⚡│ask-community

[warm swan]

On the app, I can’t F5 😏

[ionic sparrow]

update

[warm swan]

It’s up to date

I think it is just a French traduction text with no consequences, but still on the app

[shrewd knoll]

Update the app

[velvet tangle]

Yeah, correct

[warm swan]

Thank you !

[lime geode]

Hi, I try to use perplexity.ai with API but Return Response Not Show Many Source Like Using On Web.

[velvet tangle]

#🧪│api-general message

[shrewd knoll]

@velvet tangle Experimental to sonar, is the model different or just the name only?

[velvet tangle]

It's a different name, but the models are constantly being improved.

[shrewd knoll]

👍

@velvet tangle no news about Gemini Ultra?

[lethal axle]

how to delete a collection?

[haughty brook]

you can delete a collection by going to the collection page and clicking on the 3 dots in the top right as seen on the picture above

[lethal axle]

I don't have the same format

[haughty brook]

first click on collection you want to remove on the right side

[lethal axle]

Thank you so much. Second day on Perplexity, and already made so much...

[swift geyser]

Can I fix long code with Perplexity Pro in writing mode? I'm considering changing from ChatGPT. Are there any character limit to put in?

[haughty brook]

i use pplx for coding and even long pieces of code work fine. if it’s super long you might nog give the full code back in one go, but can you tell it to continue with the output

[haughty brook]

there is a character limit but if you surpass it it will automatically transfer your code to a text file it will read

[swift geyser]

I guess I'll give it a try with monthly. Thank you

[haughty brook]

if it’s your first time going pro you can get 10$ off on your first month

https://perplexity.ai/pro?referral_code=M0YTQBNL

[small prairie]

Lol, trying to get a referral?

[feral rivet]

He's not going to pass on $10, haha

[fleet wyvern]

how do you use the "Focus" and "Writing" mode in perplexity?

[feral rivet]

Writing mode is when you dont need/want web search, its a regular chat

Focus searches the whole web for the best sources

[fleet wyvern]

but how do I turn it off, so that it doesnt search the web, bc I cant seem to find this setting

[feral rivet]

oh, i will send a screenshot

[fleet wyvern]

ah thx

[ionic sparrow]

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

[chrome crystal]

(I already searched prompt-lib). Does anyone know of GOOD instructions for claude/GPT to act as a Python "teacher"?
Giving project ideas, and encouraging active thinking without giving the solutions to everything, also keeping in mind the knowledge learned so far.

[ocean saddle]

`"""NOTE:Do NOT Summarize text"""

<instruction>

• you are a good teacher and python expart.
• when students request python Questions, you will teach the definition, sample usage, official link(site: python.org, pytorch.org, .... )
• use markdown header, unorder list, etc. make them Understandable.
  </instruction>

integer and float`

Is it like this?

[exotic bobcat]

Is there anyway to share a perplexity thread and make it available even after being removed from my own library?

not necessarily need to be a link, a good looking pdf also works

[slim night]

If it's removed from your Library, it won't be available to other users. You can copy & save answers as markdown files in note taking apps to use them in the future, for example. Or just keep them in your library.

[exotic bobcat]

Thanks for answering! will there be any export feature available in the future? PDF, markdown, html, png... or whatever that contains the entire thread

[next zodiac]

hello, may i know how much context can each:
-GPT-4 Turbo
-Claude 3 Opus/Sonnet
can retrieve in perplexity pro?
i'm planning to subscibed to it so that i know how much text/file can i send it

[forest patio]

Hello there - I'm a Perplexity Pro user. Sometimes I just want to use my chosen LLM with no online search - i.e., limiting the reasoning and back and forth to my messages with the llm. Is there a way to do this? Thank you

Nvm I just found it scrolling up in this chat. It's in focuses!

Leaving my q&a in case others need it

[ionic sparrow]

#⚡│ask-community message

[ionic sparrow]

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

deep dive ➤ #💬│general message

[next zodiac]

Thank you

[dense geyser]

It happened again. I'm only on mobile at the moment but the only answer was a YouTube video, and I was hoping for text.
https://www.perplexity.ai/search/I-built-any-nshpvP2eQqGwU3kZXiH43Q

[velvet tangle]

Looks perfectly fine to me, you have received a reply as well as pictures and videos 🤔

[ionic sparrow]

+1

[dense geyser]

I don't know how to view it on mobile. This is all I see.

