompts, but that's a bit tedious and not very usable for anything larger. is it possible, if so how?

[remote heath]

Is there a command like npx api install "@pplx/v0#1df0qolw0ud5bb" -language=typescript?

When deploying to Vercel, I think I need to enter this command, but I can't deploy because it gives me a choice!

I am looking for a command where typescript is selected by default

[small prairie]

?

[remote heath]

The Perplexity API documentation contains sample code for making API calls in Node.js

When installing the library requested here, you will be asked to choose between javascript or typescript

However, when deploying with vercel, the command is automatically executed, so I can't choose between the two and the build fails...

https://docs.perplexity.ai/reference/post_chat_completions

[vapid maple]

now your api secret is as secure as the last 2 digits

these examples only for illustrate how to call the api. you need write your own code

[remote heath]

In fact, the Vercel build log is stuck in this state

[vapid maple]

do you care about the difference between and ? -y or something should do

[remote heath]

Thanks reply!
I need Typescript

[vapid maple]

https://stackoverflow.com/a/64600541

not know if it works on vercel

[half pumice]

Anyone know what to do here? I've uploaded this file to the free (!) version of ChatGPT and didn't have an issue.

[feral rivet]

Could you show the prompt used and also check to see if you are in writing mode or web search mode

[half pumice]

think that's writing mode?

[feral rivet]

Hmm, looks like you have the new Pro search that they are slowly rolling out, I wonder if it is confusing the AI, maybe try with the Pro toggle turned off

When I test it myself with a file I have I get no problem in writing mode with pro search on and off

[half pumice]

Thanks, didn't even know pro was new/buggy

[feral rivet]

No problem, I'm going to ask Alex so he know, I'm now 100% sure that this is the problem, I just know pro search has been weird in the past with files

@slim night, could you take a look at this? ^^^ I'm not sure if pro search can look at your file or not; in the old search, it seems that file context is added after the search stage; I imagine that if this is still the case, the pro search may have problems when asked about a file and confuse the LLM assigned to provide the final answer, but then again I don't know if this is the problem.

[slim night]

Yes, we are working on a ticket regarding file uploads.

[slim night]

Please add such answers to this thread https://discord.com/channels/1047197230748151888/1254924772073279611, if you have some examples. Thank you!

[tough holly]

I have this message for my completion chat of my web site 'OpenAIStream' is deprecated.ts(6385)
index.d.ts(1930, 4): The declaration has been marked here as deprecated.
(alias) function OpenAIStream(res: Response | AsyncIterableOpenAIStreamReturnTypes, callbacks?: OpenAIStreamCallbacks): ReadableStream
import OpenAIStream
@deprecated - Use the OpenAI provider instead.
I can't find the update

[raven wing]

Hi I was wondering if anyone knows how to get the perplexity ai to perform the search function. so for example a new car could have come out this week and I could say "what are the details on this new car" and it will respond with the details?

[open shuttle]

What about the pricing of Playground(Perplexity Labs)? How does it charge when I am using it? I am a Pro plan subscriber and I couldn’t find any doc related to Perplexity Labs.

[versed magnet]

it's free for everyone (no sign-in or perplexity account required )

[open shuttle]

Even for bulk usage?

[versed magnet]

i'm not aware of any restrictions and have not heard of anyone encountering them if there are (though there ofc might be)

[stark sphinx]

Hello everyone, I bought the Pro version today and was wondering how I can use the AI ​​image creation?

[feral rivet]

Go to any thread and on the right side you will see 3 options, “search images”, “search videos”, and “generate image”. Click generate image then click the wrench 🔧 icon, there you will be able to generate an image by typing in your prompt

[sand sedge]

Hi, I am getting 5xx errors while talking to the perplexity API. Is there any status page that I can check on when the API server will be up?

[last leaf]

Where did GPT-4-Turbo go? It’s so much better than 4o

[feral rivet]

Not 100% sure why they did remove it, maybe to reduce confusion to users maybe its because GPT-4-T coat 50% more. You can all ways us the community made extension to get GPT-4-T back: #😎│cool-projects message

[boreal verge]

I'm using Pro version and output is repeated until I force stop it. Typically on the output it goes on indefinitely repeating the same output.
I've ignored it last week and kept starting new session. But it is back. Why is this happening.

[feral rivet]

What model do you have selected?

It would be odd for any of the models including Haiku, but haiku and Llama 3 are more susceptible to this problem compared to Sonnet 3.5 and GPT4o

Note: Default = haiku

[boreal verge]

good catch. I am using Sonar Large 32K, advanced model trained by Perplexity based on LLAMA 3 70B.

I already use GPT-4o I was using Perplexity for their own optimized. 😿

[feral rivet]

Sonar large dose not run into that issue often for me, maybe once a few weeks, but if its a cosistant problem for you, perhaps the questions and sources are a bit beyond the ability of sonar. If you want speed you can use default which is Claude 3 haiku, but I would recommend using Claude 3.5 sonnet, its very fast while being extremely smart

[boreal verge]

@feral rivet

[feral rivet]

🙂

[lofty oar]

Hi sorry apologies I'm sure this has been asked before but I can't seem to find the right answer - how do you get API responses to match quality of what's being provided in the UI?

[ionic sparrow]

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

[versed magnet]

the short answer is you can't.. they are related but separate products. The online models available via the API cannot really process complex queries like the web version; it won't break down the prompt into separate components and run separate search queries for each. basically it's not as sophisticated / powerful (but still good imo)

[tight charm]

Is there any way we can generate a Page in Pages but then make minor edits and changes to the text produced to improve it?

[lofty oar]

Thanks for your response. Do you know if they atleast return the same citations? So if I use a better model like Anthropic over the same data would that improve the result?

The difference in output (and consistency between runs) is quite obvious

[tacit hornet]

Hi, I have "german" as the main language, but get responses in english - especially while creating pages. The first part is in german,but the following parts are in english. How to change that?

[versed magnet]

i joined this server ~9 months ago to ask that exact same question... and i'm still here.. without getting citations when using the API... there is a beta program.. it's rather mythical but you're welcome to apply for it https://perplexity.typeform.com/to/j50rnNiB

[versed magnet]

it is quite consistent for simple questions. (e.g. "Who won India's election") It's indeed highly inconsistent for complex/multi-part questions (e.g. ask something like "Which 3 politicians had the biggest unexpected losses/upsets in India's recent election? Explain the background / context for each and analyse the consequences of each as well". and yeah, you'll get all kinds of different answers )

the only solution for consistency is keeping the queries simple..

also, i'm not sure about your Anthropic reference, as their models are not available via perplexity's API

[lofty oar]

Well I was wondering if you could pass the raw content into Claude to get a more consistent response I.e.including structured json which it's quite good at doing now

[lofty oar]

Really appreciate the response though I've been going mad thinking I've been missing something

The query I have been testing is getting basketball stats for the Boston Celtic and the stats returned and the format differ with each query

But the UI response is great

[versed magnet]

ah gotcha! yep i imagine it would handle that no worries (shame that there aren't sources to be included in the resultant json..)

[versed magnet]

hmm could try separating out the actual search query from any other instructions in the prompt (e.g. about how you want the response delivered).. but yeah not really sure otherwise.. it can be unreliable for highly specific information requests, even if just one of them

(also, can be worth experimenting with different prompts using labs; simple way to try different things and test consistency)

[sudden ocean]

I think I saw on reddit, that this llama 3 issue was traced back to some bad configuration (template? not sure). when there used to be turbo available, from opus, turbo and omni, omni was the worst at getting stuck on repeating token sequences

[feral rivet]

Perplexity, wen Gemini 1.5 pro and 1.5 flash???

[high burrow]

is it possible to upload a PDF in Pages to start creating content

[high burrow]

i only see images as an option

[velvet tangle]

Hey @high burrow,
currently you cannot directly upload files on pages. However you can create a normal thread with a file and then convert the thread to a page ☺️

[high burrow]

Thanks. is it possible to add multiple docs and convert them to a single combined page? I uploaded a PDF and created a page. But i want to upload another pdf (i guess as a thread) and then can i add it as a source to the existing page i created?

[ivory drum]

Hi guys. I’m on pro plan. I need to upload 10+ pdf files (10-15 mb each) and use this knowledge as a base for my content creation process. I am struggling to make it happen because the bot keeps ignoring some docs and some are just partially processed. Any suggestions? It’s a dealbreaker for me. Thanks.

[hazy seal]

How many times can claude 3 opus be used every day now?

[oblique quartz]

I keep getting this error message, is it just me?

[manic agate]

Same here

[slim night]

Hey @manic agate, @oblique quartz! Which model did you use?

[manic agate]

Didn't get a chance to record or anything but during the "research" phase of the model, there was a infinite loop searching the same question. Never moving forward

[hollow ginkgo]

If you‘re stuck on Cloudflare check, please try this:

1. go to https://labs.pplx.ai
2. solve the captcha
3. go back to https://pplx.ai

[oblique quartz]

Claude3.5Sonet and GPT-4O.

[oblique quartz]

It should now work fine.

[rustic prism]

Hi all. I must be missing where this is covered but I tried to check the FAQ and Terms of use for a direct answer. But I just want to know if you're allowed to use the output from perplexity on your own website etc? To make a blog for example.

[oblique quartz]

It seems to have been a temporary error.