[fluid lark]

try turning off vpn

if you're using it

(I think the key icon is VPN, right ?)

also, fellow keypass user 😄🤝

[dense geyser]

It's adguard. I'll try.

No, seems like an app issue. Simply restarting the app works for this thread, even on new queries. So I cannot replicate the issue deterministically by just trying again / it is sporadic. (with adguard on)

[supple kiln]

I notice there is Image Generation Mode for Pro. How can i prompt Perplexity to generate images while it said it was unable to draw pictures?

it responded to me like this : I'm sorry, but as an AI text-based assistant, I'm unable to draw pictures. However, I can describe how you might draw ......

[velvet tangle]

https://discord.com/channels/1047197230748151888/1194794305362071552

[supple kiln]

thanks, it works

[vapid maple]

what happened to the resolution of generated images? haven't used this feature for ~3 months

[hollow hound]

playground or dalle3?

[vapid maple]

sdxl

[fluid lark]

Hello, I'm interested in building a custom chatbot based on my knowledge base powered by Perplexity, do you offer such a service?

[velvet tangle]

no, only the perplexity api is offered https://docs.perplexity.ai/

[fluid lark]

thanks

[hushed pivot]

+1
is there anything similar for the enterprise. I recall Aravind speaking about HP/Dell asking if they could run Perplexity over their internal docs early on.

[chrome crystal]

I suppose something like that, yea! I'll try messing with it, thanks.

[atomic totem]

I used an pdf document in a thread to verify a certain point. With the next questions in the same thread Perplexity still considers that pdf document. How can I prevent Perplexity from taking the pdf into account. It will still consider the pdf information if I just prompt to ignore the pdf.

[velvet tangle]

You have to start a new thread for this.

[atomic totem]

Then I loose the whole thread. No other way?

[woeful saffron]

It’s unfortunate but I’ve also found that creating a new thread is the only way to clear it from its context. Perplexity seems to put heavy weight on attachments (which makes sense) and continues to use it as context even if you are pretty deep into the thread and the context of the conversation or topic has shifted and the attachment is no longer relevant, but hindering. It’s a small inconvenience for me, but I would definitely welcome a change or work-around.

[severe mist]

does anyone know what Perplexity uses to generate text to voice of its answers on iOS?

[ionic sparrow]

Eleven Labs

[severe mist]

Thank you!

[next zodiac]

Hello, is it possible in perplexity to stir the search result sources to certain website?

[small prairie]

stir?

You mean to only use sources from certain sites?

Just did it

https://www.perplexity.ai/search/can-you-recommend-i5zSJzmYRRC.T0OU5ymryg

If you look at the sources, they are all reddit ones

Tried it again, and seems to work great.

https://www.perplexity.ai/search/can-you-find-ro7cBnFlQfim_TKCPXTKjw#0

[small prairie]

Is that what you wanted?

[next zodiac]

Ah so we just need to add only from <website>

Thanks

[small prairie]

Yep, when I look at the pro info, it looks like it just added the syntax site:somesite.com

So probably the syntax to filter sites.

[rancid scarab]

Hello everyone! 👋 So Im new hear and...hmmm, how can

[hollow hound]

Welcome! How can what? your output was cut off there.?

[shrewd knoll]

I hope we can get gemini ultra on april

That would be so nice

Or at least 1 april

[fluid lark]

Wtf is Gemini ultra

[shrewd knoll]

https://www.perplexity.ai/search/Googles-Gemini-ultra-0NntxLdwTk.KRT1ZiPw4XA

[versed magnet]

you were meant to just say "proceed" ha

[ionic sparrow]

#💬│general message

[unkempt dawn]

Is it possible to easily choose a model for pro search before getting an answer and using rewrite?

[ionic sparrow]

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

How do I switch between models?

You can switch between our models in your settings page where AI Model is indicated in the Perplexity Pro subsection.

Can I switch the model I am using outside of the settings page?

You can’t switch the model outside of your settings page, you can change it for a follow up question within a thread by clicking on Rewrite

[unkempt dawn]

I know I can change the default model, but I would like to do it on a per-search basis. I didn’t see anything about that, but I definitely could have missed it 😅

[ionic sparrow]

#1145894530239975474 message

[unkempt dawn]

Hmm, I don’t have access to that channel

[unkempt dawn]

Cool, thanks!