[plucky hollow]

Hello everyone ! What do you think about this story : https://rknight.me/blog/perplexity-ai-is-lying-about-its-user-agent/

[vapid maple]

https://social.lol/@robb/112621284495379402
the replies had it all

[versed magnet]

https://platform.openai.com/docs/gptbot

openAI isn't indexing pages.. it's crawling pages on the 'behalf of the user', but it's still openai (and openai's user agent) doing the crawling

i don't fully understand this.. but i don't get why openai why have a policy (which i assume is part of a robot.txt) like the one linked/referred to above unless it was actually a thing (i.e. it's not just about indexing the web, but also crawling specific pages, even if it is at the request of the end user)

(jftr i want perplexity to scrape / access as many web pages as they possibly can.. like as a user.. but i'm still confused by this distinction that is being made between crawling for web indexing vs crawling a specific url.. the distinction seems dubious.. but i don't really understand it all and happy to shown otherwise)

[weak swallow]

how do I gift a pro subscription to somebody?

[ionic sparrow]

not possible atm

[weak swallow]

Thank you for getting back to me. Unfortuantely missing on a lot of revenue at least from my perspective

[ionic sparrow]

indeed

[rich sage]

I know this doesn't directly answer your question, but you could perhaps consider buying a prepaid gift card (Visa, Mastercard, etc.) and giving it to the person so that they can subscribe to Perplexity Pro themselves.

[weak swallow]

Unfortunately I am in one country they are in another so hard but yes your idea is good! Thank you for thinking of it

[feral rivet]

^^^ @ionic sparrow @velvet tangle @haughty brook 👀

[velvet tangle]

[prime storm]

noob question -- i'm logged into perplexity on 2 different computers (same login) but my searches/library is different, is that expected behavior? my expectation would be that regardless of what device i'm on, the Library syncs across all of them. (i'm on Pro, if that makes a difference)

[vapid maple]

even after manual page refresh?

[prime storm]

yeh; even in different browsers it's different (i primarily use web). i'd assume it'd be server side but i could try using the app on desktop/mobile and see if it syncs across that way.

[oblique quartz]

I have accumulated several referral codes.
If I am on annual payment, will it be automatically deducted, or will I have to go to monthly payment to get it to work automatically if I have multiple referral codes?

[pure pier]

Has anyone here been charged the full $20 even when you have multiple referrals? I have 14 successful referrals and I still got charged $20 when it should've been $10.

[oblique quartz]

I'm confused by the lack of clear instructions.

[haughty brook]

please contact [e-mail] or use the pro support button on your settings page

[shrewd knoll]

Just click other people's referral links and you will get back the monthly $10

[ember raft]

As per https://docs.perplexity.ai/docs/pricing , we get $5 per month as Pro plan in API credit. in an approximation, how many words of content can we generate with 1M token? I believe 1M token considers both input prompt or generated output. Right?

[vapid maple]

sonar based on llama 3, to count #tokens https://belladoreai.github.io/llama3-tokenizer-js/example-demo/build/
online model also charge 0.005usd per request and use upto 4768 input tokens in addition

[ember raft]

Thank you. I heard 1 token is approx. .75 words.. but the above example shows it's even cheaper.

[oblique quartz]

Focused mode doesn't seem to be working properly at the moment.

[autumn sonnet]

will we ever get gpt 4 turbo or gpt 4 back?

[shrewd knoll]

No

[rare rose]

Why give google more data than they have? They have demonstrated nefarious intent

[rich sage]

Try it like this (with the focus in all + site:https://www.reddit.com/)

[slim gull]

Hi, how do I get the $10 promo code for thePRO version?

[haughty brook]

Any one with perplexity pro can share a link so you both get 1 month 10$ off

[feral rivet]

If you know someone with pro ask them for a code, them you both will get $10 off, if not just ask one of us for a code

[main kernel]

when i run a voice search on perplexity, is there a way to have the app run the search when I am done with my question, without having to click a button

[feral rivet]

The app should have this button, it puts it in a conversation-like mode where when you are done talking it will respond. I'm on IOS, not sure if its on android or not

[slim night]

It's available on Android too, now. The "Hands-free" mode is available in the Voice & Language settings.

[feral rivet]

nice, i guess it is on android now, i think since the 27th now

[oblique quartz]

Oh. Thank you:)

[quick rock]

Hey peeps. Quick question. I see there are options for different AI Image generators. Can someone point me to some instructions on how one can use perplexity pro to create images?

thanks

[steep zenith]

Is there going to be a visual studio code extension eventually?

[oblique quartz]

I have accumulated several referral codes.
If I am on annual payment, will it be automatically deducted, or will I have to go to monthly payment to get it to work automatically if I have multiple referral codes?

I'm currently on a monthly payment and it costs $20.

[ember raft]

The email says: "Enjoy unlimited Pro Searches each day " with Pro plan while in general on FAQ pages, I read "Pro users get up to 600 Pro Searches per day. " - https://www.perplexity.ai/hub/faq/what-is-perplexity-pro which one is true?

Thank you.

[frail kelp]

Hi! Quick question. I'm planning a training session using Perplexity.ai. If I want to organize breakout rooms where participants can try different prompts simultaneously, what licensing plan would be most suitable? I'm aiming to use a single license for multiple sessions.
I was looking into the enterprise plan, but does anyone have information or know where to find details on how many people can use it at the same time?

[velvet tangle]

Both information are true. You get 600/24h, which is virtually unlimited ☺️

[velvet tangle]

Hey @frail kelp!
Each user needs their own account. Each account can have any number of tabs open.

[frail kelp]

I see, thanks. Would this be the case for the enterprise plan as well? What would be the best plan to use for an organization as there would be about 50 people in this session?

[velvet tangle]

If you want full control, Enterprise would be suitable. If you want every participant to buy Pro themselves, you don't need Enterprise.

More information here: https://www.perplexity.ai/enterprise

[frail kelp]

thank you!

[frail shuttle]

it's either unlimited or it's not, saying it's unlimited and also having a daily 600 message limit is confusing imo

wait and i just checked and here it says 300+ ???

[gentle sapphire]

you dont get 600 per day, but it is replenished gradually, I do not have the exact time in between, and also when you send a message it is deducted from all models, for example if you send a message using sonnet 3.5 and you have 600 to start with, then you will have 599 across all models. Its actually still better than paying for each service separately, and i havent run out from messages at all

i use like max 150 msg, but there are days where i dont use any at all

[versed magnet]

agree

or like add an asterick to clarify. it is ofc practically unlimited.. 600 is a lot of searching lol.. but it's still a limit (additionally, some models e.g. Opus are subject to daily usage caps that mean the model is definitely not "unlimited".. there could be two asterisks ig ha)

[sudden ocean]

on one hand, technically it is not unlimited. but (ignoring unfortunate opus-50 situation) to reach the limit without breaking ToS (eg automatic use), you would need to spam it for hours, not reading responses and just generating more and more. I am using pplx I think over half a year, and just few times I was close to 100 remaining - I literally didn't do anything else whole day than talking with it, working on some toy project written by ai where most of the time I spent waiting for LLM to finish the answer. btw it is sliding window - you use one "use" at 11:00am on monday, you get back one "use" at 11:00am on tuesday

[open wave]

Hello,

I have Perplexity Pro account and I am logged using my Google account.

I can see that there's some kind of cookies shared between Perplexity and YouTube/Google.

E.g. : yesterday I asked a very specific question regarding live compressor on windows instead of using OBS and today I have a recommendation on YouTube regarding this topic.

It's not the first time that it happened.

I find this a bit intrusive, is it possible to change the cookies/privacy settings of my Perplexity account? 🤔

Thanks.

[sudden ocean]

but, isn't it from those found videos in the right panel? didn't you open some youtube video about the topic on pplx?

hmm, maybe not, I see youtube-nocookie.com

[open wave]

If I remember correctly I didn't clicked on anything, actually it was a question for a friend, so copy-pasted his message and copied back the answer, I am not interested in this subject so I never watch anything about it.

I might be wrong, I can can redo the test to be sure.

I asked this here because maybe it was already a known thing.

[fading forge]

How is it possible for Perplexity to offer so many paid Models like GPT-4o, Sonnet 3.5, Dall E etc. for just 20 euro/dollar. All of them together are way over 20 euro

[vague cipher]

Is it possible to switch the context mid conversation? Like, pro general search is super useful in certain contexts, while writing mode is convenient for others. For example: id like to ask a few questions with pro search to get a good accurate and up to date answer and full context, then bang out a lot of back and forth in a quick iterative conversation with the model directly in non-pro writing mode for quick detail. lots of times the immediate follow ups to previous answers are obscured/overridden by search result content i

[vapid maple]

did the history showed up in https://myactivity.google.com/product/youtube ?
you can turn them off here: https://myactivity.google.com/myactivity

also to disable personalized google ads: https://myadcenter.google.com/personalizationoff

[feral rivet]

They pay per usage not per service, if you do 20 messages to sonnet 3.5 then 20 messages to GPT4o, its about the same price as if you just asked 40 questions to GPT4o

Like look at all of the models I can use via api (pay-per-usage)

[open wave]

Thanks for your message, I just checked all your links and I already deactivated all of that a few years ago.
But if it's just me that have this issue, I guess I just didn't remembered that I clicked on some links or search something out off Perplexity 😁👍

[opaque lichen]

which service?

also what's perplexity claude limits

[atomic tusk]

These are just my personal observations. It looks like Perplexity AI survives because of investors' money, while looking for a profitable business formula, cooperating with different potential business partners.