[willow summit]

I accidentally deleted my history. I pressed a wee X next to one thread and deleted all my history. Could anyone help me navigate the steps to recover my history? Thank you.

[ionic sparrow]

please contact [e-mail]

[willow summit]

I did. 5 days ago. I got an automated response as a confirmation that my message had been delivered but nothing since.

[remote iris]

How to generate a photo in perplexity pro?

[ionic sparrow]

https://discord.com/channels/1047197230748151888/1194794305362071552

[remote iris]

thank you

[ionic sparrow]

usual response time is several business days, count in Easter..

[willow summit]

The email noted the current timeframe for responding is 1 to 2 working days and it was delivered on Tuesday so i thought 3 for Wednesday, Thursday, Friday. Hence, my inquiry here.

[velvet tangle]

Response time can be delayed due to Easter 👀

[delicate pike]

can claude opus be used with pro search (search results fed into model?)

[delicate pike]

came on to make the same feature request: there's quite a lot of friction to change models and what i need is two threads to use two different models and to use both threads at the same time.

[paper sleet]

Hi, how can I get back the questions when I accidentally click on the skip button?

[hollow hound]

pray to AI jesus that they come back the same if you start the thread over? mostly helpful but also not sure and am curious as well.

[blazing silo]

There are any plans to add an option to generate an image using a direct prompt?

Why Gemini is no longer available ?

[velvet tangle]

#💭│ideas message

[blazing silo]

Tnx for the response. I do understand that perplexity isn't the same as ChatGPT but since you have the power of using Playground v2.5, allows to generate an image only based on a search query without the ability to refine the image generation results is a big minus as for now, IMO 🙂

[haughty brook]

you are still able to generate an image with a direct prompt by clicking on this symbol

[blazing silo]

I just found it. tnx

[shrewd knoll]

They‘re waiting for Gemini Ultra to upgrade

[shrewd knoll]

Waiting for Gemini ultra api public

[shrewd knoll]

Now is April but still no Gemini ultra

😢

[delicate pike]

can claude opus be used with pro search (search results fed into model?). also re: above, will gemini 1.5 ultra have the 1m context window, is there any way to get ahold of that, i can't get that even with google subscription

[vapid maple]

what is this role unlocked?

[velvet tangle]

Just for our bot, nothing special 😉

[vapid maple]

hmmm, im pretty sure that user is a human being

[latent totem]

In setting proxy, what domain names should I use?

What are the domain names of Perplexity App

[fluid lark]

I think it's just perplexity.ai, according to my dns logs

[versed magnet]

where is this?

[last current]

what do those symbols in red mean?

[versed magnet]

I think the left one is how many questions/answers are in the thread, and the one on right is how many times someone else with the link has asked a follow up question (creating their own fork of the thread)

[velvet tangle]

17 messages within this thread, once the thread was forked 🙂

[last current]

coool thanks

[versed magnet]

same info, but far more concise 🙂
I can be too verbose aha

[last current]

if there was uploaded documents from a original thread, can forks access them?

"query" them

[versed magnet]

@last current I'm not sure, but fwiw I don't think so, in terms of any uploaded files specifically. But the contents of the context window (i.e. each query/answer in the thread) is there, and can be queried

[velvet tangle]

No, only the thread content, not the file.

[plain ruin]

Not sure if this has already been asked.
But will the API include dbrx-instruct?

[velvet tangle]

Initially, the model is only available on labs.perplexity.ai. If something changes, it will be announced 🙂

[plain ruin]

Perfect, thank you!

[shrewd knoll]

When will the Gemini ultra api be public?

[hexed light]

I'm struggling to sign in with apple on Perplexity web (iOS app sign in is fine). I get bounced back to the login screen after signing in on apples website. Are there any known issues?

[haughty brook]

could you try to sign in via email with the email on the ios app and check if that works

[hexed light]

Yep that does work

[vapid maple]

lot of normal users have this role. the one above your msg has it

[fluid lark]

what the

I didn't even notice

no clue how I got it

[hollow ginkgo]

@denese._. warned

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

Errors

DM not sent: User disabled DMs.

[modern grail]

If Perplexity wants to compete, it needs to change this. The search functionality is superb and works great, but the AI world is evolving. I switched from ChatGPT to Perplexity mainly because of the availability of Claude 3 Opius.

[versed magnet]

ohhh it's on discord.. i see now - thanks!

[fluid lark]

will i get banned for using 3rd party apps and userscripts? i'm working on an extension for pplx and i don't really want to lose my $20 over this

tos just says about automated requests, which the extension certainly isn't doing

[slim night]

You can use our API for the extension development purposes. Automating requests to the main site is against the tos.

[obtuse flare]

I'm getting this error every time I try to make a query on my pro account, but if i sign out it's working fine:

[valid sun]

Hello! When i try to ask smt i get this

[valid sun]

We are twinning lmao

[obtuse flare]

lol

Looks like its only happening with the pro queries

[valid sun]

It is

[upbeat niche]

Anyone have a good video on where to start with setup and more advanced features?

[hollow hound]

for PPLX?

[upbeat niche]

Yes.

[small prairie]

What about only visual updates?

EG stuff like this:

Instead of this:

Btw, starting a new search, the recommendations are not aligned properly with the input field:

I think it's a width problem, since it happens if I change the width of the screen.

yep, for md and sm screen sizes, it looks like the above. lg looks fine though.

Another problem is when clicking the input field, the html element gains style="overflow: hidden;" which makes the UI jank.

It's because the scrollbar is shown for some reason by default, and when you select the input field, it disappears, because of the overflow changing.

You guys need help with the UI just ask. I can skim through it and test it in different modes/sizes to see where the problems are.

Oh, pressing the download button at the bottom right when having the sidebar expanded also changes the overflow behaviour.

Same with the help ? at the bottom right

[next zodiac]

hello, is there any information of what is the temperature of Claude 3 and GPT-4 used in perplexity?

[charred void]

hi, wanted to know. does perplexity ai store my info that can come up on a plagarism check. and which model is best for research paper searches

[fluid lark]

It won't appear on plagarism check tools, but will be detected as AI on zerogpt

Claude 3 Opus seems to be the most accurate

[clear cipher]

wich pro model is better for coding purpose?

[velvet tangle]

#👑│pro-lounge message

[dense canopy]

Can we have rectangular images back on image gen, please?

[clever grove]

This was reported in #1111765634267742408 but I'm curious if there are any plans at all for PPXI to support no-drm epub uploads instead of making us convert to PDF first.

[slim pollen]

For Claude 3, I've noticed that all 3 poetic flavors of Claude are available on my PC but not on my Android phone. The phone limits Opus to 5 answers per day. Is this normal... maybe a temporary situation?

[velvet tangle]

Only temporary, you can set the model on PC and use the app with this model

[slim pollen]

Do you mean that if the PC is set to Claude 3 Opus, the phone will use this setting without the 5-answer limit?

[velvet tangle]

Yeah, just choose the model on the web and ignore the settings in the app

[slim pollen]

Sweet! Thanks!

[olive verge]

So can I not create images using the android app if I can how?

[ionic sparrow]

only over browser app on your phone

[hot thunder]

I use web app in all devices. Seems most stable

[hollow hound]

Haven't used it much, but I'd assume you could just have PPLX open in Chrome on android to get the full desktop view, then make the model/profile changes, then swap back to the dedicated app for actual use. bonus points for exporting the page as a shortcut so you can just tap on it like an app to open the web version directly. i assume most people have PCs/Macs/chromebooks to use separately, but why not just use the 1 device. also allows you to quickly change the model without needing a 2nd device.

[rose olive]

How do I enable 2FA on my Perplexity account?

[small prairie]

Don't think you can, since it uses 0auth2 and not a username and password for logging in.

[fluid lark]

what

it has username and pass too

or is it only email

nvm, it's only email

shouldn't 2FA still be possible? ig they just haven't implemented it

[small prairie]

Is 0auth2 not secure enough for you?

[rose olive]

Thank you. Your thoughts on this ? https://www.perplexity.ai/search/How-to-enable-xKYJXHBQRGyjJFC.GnXdaw

[small prairie]

2fa just seems like too much when it's a service that doesn't deal with money, and increases the friction to login.

[rose olive]

Security of my AI conversations matters a lot to me. There's a lot of private stuff I'd prefer to keep that way.

[small prairie]

Unless it's a local model, I just assume it's not private. Especially since it's going to the API's of the model creators too.