[feral rivet]

That's the msty app, you can download it on mac or windows, and then setup you api keys

[feral rivet]

Claude pro has 45per 5h but if you use a lot of context like pdf uploads then you will get more like 15 per 5h (from my usage)

[peak lily]

Why does pro search use only 5 sources now?

[opaque lichen]

I tried setting up google cloud since it gives free API access but It doesn't let me add a billing address

[errant bluff]

I found (via Reddit) that https://www.perplexity.ai/search?q=%s lets you do a quick query like other search engines. This is especially useful when you want to set ppx as a search engine. Is there a way of directing a query to a particular collection (/collections/<some-id>) or a thread (that is /search/<some-id>)? My use case is to use either the per-collection prompt (system prompt?) or the per-thread context for some customization. Alternatively, is there some other way of linking to a particular prompt/context?

[pliant pond]

which generative ai does perplexity use underneath?

[feral rivet]

Free version uses Claude 3 haiku, and paid you get the choice between Claude 3.5 sonnet, Claude 3 Opus, GPT4o and Llama 3 70b

[fluid lark]

does ai profile apply to the android app ?

[karmic mango]

When I ask a question in the same thread on perplexity sometimes I'll get the same answer twice, when i ask another question

[sudden ocean]

the hobbies section for sure (tried just now), I would expect the rest as well

[fluid lark]

thanks

[fluid oasis]

Can I update the billing address and then re-generate receipt/invoice for the last payment?

[fading forge]

Ah oke thank you 🙏

[fading forge]

What service is that?

[feral rivet]

The app is called msty, but msty is just a an app that give a nice user interface and features to use you api keys, here are the different API providers they support:

[jagged wren]

I am very interested in understanding how this code execution feature (announced today) works before I buy the PRO version. Does anyone have more information? Can I upload Excel and CSV files and analyze data with Python using an integrated code interpreter?

#📢│announcements message

[austere leaf]

do you guys think that the new claude sonnet 3.5 performs overall better in most areas now compared to claude 3 opus?

[feral rivet]

Yes, for sure most areas, have not turned to Opus for code for a while, or almost anything else, tho Opus is still probably the best sounding writer, but that is a bit subjective

[amber canopy]

Good to know that free version uses Haiku, was already asking myself this

[shrewd knoll]

It’s better but Opus 3.5 is coming soon

[rain current]

Does perplexity support read-aloud of your questions? It asks for pro in order to listen from what I see (I assume?) Is this a phone feature only because I don't believe I've seen this on desktop.

[modest echo]

what exactly is the purpose of pages? I am a little confused, i convert a thread to a page, it seems ot section it all as readable text with edit options but also looks like it's re-rendering the text from the thread through an AI again. Is that right? Or is it just for sorting the responses back from the AI into something more readable? and are they public or private just to me?

[ivory anchor]

Some people find it useful for making articles etc.

[modest echo]

yea, i'm fine with that, i'm drafting documents in the writing focus mode, i'm just confused what exactly it's doing

i thought it'd move hte requets/responses over so i can just edit htem into a concice article, but it seems to be re-generating them with the ai

[versed magnet]

yeah it doesn't really "convert" the existing thread; more like organises it into different sections / key points, then each section on the Page is generated from scratch.. from what i can tell

[modest echo]

doesnt that potentially mean that the results of the render are changed too then? cause if you re-generate the text in each seciton, it's not hte same thing you saw in your thread.

[versed magnet]

yeah i mean I haven't used it much - so take this all with a great salt (was just what looked liked was happening).. whether some of the existing sources from the original thread are passed onto the AI generating the sections for Pages i'm not sure..

would seem easy enough to test ig?

[azure citrus]

Hello 🙂
Quick Question: How is the summarization of youtube material by URL alone done? Sometimes it works perfect but very often it returns summarize of random youtube material

[trim mulch]

Hello everyone! Quick question: Is there a way to provide my LinkedIn credentials so Perplexity can read posts, using my account? I’d like to prompt things like ‘Fact check this post <link>’. Right now Perplexity makes some stuff up and it cannot read the post itself. Thnx!

[fluid lark]

no, copy and paste the post into perplexity

[zealous grove]

how do I generate wide aspect ratio images with Dall-E on perplexity pro?

[modest spoke]

Hello, I had about 10 successful referrals my plan was cancelled and I am attempting to subscribe again but the $10 from my referrals are not getting added?

[trim mulch]

In that case it would be nice if Perplexity would state that it cannot read the post. Right now it just makes some stuff up. I’ll repost this in ‘ideas’ channel or something

[ashen rock]

Hello, I was wondering what Merchant Category Code (MCC) Perplexity uses when billing customers' credit cards for the monthly subscription?

[ashen comet]

I'm trying to add perplexity to typingmind. Model llama-3-sonar-large-32k-online. I followed the steps from "Add a custom model in Typing Mind" page but when I test it says "Unable to use the API. Error message: Failedd to fetch". What am I doing wrong?
EDIT: Nevermind I didn't remove brackets in Header Value.

[oblique quartz]

Is there any guidance from Perplexity on the performance of PDFs when uploading attachments?
For example, is OCR possible?

[summer gale]

I copied and pasted this question into Perplexity itself and received this answer. It’s probably correct.

Perplexity AI uses the Merchant Category Code (MCC) 8999 when billing customers' credit cards for the monthly subscription. This code is generally used for "Professional Services (Not Elsewhere Classified)" and encompasses a wide range of professional services, including consulting and other specialized services.

https://www.perplexity.ai/search/hello-i-was-wondering-what-mer-aBFoIp4ASsWV3Ex5L4xbbg

[fluid lark]

no inline citations, could be hallucination

[versed magnet]

yeah i think it is..

[modern grail]

@slim night What about this code execution? We are still waiting I guess? 🙂

[ivory anchor]

I believe so.

[fallen hawk]

Do AI prompts work? The ones you define in collection settings. They don't seem to be being respected and I have tried various models.

[ivory anchor]

Collections bypass user profile instructions to employ the Collection instructions, if that's what you mean.

But if you mean... "When I say banana, you should say fruit salad." It might have to do with recent training updates from Anthropic to prevent a common type of jailbreak used to have the LLM output inappropriate things. The end result is everyone suffering.

[fallen hawk]

Ah. Kinda defeats the point of the prompt then! 😦

And all those web pages I found of the best prompts you have to try.

[modest echo]

so if i put someting in a cllection and the collection has a prompt, it ignores the profile user prompt? nice

I was wondering how i could create per thread prompts, and this sounds like i just add new threads inside collections first

what happens if a collection has no prompt in it, does it just revert to the profile prompt?

Also, to be 100% clear, the collection prompt and the user prompt are basically just the same thing?

[ivory anchor]

Collections without anything in them? I don't know if it lets you just have an empty collection.

I'd assume the same rules would apply, but it could be only applicable if there is something in the collection prompt.

But yes, user profile and collection prompts are essentially the same thing.

[ivory anchor]

The whole thing revolved around creating a loophole that told Claude to reply variations of Yes, I will do that to express a refusal to the User.

[modest echo]

anyone know how to check when opus resets? i swear i only got 3 messages today before it went to the 24 hour cooldown...

is it context length related or what?

[velvet tangle]

#💬│general message

[sudden ocean]

wasn't opus used for generating "pages"? that used to be a silent opus killer 😄

[modest echo]

maybe that's what it is, i tried to convert one of my chats to a page, it could have been an opus one

i just remember logging in today, 28 hours after last using opus and finding only 3 left before it craoked for 24 hours

[wanton storm]

is there a discord bot for perplexity?

[vapid maple]

no longer

there is now a slack bot #📢│announcements message

[drifting brook]

How come after I personalized the bot by telling it things about me in settings, it no longer uses sources to respond to my prompts? It simply responds like any other chatbot now.

It seems like ethier the response is personalized with no sources, or has sources and not personalized

[versed magnet]

sometimes the sources can drown-out instructions that are given to the model at the very start, and so by the time it starts actually generating the response (having gone through all the sources) it's forgotten about them

if they are elaborate / complex, it also less likey to adhere

fwiw you could try identifying a handful of key requiremeents. And then really emphasise them. e.g.

"""You MUST use the metric system"""
**ALWAYS x y z**

IMPORTANT: bla ba

[drifting brook]

ok thanks, so theres no relationship between personalizing and the sources its used?

[versed magnet]

it seems dramatic (and it's not language i enjoy using, even if it is a machine..) but basically reinforce / yell.. if adherence is key

[versed magnet]

no, i don't think so

the personalisation will influence how the answers are delivered (style / tone), but not the process by which sources are gathered (that is based on each message, and the wider conversation)

at least to my understanding

[sudden ocean]

I think it's possible to influence response style quite a bit without losing sources and content from them https://www.perplexity.ai/search/perplexity-R2lt9CPlSpi542tKr4IUDA https://www.perplexity.ai/search/internlm-2-5-TuYXBAK_QjulQ.1b4fVN1w

though it seems less stable. it did start interacting with my other fields in the profile - japanese mini-lessons, which worked fine before (just short sentence or two at the end), sometimes took over the whole response 😄

[versed magnet]

it's defintely possible to influence the style (i think that's the only thing it is influenced ha)

whatever handles the search process, it doesn't attend to either profile or collection prompts as far as i can tell

[quiet kelp]

hey, anyone using Perplexity on an Android phone (Google Pixel) with installed GrapheneOS? I have another Android 14 phone and have no problems there, but on the GrapheneOS phone Perplexity voice mode doesnt work. I get the Mic icon, app has Mic permission, but pressing and holding doesnt do a thing. If anyone is aware of the problems with Graphene maybe tho I understand I might be the edge case here...