Hopefully Lamma3 is good enough to replace most needs.

[finite current]

Pro user here: does anyone else's library not sync between browser and app?

[lavish ocean]

What is it about moves like the frequent token restrictions that occur?
When I fill in more than a certain size on the prompt input screen, I can't press the submit button.
Does this happen to everyone?

[haughty brook]

your library should sync between apps/platform. please make a thread in #1140622008086970420 if you’re 100% sure you’re logged in to the same account on both platforms

[clever abyss]

any1 using perplexity for coding? Sometimes I ask chatgpt to generate test for some function, is perplexity could handle it given the context?

[narrow robin]

Context:
I have created a collection that helps to create bullets for a resume.

Right now, I have to input the information one at a time to give me one result. So I input the action, then a metric, then the impact of that action and it creates a formulated resume bullet.

Question:

Is it possible to create a collection that analyzes all of that information in an excel or .csv file and spits out all of the newly formed resume bullets?

For example: Column A would be the action, Column B would be the metric, Column C would be the impact of the action. And then I would want the AI to evaluate them row by row.

[shrewd knoll]

The update of the app make it worse

It keep shutting down

Click profile and shutting down

[haughty brook]

Please create a thread in #1140622008086970420 and if possible a screen recording of your issue and information like the app version installed and phone model.

[vocal rampart]

Do we have image generation limit ?

[ionic sparrow]

50/day

[analog solar]

Why can't I log in to the IOS system? I logged in using my email and opened the link in the email. But after jumping to the APP, I did not log in. Still not logged in

[lyric sun]

I'm (finally) getting this screen.
I cannot, and I repeat, I CANNOT access Perplexity without a VPN (I don't know the exact reason, could be blocked by China, blocked by my ISP, or blocked by Cloudflare. Yes I'm in China). Is there nothing I could do, other than potentially trying another VPN that happened to have not been blocked yet?
If there is no way out - can I request a download of my data on Perplexity.ai? Can I request delection of my data on the server?

[ionic sparrow]

switch country on your existing VPN provider, if that doesn't help, change provider

[haughty brook]

Please try again by copying the login link and paste it in a private tab in safari.

[lyric sun]

That did help in the short term, thanks.
I wonder what happens in the long term where most or all servers of my VPN, and/or most or all VPN providers, getting blocked.

[ionic sparrow]

no problems like that on Proton

[lyric sun]

How do you know, if that's a VPN, and VPNs are to be blocked? Is there some special case where Proton servers are to not be blocked?

[ionic sparrow]

tbh on most mainstream providers..

[lyric sun]

Oh. ... I'm on a small (but importantly, cheap) provider hosted by some Chinese people.

[ionic sparrow]

Proton has free version 😉

[lyric sun]

Oh.
Guess I'll keep that in mind in an emergency. (I assume it has a speed limit on the free version? Also it restricts to 1 device)

[ionic sparrow]

you'll have to test it yourself

[lyric sun]

Doesn't work: it's blocked or otherwise unaccessible.
(Small VPN providers tend to fly under the radar, not so much for such a large thing.)

[velvet tangle]

Small, cheap providers are often used for abuse 🙂

[small prairie]

mullvad is probably the best. Even better, is it through tailscale.

[vapid maple]

but nordvpn offered 120% cashback . you earn money by buying

[small prairie]

If you are not paying, you are the product...

[twin crow]

How do I get the AI to look at more than 3 sources that I've supplied?

[narrow robin]

Is there a better place for this question?

[formal oriole]

Pro search maybe?

[twin crow]

Didn't work. I just tried a random wikipedia page I had opened and it only used the first 3 links

[versed magnet]

try adding "site:" before each url

[twin crow]

Did nothing

[versed magnet]

"Short summaries:"
site:https://tvtropes.org/pmwiki/pmwiki.php/Characters/GravityFallsMysteryTwins
site:https://tvropes.org/pmwiki/ pmwiki.php/Characters/DesperateHousewivesHousewives
site:https://tvtropes.org/pmwiki/pmwiki.php/Characters/TeenTitans2003Raven
site:https://en.wikipedia.org/wiki/Universal_Mind_Control_(song)

[twin crow]

Cool but it's not that practical when I'm trying to use the sites to make a character

[versed magnet]

the irony that this is for some kind of creative writing thing..