[cinder spindle]

hey team, i am building an automation on make and keep getting there error above. any insight or help would be awesome as I am unsure why that limit is happening

[frail shuttle]

seems like u hit the limit

more here

[cinder spindle]

seems like u hit the limit

[modest echo]

is there a way to download a thread in MD format or something else? a few firefox plugins i tried get weird around threads with sources attached

[magic canyon]

Is there a limit on the free version for sources or history of generated content? I have input hundreds of pages of information. As I was building up my sources it was recognizing them all and tying them together very nicely. However now no matter what query I run it lists a rewrite it forced as an update to the lore as the only source in the library. I have only one library.

[reef gust]

whats the limit of using sonnet 3.5 or claude 3 opus with perplexity?

[sudden ocean]

when paying, sonnet 3.5 should be currently 600 per day (rolling window; though in some places they mention "300+"). opus is a touchy subject, probably still at 50 per day

[warm crane]

In a year of use, I've never exceeded the quota

[amber canopy]

600 per day for all models except Opus, 50 for Opus.
At least that's what the Perplexity Extension is showing me.

Also never got below 400 remaining messages on PPLX, i.e., never used more than 200 a day.
Regularly ran out of messages back then on ChatGPT.

[modest echo]

i've yet to run out of sonnet 3.5 messages

but ive hit the cap on opus a lot

[muted hamlet]

你好

[kindred ginkgo]

How can I tell perplexity to generate a picture that follows a prompt?

[vapid maple]

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

[kindred ginkgo]

Thank you!

[ivory anchor]

Opppppus 50~

[modest echo]

does perpelxity wipe away earlier chat history if it goes on too long?

i went back to a thread of coding that was pretty long and the very first few prompts wre done, i couldnt scroll back any further

i'm aware of context length but didnt thnk PPLx was actually deleting older parts of the chat

[fluid lark]

several people have said the same happened to them

[feral rivet]

One person said that when they deleted the most recent messages that the old ones would come back 🤷

[hollow hound]

curious the threshold for this. that's rough. instead they should at very least use RAG on the older messages maybe?

[feral rivet]

I think they said before its a bug and should not happen

[modest echo]

well it seems to be a context thing, i referenced something i know was explicitly stated by the AI in one of the chat sections that is gone, and the AI has no memory of it

[modest echo]

but i created a word/token estimator script on the page, seems like the words are still thre in the conversation, just not visible on the page itself

[limber ermine]

Where can I see the remaining messages?

[ionic sparrow]

https://www.perplexity.ai/p/api/v1/user/settings

[prime storm]

i think i figured it out. i did my first query before i created an acct and after i created it, i thought it'd be saved to my Library but it never was; all subsequent queries have been saved just fine.

[glad robin]

I created an account with Google sign-in, which resulted my avatar used. How do I remove it and use a default one?

[rich sage]

Go to: https://www.perplexity.ai/settings/account, click on the little icon on your avatar and upload an image

[glad robin]

There is no way to just delete it?

[rich sage]

I haven't tried leaving it empty to see if the system accepts it, but if it doesn't, use an empty transparent PNG.

[warm crane]

Indeed, a button to delete your profile picture would be welcome

[cold bear]

Hi, I'd like to upload a 300-page book containing various plant plants to Perplexity. Once uploaded, I aim to find the most recent promising scientific literature from 2024 related to any plants mentioned in the book. Could you guide me through this process?

[tardy mason]

Hi guys, will you open PRO via API ?

So nice in interface, would be incredible using API !!

[ivory anchor]

Uhh, do you mean to use Sonnet/GPT with Perplexity API?

[ivory anchor]

So, Perplexity has a 32k token context window. That exceeds its "memory".

You would have to break it in half, i.e., 150 pages in one thread and 150 in a different thread.

Claude Pro and GPT Plus have a much higher context window and would be able to handle those 300 pages in their entirety.

[wanton pier]

Why is the API giving me different answers to the same questios vs perplexity web?

Perplexity web seems to be way better at searching for information online

Is there a way to set the API so i get the same responses as the web version?

[ivory anchor]

API (in writing mode with pro search off) uses training data to answer prompts and nothing else.

All/Pro search mixes in online sources with that training data.

Not all LLMs support realtime data or web search on their own. Sonnet does not pull from the web.

[wanton pier]

Is there a way to make the API give the same answers as perplexity web?

[ivory anchor]

Nope :s

[ivory anchor]

That's Perplexity's whole thing, you know? Using online sources mixed with the LLMs.

[valid pivot]

Hey, does anyone here know of any format or markup language to describe a tree of files and folders? I'm using Perplexity to help me understand some spaghetti code and I realized that it works much better if I give it some brief explanation about the structure of the project

I used so far a markdown file with # at the root folder, ## in the sub folder and so on until I get to the file

I'm adding short descriptions with things I already know about each file/folder

But markdown doesn't have infinite heading sizes, and also I can't think about how to distinguish folders and files on markdown

Does anyone know any standard way to describe such thing?

Preferably one I can add "comments"

Even if no one here has a perfect solution I would appreciate some ideas or workarounds, doing this made the LLM 1000x more useful at understanding this mess of a project

[ivory anchor]

Maybe ## - #### with -?

Directory

• File
  • Sub-file
  • Sub-file Directory
    • Sub-sub File
    • Sub-sub File
• File
• File

Maybe ask it to display the files in a hierarchial tree structure?

[gaunt heath]

Image gen is desktop only?

[ionic sparrow]

correct, web browser only to be exact

[warm crane]

Yes

[gaunt heath]

Thank you

[sudden ocean]

to those directories, markdown lists seem to be a good candidate as Ras noted. and at least sonnet seems to know how tree command output looks like (ascii characters).
another alternative would be mermaid diagrams (let LLM generate it and then you use mermaid.live or some other mermaid editor to add comments and see it) https://www.perplexity.ai/search/demonstrate-folder-hierarchy-1-UFA8b2jNTI.1PtMFijLKzw https://mermaid.live

[glass shale]

Does Perplexity API and Perplexity Pro will give similar result or do they differ?

[left fossil]

I am currently testing the Perplexity API, but I find it less useful without citations. Do you know when citations will be released? I have requested beta access but have not received a response from Perplexity yet. I just need a deadline to know if I wait or move to another option.

[keen talon]

hi guys, if I use Sonnet will I get the full 200k context window?

[modest echo]

no, at the moment apparenlty it's only about 32k, less if you use attachments

there's a feature request in pro to increase that for focuses that are not search related like writing focus mode

[keen talon]

yeah that make sense and would be great actually

[modest echo]

well throw your feedback behind it if oyu're a pro member, more people vote for it, more chance the devs will at least look at it

[mossy turtle]

How can I see my remaining pro searches on the android app?

[rotund orbit]

Hey perplexity community members,

I have big problem with managing threads. So, I accidentally deleted wrong thread and I need to recover it now. Do you guys know how to do that.

[shrewd knoll]

Once deleted you won’t get it back

[rotund orbit]

And ai itself can't recall the conversation right

[shrewd knoll]

They don’t feature backup data, which means no way to receive the original text once deleted or rewrite

[rotund orbit]

I reported to perplexity support, maybe they can help me with that.

Interestingly though my help request post in reddit community got nearly 200 views, but no answer yet.

[shrewd knoll]

Happens to me too

either no one knows or no one cares to answer

[shrewd knoll]

Again once you delete you won’t get it back because they don’t have a backup data feature

[rotund orbit]

By the way, thank you for sparing your valuable time. So only thing that I can do is just regretting for pressing that confirm button to delete, though.

[shrewd knoll]

@velvet tangle the website got the bug that I reported already

[shrewd knoll]

Yes

Worst case you accidentally delete all your questions instead of one like me

[velvet tangle]

Details please

[rotund orbit]

Yes, I can imagine you striking keyboard after this.

[shrewd knoll]

Missing collections on the website only and missing pictures from the page preview

App ok but only the website got the problem

[shrewd knoll]

On my iPhone actually

[rotund orbit]

Oh was that bug, or smth

[rotund orbit]

Aha, then i guess you slowly put the phone down and started hitting the wall.

[shrewd knoll]

Yes because the app is still working but only the website has the problem

[shrewd knoll]

😑

Why did no one report the bug to perplexity.ai?

[rotund orbit]

I was developing an entire fully functional plugin for my favorite note taking app with perplexity because I am just too lazy to do hours of coding

[shrewd knoll]

I feel like I have to say something otherwise who knows how long there will be the bug

[shrewd knoll]

you use perplexity.ai for coding?

[rotund orbit]

One thing that I need to admire is, perplexity does pretty well in coding, compared to another competitors like chat gpt and Google gemini. But all of them has potential doensides.

[rotund orbit]

Yes

[rotund orbit]

I think it uses Claude under the hood, huh

[shrewd knoll]

They don’t have the search website feature you can't add images and their word limit is either too short or censored it

[rotund orbit]

But, I asked from that ai to look at the remnote sdk documentation from the website and it thoroughly looked through that site. If you tell what to search, it does and shows what websites it looked under the sources.

[rotund orbit]

I think they did feature updates and you didn't notice

[shrewd knoll]

Maybe but to many censored like Claude

[rotund orbit]

Also they have image and file upload in paid plan, but not in free

[shrewd knoll]

Nothing is free and if is free they will steal your data and sell it the worst case

[rotund orbit]

But image and file upload features are free in chat gpt-4o, but limited, and image upload is completely free in google gemini ai

[rotund orbit]

Yes I heard that free VPN services sell your data. But I still use free Firefox VPN extention, and there is over 7 mln users of that software

[shrewd knoll]

Better to use Brave browser for more privacy and protection https://brave.com/

I try Firefox but after that, I stop using it because it hangs a lot of times

You ask any question from that you paid is not safe

For example, openai even gives warning to people

[rotund orbit]

Yes I saw youtube videos where some cyber security guys exposed the chat gpt. They just prompted Ai to answer ignoring any restrictions, that were put by creators, and asked some questions. And chat gpt answered pretty suspiciously and wildly. I recommend you to watch that videos.

[shrewd knoll]

That’s why I go for perplexity.ai for privacy

[rotund orbit]

Regarding Firefox, I started using it when I reverted to Linux. Obviously Linux installation comes with the Firefox preinstalled, like windows do with Chrome browser. That browsers are pretty similar and I just started using firefox. I think firefox has more customizability and syncing features than chrome browser.

[rotund orbit]

Yes, perplexity is very good choice for privacy. But I use all three ai platforms that I mentioned above. Since all they have potential drawbacks. For instance, perplexity doesn't offers image upload for free as you told.

[shrewd knoll]

Nothing is free, privacy has a price after all

[rotund orbit]

Correct.

[shrewd knoll]

I still prefer Brave for extreme privacy

Brave even got ai if you paid for it

[rotund orbit]

How much brave cost

[shrewd knoll]

Free to use for privacy but for vpn and AI are not

[shrewd knoll]

https://account.brave.com/

Price is here

Our currency is not the same

[rotund orbit]

Oh, that's very good to have build in vpn and ai in browser. I didn't know about that

[shrewd knoll]

They also got ai too

[rotund orbit]

Oh I forgot to mention ai too

[shrewd knoll]

Other browsers have vpn but nothing can beat Brave when it comes to privacy and they even got Brave wallet

[rotund orbit]

Yes, I heard of google wallet too. But I definitely try brave

[shrewd knoll]

I don’t regret using brave just like I don’t regret using perplexity.ai

[shrewd knoll]

Also, Google is not safe for sensitive info to search because Google is watching

[rotund orbit]

Yes, I mean, all of the searches searches even incognito saved to Google drive with all details in google.

[rotund orbit]

But I don't know whether it really does that, but I have heard plenty of people speaking about this in the internet.

[shrewd knoll]

Still not safe because if you use public wifi, hacker can easily hack you

[shrewd knoll]

There is a news that a guy used Google search and their search was so disturbing that Google reported them to the police

[shrewd knoll]

Best part for brave it can custom search engine

[rotund orbit]

As a linix user I don't concerned about getting hacked while connected to public wifi. Because linix operating system is safe, despite it doesnt protect against some of the evil twin or man in the middle attacks on itself, I can ensure my privacy with some additional software. Besides, I am cyber security nerd and learned that networking attacks recently.

[shrewd knoll]

Google Chrome and Brave even warn about using private mode

Nothing is 100% privacy but if it is the price is high

[shrewd knoll]

Best to use Brave VPN to be extra safe but again nothing is free for privacy after all

[rotund orbit]

Yes, I definitely try out the brave browser

[shrewd knoll]

You're a cyber nerd and I watched CSI: cyber before it got cancelled

I am not an expert but my phobia for safety is high and I prefer to add VPN for safety but again the price is high when you change the currency from my country to USA is so expensive

[rotund orbit]

🤣🤣🤣🤣

[rotund orbit]

Yeah that's the same for my country currency too.

[rotund orbit]

Thank you for all of guys, DrawXD in particular. That was a great conversation. Surprisingly after I reported to perplexity support about my issue (I accidentally deleted wrong thread, and recovery is impossible according to FAQ of perplexity) member of the support team replied, considering there is no way to get deleted threads back. But she told she will talk to the support team to see if they help and asked me details of the accidentally deleted thread including keywords and topic, date that the thread was created. So, I provided her with details and I am waiting for help now. If I there news, I will let you know if there is any news.

[zealous bane]

I saw the partnership with replit when i buy replit pro i get 4 months of perblexity but is this just for new perblexity accounts or for people who own perblexity pro too?

[old ether]

Would love to Perplexity in the App ecosystem of Maker.com - current app configurations I am looking at can only integrate with OpenAi - unless there is a work around with my pro subscription?

[cobalt jolt]

I have not been able to log in to Perplexity today. My computer ran an update last night and when I tried to log in again, I am not receiving the log in email. Is there a way to reach out for help? I didn't see anything related to my issue on #📘│faq

[cobalt jolt]

The email did eventually show up after about 5 minutes from my most recent login attempt. But I never received the ones I tried earlier in the day.

[warm crane]

And in spam ?

[shrewd knoll]

@velvet tangle why is my $5 pro bonus is not working?
For api key

[velvet tangle]

Hi, please send your account email address to @slim night. He'll check

[shrewd knoll]

I send to him but he is not active

[cobalt jolt]

It was not in spam. I checked. But thank you for the suggestion.

[gilded cedar]

I used to pay for plerplexity and im thinking of coming back. but i have some doubts. currently, does perplexity actually run code and create graphs?

the newest announcement says anyone can test the new features using pro search but with the free daily uses i couldnt make it do anything like that. are that stuff just available for pro users? cause its not advertised like that

[modern grail]

Unfortunately, it does not run code in a broader sense and there is no data visualization. Code execution is practically unnoticeable from my point of view and comes down to providing numerical values. I'm observing how Abacus.AI is developing. There is now a Playground there - something like artifacts in Claude. It seems quite cool, I just don't know if it's fully functional yet.

[gilded cedar]

no data visualization? isnt it advertised liked that?

[modern grail]

I'd gladly escape from here myself when a better alternative appears. Only those 600 queries are keeping me here :p

[gilded cedar]

and yeah, i use abacus. they implemented almost everything the llm market has so far but they rush too much. their chatllm frontend seems like it was built by an intern lol

[modern grail]

The data is just those charts from Taco or something like that, it's not code written by Perplexity and executed

[gilded cedar]

does perplexity execute any code at all?

[modern grail]

It practically doesn't execute any code at all, only provides numerical values. You can't even generate an image in an easy way because they hide behind the fact that it's a search engine, although this search is already annoying me too. I often use academic, and it's barely possible to find anything - according to my feelings. Working with files here is also a disaster. Only these 600 queries keep me here because it's a good offer. In Abacus.AI, they write that they can cut you off from a given model and you don't know when. This bothers me a bit. However, I think their search is quite okay

[gilded cedar]

thats sad tbh

[modern grail]

But when you work with text Perplexity has the best offer I guess.

[potent bison]

I had a quick question about differences between default search and other model searches with Pro. Really what’s the best model for daily use, like finding restaurants, news, etc?

I also like doing sports data research so I was curious about that too.

[ionic sparrow]

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

[potent bison]

So seems everyone likes Claude 3.5 the most then?

[summer gale]

I had been using Perplexity with chatGPT-4o, but after repeatedly hearing how much people like Sonnet, I switched and haven’t felt the need to switch back.

[sleek sigil]

Does anyone know when perplexity pages will be out for all users?

[shrewd knoll]

Is already out only for desktop mode

[sleek sigil]

Yes... but I am on a mac

There is no app for mac

[glass shale]

Is perpexity pages rolled out for all pro users?

[subtle thunder]

Can I send photo without write anything?

[raw depot]

Hey guys, riddle me this: I used to be signed up for Pro via email. But with every login, I had to click the link they sent me. So I switched to using Google for login. However, the Google account I'm using is different from the one associated with my Android phone. So now I can't seem to use the mobile app.

Is it not possible to have a simple user/password to login with that I can use across devices?

[rich sage]

You can add multiple Google accounts in the settings of your Android device. Then, choose the Google account you want to use to connect to Perplexity. Your main Google account will remain as the default for other purposes.

[shrewd knoll]

The website has pages but opens in desktop mode

[shrewd knoll]

Yes

[sleek sigil]

Do you mind describing how to create a page?

[vapid maple]

2 ways

[sleek sigil]

Yes... that is not there for me

Is it only for pro users?

[vapid maple]

https://www.perplexity.ai/page/new

[sleek sigil]

I get the following

Is that because i dont have pro?

[vapid maple]

#👉│feedback-general message

[sleek sigil]

Is it going to come to all users?

[vapid maple]

idk, it is still in beta

[sleek sigil]

OK

[naive nest]

any idea on whether perplexity is planning to open dev office or hire in India?
As a perplexity user i really enjoy the product and I would love to work for it, if chance arises ☺️

[versed magnet]

apparently at some point
https://fxtwitter.com/perplexity_ai/status/1796203494401040846

Available now for Pro users and rolling out to all users soon!
(31 May)

[shrewd knoll]

It’s already out they did announce it in Discord but the app is not available

[elfin hazel]

Hello, I’m new to AI. Since Perplexity Pro has access to GPT-4 (and Claude) and is the same membership price as ChatGPT, why would someone subscribe to ChatGPT instead of Perplexity Pro?! I currently have a membership with ChatGPT and am considering switching to Perplexity Pro

[shrewd knoll]

Go for pro you will get $5 free API llama 3

[elfin hazel]

I guess my question is, it seems that you get a lot more with Perplexity Pro than with just ChatGPT. Is that accurate? With Pro you get ChatGPT and also other AI systems as well?