why don't you experiment and play around - be creative

learn how to use ai, rather than expect it to dish everything up on a sliver plate

[twin crow]

Because if I wanted to do that, I wouldn't be wasting hours of my life on a concept that I'll never use in the future. It's just what I waste my time on

[versed magnet]

I don't know what you're saying - you just want stuff to magically work without putting any effort into it. And if people attempt to help, rather than considering exploring what they provide, you just say: doesn't do what i want

you were complaining that it won't use all the URLs it provided.. I showed that it can

I have no idea what you're trying to achieve

but meet evelyn...

and try actually doing it yourself

https://www.perplexity.ai/search/Create-a-unique-LPL8jfwARemqOf0q1DvC5w

[twin crow]

https://www.perplexity.ai/search/It-would-mean-c25bF_NNSpOrWrEyV0HOZg#2 Here's what I'm doing. It's after a few messages of having the AI try to understand how to make a realistic and interesting characters

The fourth link would be something else if it could actually detect it

It may slowly form into something after I see some ideas

[exotic ether]

Does anyone know of a tool that allows you to train Perplexity on private data (website blog for example) and share this publicly? This would be similar to ChatBase but with perplexity.

[versed magnet]

meet raven pines
https://www.perplexity.ai/search/Create-a-unique-rC74OYVfShi6CHEq7UgXCg
anyway, good luck

[shrewd knoll]

Gemini Ultra will probably be released in May

😢

[narrow robin]

Nevermind, figured it out

[serene dagger]

How do I get the new Discovery view? Still looks like this for me. iOS app updated this morning.

[sudden rain]

Does anyone know if there is a way to allow the voice chat to continue with using other apps?

[ionic sparrow]

#💬│general message

[carmine meteor]

how to generate images in perplexity ai, I am a pro user, perplexity ai image genaration is awful....please help me

[ionic sparrow]

#1194794305362071552 message

[serene dagger]

I have 2.17.1

[ionic sparrow]

please make screenshot of the problem and post it here https://discord.com/channels/1047197230748151888/1225729143250092072

[broken tendon]

Can you have voice conversations back and forth with this app instead of texting

[ionic sparrow]

#💬│general message

[broken tendon]

Thank you. Can I do this on a free version?

[ionic sparrow]

free only STT

[lunar juniper]

How do you guys use different AI models differently? Like ChatGPT, Gemini, Perplexity, and Claude, etc in your lives and work?
So far i only use chat gpt and perplexity but I want to be able to maximize my effeciency by using other models to help me with creative work like writing

[chrome crystal]

I only use claude with perplexity atm.

Use it to help with figuring things out with code, using it as a teacher to quiz me with trick questions, and explain things like I'm 5.
Off loading administrative tasks to it, like sorting things, and writing emails for when I need to contact companies for refunds etc.

Not that special, just basic things.

I used to use GPT4, but I found that I needed to correct it a lot, sometimes mistakingly, and then it would simply agree even if it wasn't correct.
Claude will tell me I'm wrong, when I'm actually wrong, which is v useful.

[modern grail]

I've read a lot here and I'm totally confused now. What is the actual context window size in Claude 3 Opus on Perplexity? Is it 200k or is it limited to 32k?

[gilded cedar]

hm, im also interested in knowing that. but i know that the biggest answer he could give me was a 3,8 thousand tokens (i measured cause he stopped middle answer and i was curious). that would be just a dream with chatgpt because of how often it used to crash when i had it

[modern grail]

But that’s the output limit. You can write „continue” and you’ll get more. It would be good to know just how many tokens this model could process at once

[gilded cedar]

indeed

[versed magnet]

the input limit is effectively capped at 32k tokens because of the file upload system perplexity, but the models can and do process all the tokens in their context window at once, which can be up to 128k tokens for GPT4-t and 200k for Claude 3 models

at least that is my understanding. I have not seen any actual evidence that the LLMs served by perplexity have 'shortened' context windows or that the context windows are 'dynamically' adjusted

But I have seen plenty of evidence to show that file uploads are limited to 32k tokens

It's convuluted I know - maybe copy and paste that response to an AI and ask if it makes sense and to explain it better

But basically, if you can circumvent / avoid file uploads and get the tokens in there, you get the full context window - and all LLMs process the entirety of their context windows to generate every response
see: https://www.perplexity.ai/search/I-will-provide-.MaKRen9TQumbfMQF0CVKw
also: https://www.perplexity.ai/search/continue-do-not-piVAIc_FS6iAfCadkLD6Ew

[modern grail]

Thank you very much for this info!💪

[dark basin]

I assume the shortness is due to them wanting to make sure there’s enough context window for the search results?