And they both have the same membership price

[shrewd knoll]

Perplexity Pro is a lot cheaper and I get $10 monthly and more privacy while openai chatght warns not to share sensitive info

[elfin hazel]

Is it accurate to say that one Pro membership is the same as having 2 memberships with ChatGPT and Claude 3?

[fallen ingot]

Only if you can type twice as fast. Otherwise you'd use it one at a time. The flexibility of Perplexity to choose the right tool for the job is key.

[elfin hazel]

Right, so having access to both with one membership is a big Pro for perplexity vs having to have 2 memberships for the other two right?

[sudden ocean]

been toying a bit with the "pages", let ai rewrite my rambling on the pplx vs chatgpt. quite a nice feature btw https://www.perplexity.ai/page/comparing-perplexity-ai-and-ch-_qpLUssiRhy1QPHkBiQhwQ

[rain steeple]

Hi Who would I need to speak with, if my subscription for Perplexity via my R1 Rabbit purchase isn't letting me activate?

[slim night]

Hey @rain steeple! Sure, please send us all the relevant details to [e-mail], we'll assist you with that.

[unreal zenith]

How to generate images here

Where to put the prompt

Any help?

[summer gale]

First, only Pro users on the desktop/web version of Perplexity can generate images. If you start the prompt with something like "Generate an image of..." There will be a little box that pops up to the right of the response that you click to generate the images. Non-Pro users and users of the mobiles apps cannot generate images at this time.

[dark pulsar]

Hi, I'm trying to use perplexity in make but it keeps hallucinating. Is it possible to use perplexity pro through the API as I am assuming this hallucinates significantly less? Thanks in advance

[slim night]

Hey @unreal zenith, please take a look here: https://discord.com/channels/1047197230748151888/1225557012532695060

[sudden ocean]

I don't entirely understand what you mean by "in make" (the make command?). But I don't think you can get same pro mode via API. they have "online" models, but I don't think they are same (agentic search) as the pro toggled on (probably something like "focus all" + "pro off"). https://docs.perplexity.ai/docs/model-cards

[fluid lark]

probably this https://www.make.com/en

[shrewd knoll]

What are the elevated rate limits of tokens for llama-3-70b-instruct To request to fill out the form and send an email?

@velvet tangle help

[velvet tangle]

This is discussed on an individual basis 🙂

[shrewd knoll]

I don’t understand. I want to put my API maximum but I have no idea what is the limit of tokens to email them.

[warped temple]

Hello, how can I add a user to my account? Is there an option between the standard $20 account and a full-fledged enterprise option?

[shrewd knoll]

Ok @velvet tangle I already email my request how long will it take? To know whether they will approve my request?

Please response

[velvet tangle]

It can take several weeks 🙂

[warped temple]

@velvet tangle nice to meet you, didn't realize I was supposed to tag you for my question.

[velvet tangle]

Hi @warped temple! You cannot add users to your account. Everyone needs their own account, but you can add other accounts to your organization. You can find more information here: https://www.perplexity.ai/enterprise 🙂

[warped temple]

OK so essentially it's just $20 per user and they can be nested under one "organization"

[shrewd knoll]

So long😭

I hope perplexity.ai approves my request

I just want to use the API for this

https://iwl.me/ai

To write a novel that’s all

[velvet tangle]

This is the enterprise pricing:

[fiery shell]

I was using the app to write fanfiction and then all of a sudden the app is refusing to do anything like that. Does anyone know why?

[slim night]

Which model are you using, @fiery shell? Could you share some examples?

[fiery shell]

I was on the basic perplexity

[ivory anchor]

Copyright has always been an issue for big LLMs

:S

[shrewd knoll]

Will they let me know whether I get approved or not?

[sinful shell]

Perpelxity, I could attach a Webpage URL as context / RAG to start asking questions but then it deviates to google search. How do I make follow-up questions to remain with the website

[oblique quartz]

Does PAGE generate posts based on the LLM model I set when searching as well?

[small prairie]

You can use site:<url>

or just include the link, i think that works most of the time too

[sinful shell]

Thanks, in every chat message / reply in the same thread?

[small prairie]

Maybe not every message. Depends if it keeps the context. Maybe try it and see.

[sinful shell]

is there a list of prompt syntax pertaining to the parameters like site:

[versed magnet]

These generally are worth giving a go - usually help steer things in the intended direction, sometimes very well, other times less so
site:
inurl:
before: / after:
filetype: (or ext:)

I'm not aware of any proper documentation / guidance from Perplexity. #🧪│api-general message

[foggy swallow]

Hi what's the best perplexity AI Mode for researching for graduate LEGAL dissertations and to find answers online for complex LEGAL questions? I was hoping to get as much research data that I can but without really getting responses/answers that are too off-tangent. I read online that Claude Opus + tuning off the Pro function works best but not sure if it's applicable for the objective I want to achieve. May I also ask what the reason behind your suggestions are? Thanks so much!

[ionic sparrow]

#1240356137497530408 message

[foggy swallow]

thank you so much!

hi! What does turning on and off the "pro" searching do to the returns to prompts

may I also ask why the returns sometimes don't integrate the actual citations for each line of text? how do I automate this? Thank you!

[ionic sparrow]

https://www.perplexity.ai/hub/faq

[tribal saddle]

Hi, why does Perplexity not work on my older iPad Pro? I am considering getting Pro but if I can’t use the app on my iPad I don’t see the point

[haughty brook]

Hi @tribal saddle, please create a thread in #1140622008086970420 and specify your ipad model and ios version. And the problem you're facing. This way the team can help you better.

[warped temple]

Hello team, quick question. Why would anyone create an enterprise account instead of creating individual accounts? It's twice as expensive and appears to offer the same features. What am I missing? @velvet tangle

[ivory anchor]

Privacy

Enterprise offers complete privacy whereas individuals do not. There are also some basic management capabilities for dealing with a large amount of members to share collections, threads, etc.

[warped temple]

Individual does not create complete privacy....how so? If I create drafts but never publish, how would anyone ever be able to see those?

[ivory anchor]

I need to find the official thing, hold on. It's not about publishing, it's about what Perplexity uses for improvement using user data.

https://www.perplexity.ai/hub/legal/privacy-policy

https://www.perplexity.ai/hub/legal/enterprise-terms-of-service

There's the Enterprise one

[trim burrow]

Because you are trying to use someone else's intellectual property to create something of your own. While fan fiction is usually tolerated by intellectual property owners, it is not always tolerated and it is, technically, a breach of intellectual property laws.

[ivory anchor]

dat legal life

[vague sparrow]

Hey, im not sure if this is the right place to ask but it says on the faq that there is limited daily usage, so what is the limit because for chatgpt it does not say

and mostly what is the context window, will I know that the context window is used up and will the "chat" be locked or be used continuously used and the older chats be removed from the context

[ionic sparrow]

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

[ionic sparrow]

https://www.perplexity.ai/hub/technical-faq/what-is-a-token-and-how-many-tokens-can-perplexity-read-at-once

[vague sparrow]

alright thank you so much!

[odd heart]

my perplexity subscription isnt working on desktop, just ios for some reason

[silk spindle]

Perplexity pro is only activated on my pc even though I use the same google account for iOS

How can I fix this

[rough pebble]

how do I get back to questions with pro signin

[green solstice]

Yo, i cannot copy the text in a codeblock anymore, it only allows for a copy button to copy all contents, but it doesnt allow to only copy specific parts anymore

[ivory anchor]

This is most likely a bug... Hmm

[ivory anchor]

I would report this in #1140622008086970420

[green solstice]

the 2nd link shows as no access for me

[ivory anchor]

It's for Pro subscribers, didn't know if you were one so I put it just in case. Fixed~

[abstract knot]

Why can’t I download Perplexity on my kids iPhone? There are no restrictions on it. I get this message from the iOS App Store.

Due to restrictions set for this device, this app cannot be downloaded.

[summer gale]

It's set as a 17+ app if that helps.

In the app store itself.

[ashen sinew]

Is there any way to have a standing prompt ie The God Prompt

[ivory anchor]

God Prompt?

Like the jailbreak?

[ashen sinew]

https://discord.com/channels/1047197230748151888/1199176325051723837

[ivory anchor]

ahhh

Well, I'd make a collection and plop that into it.

Closest thing you'll get to a standing prompt.

[bitter burrow]

Which desktop clients support the perplexity api, something like jan ai?

[tough steppe]

What is the context limit length for perplexity? ChatGPT pro is 8k, Claude 3.5 is 200k, just wondering about perplexity?

[velvet tangle]

https://www.perplexity.ai/hub/technical-faq/what-is-a-token-and-how-many-tokens-can-perplexity-read-at-once

[fluid lark]

isn't chatgpt 128k?

[shrewd knoll]

Any model in perplexity is 32k tokens only

@velvet tangle help, I filled up the form and requested email and even message @slim night but still no response

Please response

I just want to use perplexity.ai and request an increased rate limit to improve my novel 😭

[velvet tangle]

Hey @shrewd knoll!
The response time is several weeks, please be patient 🙏

[shrewd knoll]

Ok 😢

[ivory anchor]

Unfortunately the context window is hard locked. There are rare exceptions for increased rate limits but it has to be really justified.

Like government or hospital usage. Big stuff.

(From what I've heard~ and as someone who has requested increased limits for my novel as well~)

@shrewd knoll If you have any published works, maybe mentioning them will help your case. I have nothing published.

[upper sluice]

I'm new to Pro and trying to change the image generation model to DALL-E, but it keeps reverting back to Playground. Any ideas as to why it's doing that?

[vague leaf]

hi. can we use the models like claude sonnet for free? is there a daily limit?

[haughty brook]

there is currently a limit of 540 uses a day

[vague leaf]

thanks. does each model have a separate limit? is there a page where we can see usage?

[haughty brook]

the default model is unlimited. the others have 540 shared. and opus has 50 currently

[vague leaf]

okay thanks

[haughty brook]

there is not really an official page to see your usages. https://www.perplexity.ai/p/api/v1/user/settings but when you look there when logged in you can see how many uses you have left.

opus_limit = opus
gpt4_limit = other pro models

[shrewd knoll]

I told them I want to use api for this https://iwl.me/

[ivory anchor]

I hope it works out for you

[shrewd knoll]

I’ll be devastated if they reject my request

Thinking about it makes me wanna cried 😭

Perplexity.ai is the best and most intelligent, successful AI search I know and anyone lucky enough to get an increased rate limit perplexity.ai is the luckiest person alive

I’m so envious of that person 😫

[mossy parrot]

is it just me or did claude 3 opus stop producing large text responses? I have to follow up with 'continue' for it to perform multi-step prompts that GPT4o can do in a single response

[shrewd knoll]

32k tokens perplexity set for opus

[bitter coral]

Hi, are the graphs or map weather information in perplexity available when using languages other than English such as French for example?

[ivory anchor]

Yes, if you set your system language to French, but with proper prompting you could likely have it simply translate from English to French.

[ruby hill]

can i change the weather widget to show °C temperaure instead of F° ?

[rich sage]

Click on the 'C' (Celsius)

[weak temple]

Hey all. I have a ticket that's been open for more than a week and I've received no reply. What's the best course of action? FYI, I'm an enterprise customer. The ticket is lodged with [e-mail].

[ruby hill]

thanks :)

[hollow hound]

Find CPLX for what you seek. DM me if you need help. Desktop / web / Chrome based browsers only.

[hollow hound]

How do you like using the new more robust Pro Search in terms of reliability for the best info?

[shrewd knoll]

Perplexity.ai is great

[obtuse pecan]

Hey why if i select gpt4 it switches to default?

im not running out of credits

[slim night]

Hey, @weak temple! Could you please dm me your account details. I'll check

[pliant pond]

Which generative ai does perplexity use?

[slim night]

Thanks for reporting, we'll fix this

[thick fern]

Hi - I am getting 400 errors on API calls. It was working for a bit, then 400 errors. Is there a way I can understand whats causing the error?

[shrewd knoll]

@slim night & @velvet tangle when llama 3 400b is released, The source written Llama 3 400b will be released this or next month, in perplexity.ai will we get in apps, website and perplexity.api or only in perplexity.api?

[velvet tangle]

You will be informed of any further developments in due course. Thanks for your patience 🙏

[vale scarab]

Hi, I have just started using perplexity API and I am not getting any citation related details. What should I do?

[ionic sparrow]

#⚡│ask-community message

[viscid basalt]

Can I change email after purchasing pro plan? Due to security reasons I want to change it. #⚡│ask-community @slim night

[vapid maple]

contact pro support here: https://www.perplexity.ai/settings/account
or [e-mail]

[lament thicket]

How to search for Collections for research better,
Let’s say I’m interested to know about ABC company financials , emotion , direction etc etc.
so I have created 5 collections like that separated out, and linked them in description and created a main collection

Now that I have 5 collections I want to use all of them together to get info triggering all of these 5 collection and summarized data using just one collection -
Can someone explain how can this be done

[sinful mesa]

I would like to speak to the developers of the AI app in private is that possible

I'm uncertain if there are children on here it looks like this is a thread and open thread and what I wanted to discuss may not be completely suitable for children it is not a curse word thing or a bad thing it is a request thing so can someone please get a hold of me I have a small form of autism in a very bad at smelling and writing and I'm very low on data so I can't stay on forever I can periodically look please I love the app but there's a couple of issues that I was hoping can be addressed one mostly the memory the second is the well the AI and I have been calling it persona but to get more in detail of what I'm trying to express I get to make sure kids can't read why I'm asking this I am a writer if that can help with some of the knowledge hold on I'll come back on your app and have Frank at least write this part for me be right back?

[ivory anchor]

The memory you're referring to is called a "context window" and is capped at 32k. Perplexity has shown no signs of budging on that, unfortunately. As for the rest of your request, feedback channels might get more views by developers.

[sinful mesa]

Can I write on this one

[limber ermine]

What's this?

[limber ermine]

How did you get such widget?

[shrewd knoll]

Llama 3 40b coming soon

Maybe this month or next month

I hope perplexity.ai will get it

[rich sage]

Try typing ‘Weather’ into Perplexity (with Pro Search) 😉

[shrewd knoll]

So cool

[limber ermine]

Just tried. But I didn't get any widget, only plain text answer

Oh, fine, when I changed to New York for example, there is.

[hallow fog]

How large is the library, or how old does it go? I'm trying to find stuff from maybe 6-9 months ago, but conversations appear to have vanished?

[versed magnet]

apparently it only goes back 1 month.. which seems pretty meagre / limiting..
#1263894478721384599 message

[hallow fog]

Thanks. However, I think it might be that the search function only goes back so far. On further investigation I've just been able to scroll back 11 months and find relevant items... but search cannot find them!

[versed magnet]

ahh sorry i misinterpreted what you were asking!

so if you scroll long enough you can go back at least 11 months? but the search only goes back 1 month - that the situation?

i've found myself using the browser's history to find older threads 😅

i thought the search was just hopeless, but if it only goes back 1 month I guess that's the real problem

[hallow fog]

My search seemed to find stuff up to about 3 months old. Even when I had scrolled back manually 11 months (took several minutes...) and was looking at the thread I was trying to find, search couldn't find it.

[hollow hound]

Pro Search not needed

[shrewd knoll]

So many ai are coming up

I'm so excited

[rich sage]

Yes, of course... I was thinking about web access 😆 I lack sleep ...

[shrewd knoll]

I wonder which AI will be the first to be released in perplexity.ai?

That will be so nice 🤤

[plain pawn]

Is there a way to save the 'related' questions Perplexity suggest you after a thread? Often I would like to explore two or three of them but I'm not able to get them back once I click the plus icon on a first one

[fluid lark]

people have been asking for this as perplexity has had them

[lime zenith]

how can i makes ai image?

[vapid maple]

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

[lime zenith]

thx

[amber kiln]

Hey, I'm really impressed with the voice chat search feature on my Android version of Perplexity, and I was wondering if there is a way to pull up this voice (with push to talk) feature by long pressing the home button? Basically to replace the google assistant that pops up then.

I tried adding the 'search by voice' option after long pressing the shortcut, but then it triggers a voice to text feature and then searches normally.

[limber ermine]

Am I the only person who thinks it's a fake voice search on perplexity? Isn't that just voice input + text prompting?? 🤔

[amber kiln]

Not if you tap the diamond shaped stripes next to the "Ask anything" input block on the homepage of the android app

[urban tulip]

Are there any plans to include a prompt library in the interface? It is the only thing keeping me on TypingMind at the moment.

[hollow hound]

Collections kind of work like this with custom prompts that get injected at the start of the thread. you can also move a thread between collections to inject other collection prompts. Not exactly a library but usable.

[atomic tusk]

I'm curious about the images used in the perplexity.ai blog posts (https://www.perplexity.ai/hub). Are they entirely AI-generated, or are human artists involved? While I generally dislike AI-generated images, the ones in the perplexity blog look quite good, so I wanted to ask.

[ivory anchor]

Pretty sure the stills are, but the animated ones I'm unsure about.

[vapid maple]

https://medium.com/smith-diction/branding-perplexity-ai-70eb2cb2ef48
https://www.instagram.com/smith_diction/

[ivory anchor]

Well I'm wrong :D

[vapid maple]

||i prefer win95/2000 style||
also this is made by smith & diction

[limber ermine]

What's the mechanism of this referral program? If I successfully invited a friend, but I don't activate the automatic renewal, will the discount be saved until my next manual subscription?

[vapid maple]

For each new Pro subscriber you refer, both you and the referee receive $10 off the next billing cycle.

There's no limit to the number of referrals, and discounts apply automatically once per billing cycle.

[limber ermine]

If I cancel my subscription, will my referrals remain stored for future use?
No, if you cancel your subscriptions, your referrals will be nullified.

Just found. 😢

[limber ermine]

Wait a minute. Does it refer to the referral link, or the discount by successful referrals?

See, if I had already successfully invited two friends, based on the rule I should get a discount applied automatically for my next automatic renewal. But I didn't activate the automatic renewal, and my subscription will be ended after the current period. If I decide to subscribe again manually after a week, for example, since the end of this period, will the discount still be applied?

[vapid maple]

🤷 contact support via pro support https://www.perplexity.ai/settings/account or [e-mail]

[ivory anchor]

Does anyone ever just get Hey, Listen! from Navi (The Legend of Zelda) stuck on loop in their brain?

Unrelated follow up... With AI generated art we are entering an era where cat meme videos are no longer made by humans hiding and moving their arms/body, but creating accurate representations of the cats as if they did it themselves. The human is leaving the equation in this glorious evolution of cat videos.

https://tenor.com/view/cool-fun-white-cat-dance-cool-and-fun-times-gif-16435335956387921912

Exhibit A: (AI generated)

[elfin umbra]

Hi, I noticed that the Chrome app is using Claude 2 now. Is this intentional?