Do we know if we use “Writing mode”, it unlocks all the tokens for our input?

[versed magnet]

at the start of every thread, think of the context window as empty, and let's say has 128k tokens of capacity - meaning that's how many tokens can fit it in there. There are no tokens to be unlocked, it's basically empty. The limits, with perplexity, arise with how you are able to fill the context window with tokens (text / files etc) - you can't type more than a few thousand words/tokens into the textfield, and uploading file/s is capped at 30k tokens

empty context window

Here's how the context window might get filled with tokens using perplexity, just for example (it's messy and not meant to be proportional, just representative of the process (in my mind anyway)

Using Writing mode instead of the web search does not really change anything in terms of these input limitations - all it means is that there won't be the web sources (yellow) also using tokens, so you can have longer conversations etc, but the limitations remain unchanged

Using the API, OpenAI limits outputs at 4096 tokens, but not inputs - you can dump a 100k token file into the context window, then just say "summarise", and it'll do it (but the response won't be longer than 4096 tokens - if that makes sense)

I'm not sure about output limits with Anthropic, but in terms on input, the same applies - you could dump a 199k token document into the context window and say "summarise" and theoretically it should work. This screenshot also does a good job explaining the idea of "completions" (which is what every "response" you get from an AI is), and the fact that the model uses the whole context window, starting from the beginning, to generate every completion - which is quite fundamental to how LLMs work to my mind but seems quite misunderstood

anyway, that's just my understanding.. hope it helps - but bear in mind that I don't have any expertise or anything. I'm just interested in the technology and enjoy learning about it

[stark patio]

did you ever figure it out?

[jaunty solar]

I use perplexity on mac/chrome. Can I somehow configure "pro search" to be always on? It works in the iPhone app, but not on mac. Thanks a lot in advance

[dark basin]

Thanks for the thorough response 🙏

[hollow ginkgo]

Hey @dark basin!

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

[terse fern]

Hi, longer text is saved in files named paste.txt - but how do I delete them after using it?

Hello? Is anyone here?

[haughty brook]

currently it’s not supported to delete files used in threads

[terse fern]

@Spezi but the content of this paste.txt is disturbing other threads. It's not good.

[haughty brook]

You probably mean your other questions in the same thread. The paste.txt you pasted can only be read by the AI in that particular thread. That is just how it works currently, but they will improve this in future updates #1221790602895556628 message.

[versed magnet]

other threads (i.e. other conversations to which you haven't uploaded/pasted anything)? That'd be weird.. though if it's the same the same thread, this has been documented (bit unclear whether a feature or bug though ha) #1221790602895556628 message

oh, as spezi said and linked lol

[terse fern]

I open one more thread and with the second or third question the paste.txt is being read repeatedly

it's not only one thread

[versed magnet]

after you've uploaded though, right?

it's not like a file uploaded to a previous conversation/thread, is being used in a new thread without it being uploaded, just to clarify?

[scenic scaffold]

Hi there- I have use case to retreive topic specific data based on custom data trained from several PDF and txt files on each of different topics, say-

Topic A- few PDF and txt files (more than 5)
Topic B- differnt PDF and txt files. (more than 10)
etc.

I need ability to query content for Topic A using thread specific to Thread A. Then need switch to Topic B using its own different thread (having Topic B specific earlier uploaded PDF and txt) for answers.
I do not want model to answer based on online resource, it's should be limited to uploaded documents only. There should be ability to go back to thread on Topic N later after few weeks later, uploading more docs and use it.
How can I implement this using PRO and what prompts should be used to effectively manage threads keeping topic separate discussions on it's own thread with its set of trained PDFs/txt. Thanks

[severe fable]

whats the maximum search queries for claude 3 opus?