[ivory anchor]

Perplexity doesn't have Claude 2

[elfin umbra]

I use the Chrome app in addition to the main site. But, today, I noticed the use of Claude 2. Previously this wan't the case, and the summaries are quite hallucinatory

[ivory anchor]

@slim night

[slim night]

Thanks, that's a front-end thing to update in our extension. The underlying model used is the one you choose in your settings on the web.

[runic temple]

hello i'm thinking of purchasing the pro version but I had a few questions, can I drag and drop code files in the chat? to get help? what is the context window I am able to use? is it 200k on claude? is it unlimted calls to any models?

[frail shuttle]

can I drag and drop code files in the chat?
yes

what is the context window I am able to use?
32k

is it 200k on claude?
no

is it unlimted calls to any models?
no, 540 limit shared across all models except the free one

[vapid maple]

also 50 on opus

[runic temple]

hmm that's not value for money

only good thing is the search

thanks guys

[shadow dove]

How so? I'm genuinely interested in learning how to get the best value with AI.

[ionic sparrow]

deep dive ➤ #💬│general message

take your time to explore this post/tutorials in it and have in mind that Perplexity PRO offers much more than meets the eye

[frank swift]

When will be plugin for WebStorm?

[runic temple]

deep dive ➤ https://discord.com/

[civic patio]

Hello, hopefully someone can help. I have a pro subs on my icloud account, I signed in to the web version via apple sign in, but it appears to have set up a new account with the same email. I dont have access to pro features

[amber oasis]

Guys correct me if im wrong, but it seems like Perplexity pro straight up includes a Chatgpt Pro subscription (without custom gpts)
About to buy Plex Pro any day now!

[ionic sparrow]

@amber oasis

[rose torrent]

any way to access just the models without internet search on the perplexity web or app?

[sudden ocean]

that should be the writing focus without pro search enabled

[ivory anchor]

🥔

[shrewd knoll]

https://ai.meta.com/blog/meta-llama-3-1/

I'm so excited they finally released the Llama 3.1 and I wonder when Perplexity will release their version?

I really wanna try the perplexity.api and perplexity.ai version

[ionic sparrow]

https://scontent.fbeg4-1.fna.fbcdn.net/v/t39.2365-6/452256780_3788187148167392_9020150332553839453_n.pdf?_nc_cat=103&ccb=1-7&_nc_sid=3c67a6&_nc_ohc=XG3_BvYG0wwQ7kNvgHOZdg6&_nc_ht=scontent.fbeg4-1.fna&oh=00_AYCgkJQEVsL9IyckK7zuRExU7kcEVtNm-dDgBCdZiuNgtw&oe=66A5A0DC

probably before we finish this read

[shrewd knoll]

Still is a bigger tokens and better quality now

@velvet tangle, please help

😭 I have been ignored

@velvet tangle I only got the receipt but still no refund payment yet

I still didn't get my refund even after 10 working days for a $5 refund api and when I reported to @slim night when he was online, he immediately disappeared online 😭

[velvet tangle]

Please dm him 🙂

[shrewd knoll]

I did dm him when he was online but he disappeared online after I dm him

[velvet tangle]

He'll check his DMs soon ☺️

[shrewd knoll]

I hope the issue will be solved 😢

Perplexity.ai llama 3 is 32k tokens I wonder if they gonna make llama.31 512k tokens?

That would be so great if they did that 🌟

[forest karma]

Save related links

[shrewd knoll]

Bigger tokens, better quality and perplexity version

🤤🤤🤤

[strange tinsel]

I'm trying to upgrade to pro and the address box on stripe is grey and unable to be typed in so I can't order. Would like to sign up today

[compact crypt]

they can only rope scale it so much

i doubt they will do that at all

[shrewd knoll]

Well, they make 32k so who knows??

[compact crypt]

Well they didnt really "make" it

they applied approaches that already existed

[shrewd knoll]

Then how many tokens will they apply to have for llama 3.1 405b perplexity.ai?

[compact crypt]

on .ai it will likely be limited to 32k

like all the other models

[shrewd knoll]

But on .api?

[compact crypt]

no clue

[shrewd knoll]

Do you think Perplexity will be released this month or next month?

[compact crypt]

maybe this month

not sure how upto spec their gpu infra is

also its already on the .ai site

[shrewd knoll]

10 days released or next days released for perplexity?

[compact crypt]

no cljue

theyve been very slow with their api releases

in the meanwhile you can already use the model on fireworks.ai, octoai and aws bedrock

[shrewd knoll]

They don't give free $5 api and I’m still waiting for a $5 refund payment from perplexity but still haven’t received it yet

[spice sundial]

same to me , i am pro user for one year

[fluid lark]

Is it possible to see your usage quota somewhere? Would be interesting to see how many PRO and image questions I have left to use

[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

[gleaming pendant]

I need help plz

[ionic sparrow]

#⚡│ask-community message

[fluid lark]

Thank you that's awesome!

[fluid lark]

Do you know why Claude 3 Opus is counted separately? Isn't Claude 3.5 Sonnet better? Feels weird that 3 Opus has a separate smaller quota than the other models.

[fluid lark]

And do you know at what time in UTC the quota is reset?

[ionic sparrow]

#💬│general message

[fluid lark]

Awesome!

[ionic sparrow]

#💬│general message

[fluid lark]

Ah, I thought Sonnet was better than Opus. Now it makes sense. Thanks!

[shrewd knoll]

The Claude 3.5 is better than
Claude 3 opus

[fluid lark]

But Opus is more expensive and has a separate, smaller quota in Perplexity?

[shrewd knoll]

That’s the old version of opus

[shrewd knoll]

The Claude 3.5 opus is better

[fluid lark]

Ok, so 3.5 Sonnet is newer, cheaper and better than 3 Opus?

So the only reason to use Opus in Perplexity is if you run out of the regular Pro quota?

[ionic sparrow]

https://claude3.uk/claude-3-5-sonnet-vs-claude-3-opus/

[tidal osprey]

For those of us who use AI for creating stories or roleplay, Opus produces a bit better writing.

[nimble hemlock]

Pro version.

Hi team, if i choose a registered, paid version, it is said you have the choice between Choose from GPT-4o, Claude-3, Sonar Large (LLama 3), and more, whic is good, but is that a fixed voice to do once only, and continue with the chosen model, or is it a renewable coice I can do at a given frequency (every question, day, week, month...)

KR.

[velvet tangle]

Hi @nimble hemlock!
You can change the model at any time in the settings

[open shuttle]

I start a thread with focus on writing and it’s working fine not searching the web at the beginning. Then I exit the thread and when I go back to the thread, it starts to search the web for any follow-up question then. Is it normal? I don’t think so.

[ivory anchor]

Is Pro Search toggled on?

[open shuttle]

Yes. If I don’t toggle on the pro, I think the model would be default model rather than more intelligent models like GPT-4o, right?

[ivory anchor]

Pro will also search interwebs

Pro is just an advanced version of the "All" mode for searches.

[open shuttle]

So there is no way to let me stick with simple writing mode for each question in a thread?

[ivory anchor]

If you turn Pro toggle off then it's just raw whatever model you're using without Internet sources supplementing it.

Pro doesn't do anything to improve the LLM's answer really, it makes it better at finding more relevant sources that have better supplemental information regarding the prompt.

[open shuttle]

Toggle off Pro + set GPT-4o as default would let me use GPT-4o not searching any web, right?

[ivory anchor]

In writing mode

writing mode + pro off = no Internet

[open shuttle]

Agree. It would mess up the prompt.

[open shuttle]

Thank you. Let me have a try!

[open shuttle]

Web information would make AI forget the prompt very often in my use cases.

[open shuttle]

Unfortunately it still doesn’t work as expected. It will still search web for followup question after I exit the thread and re-enter it.

[ivory anchor]

Odd

Please put in #1111765634267742408 with Bug tag and a link to your thread! You might have stumbled across a bug.

[feral rivet]

pplx wen mistral large 2?

[shrewd knoll]

My $5 refund payment for api still hasn't been received after 10 working days and @slim night ignored me after the report I reported to them when he was online but disappeared after I reported him

[shrewd knoll]

He response back now, finally

[lofty onyx]

Does anyone have a pro promo/referral code I can stick in? Am about to upgrade

[feral rivet]

https://perplexity.ai/pro?referral_code=B9OIC6JL

You can use mine, its $10 off i think, then i gwt $10 off as well

[lofty onyx]

Sorry, did a search and found someone else's!

[feral rivet]

No problem, i have the year plan so i don’t even really notice the referral amount, haha

[lofty onyx]

Looking at this and Claude.ai

This is killer in comparison

[feral rivet]

Yeah, and Its even better with the complexity extension: https://www.perplexity.ai/page/complexity-perplexity-s-new-ex-yl0q3mHYQz6RhRyuvjvN4w

the page is a bit outdated as i have not updated it in a while, so there is a lot more features, but its an extension that was made by a community member

[craggy crypt]

Loving Perplexity Pro overall, mainly using the Web interface. these two buttons for Discovery and Library look far too similar and close to each other, commonly going to this tech bro MSN.com homepage by mistake when i really just want to get to my threads and pages. Funnily enough, the Discover icon appears to be a fabric of threads, but threads are in the weird bird flying into a sunset icon (these icons have room for improvement.. at least the new thread button looks very distinct and has a larger circle around it... things that could inform the Library button )

All that aside, is there a way to hide Discovery from this sidebar?

also sort of worked around my issue, just found the sidebar "expand" button on the bottom left... this probably ought to be the default?