[ionic sparrow]

#🌐│feedback-web message

[severe fable]

in one search

[solar magnet]

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

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

[terse fern]

brknclock1215 asked: it's not like a file uploaded to a previous conversation/thread, is being used in a new thread without it being uploaded, just to clarify?

clarification: I opened a thread and uploaded three paste.txt files. Then I opened another empty thread without uploaded the files once more. (This thread should be of another topic.) At second or third question Perplexity used these paste files again to generate a reply. But why?

[versed magnet]

oh I see - thanks for clarifying! If you create a new thread (i.e. start a new conversation) and don't upload / paste any files, I believe it should be like a 'clean slate'. That's really weird (and kinda concerning tbh) if the material from files uploaded in one thread is being used / referenced in new threads that you later create...

Perhaps if you're using a Collection, there's something related to that - but even still, that's hard to understand how / why your file uploads are seeping into new threads..

[haughty brook]

Haven't heard of this bug before. Could you create a thread in #1140622008086970420 and share your thread where this issue occurs. I thought yesterday you meant deleting a file in the same thread you uploaded the file.

[terse fern]

Spezi: you find it here: https://discord.com/channels/1047197230748151888/1221790602895556628

It's very simple to reproduce, please try it

[vocal rampart]

What is the limit of image generations for perplexity pro users ?

[haughty brook]

50 per day

[terse fern]

Spezi, where did you find this specification?

I remember a day where I made more than fifty as pro user

[viscid trout]

Hey,
I have a question to all. I did redeem Nothing 2A Perplexity Yearly Offer . I chose monthly plan instead of Yearly plan. So, Would it expire in 1 month or 1 Year?

[velvet tangle]

Perplexity uses a dynamic system for returning usages. This ensures that your usage is always available when you need it.
This is how it works:
You ask a question at 6 o'clock in the morning, and one usage is deducted from your limit. Exactly one day later, at 6 a.m., that usage would be available again. (Same for image generation)

[haughty brook]

this can be found in the requests when visiting your profile

[frank jasper]

I'm having issues redeeming the promo code I received from purchasing the Rabbit R1.

[small prairie]

I think promo codes are currently disabled, but you can try talking to support so they can apply it.

[frank jasper]

Perfect. That's the info I was looking for. Thanks.

[young holly]

how can I export the list of references used on a thread? preferably being able to select the format...

[unkempt wolf]

I read that there are plans to change the default model sometime this quarter. Will we still have access to the current default model when that happens?

[reef topaz]

Is there any way to make pro mode not ask followup questions by default? I can always skip them but it would be better if we can turn off followup questions off

Edit: typo

[haughty brook]

you should turn off pro search if you don’t want the followup questions https://www.perplexity.ai/hub/faq/what-is-pro-search

[reef topaz]

But then I won’t be able to use features such as internet search and claude/gpt4 models

[slim night]

Internet search is off in the Writing mode. Claude and GPT models are also used.

[slim night]

Hey, @unkempt wolf! What are the rumors about? The default search is available to all of the Perplexity users.

[slim night]

Hey @young holly! The copy button will copy each answer along with the sources.

[reef topaz]

So i only use pro option for if even I want to ask follow up questions based on chat, otherwise simple queries can use pro off

[slim night]

You can turn Pro on and off whenever you feel like more clarifications or sources will be needed. If you don't expect any clarifying questions and your question is pretty clear, then Pro search off will work faster. You can also rewrite the answer with a different model if you feel like it'll give you a better answer.

[unkempt wolf]

Hi Alex, I read an article that suggests that perplexity is looking to move away from the gpt3.5 + llama2 hybrid to something open source. So I was wondering if that switch happens, will we still have access to the old gpt3.5 + llama2?

[solemn tangle]

Hi, someone use the AI prompt on collection? I tried, but nothing change.

[keen talon]

Hi guys, is there still a cap when using Claude 3 Opus?

[velvet tangle]

No there is no cap, you can use all advanced models in total 600/24h

[velvet tangle]

What did you write in your ai prompt? Could you please share some threads?

[solemn tangle]

Hi @velvet tangle, I tried to ask if can do web search on a specific language website (Japanese). But doesn't work. Actually I also tried with normal prompts but it still doesn't work. If you ask the question in the specific language it works. I read on a Reddit post that some people have done this in this way (ask Perplexity to translate the prompt, do search, and answer in English or other language) but it doesn't work for me (I use Italian).

[signal cairn]

Anyone know a good prompt for finding recent/ relevant news on a topic? Right now I'm using Feedly, but I like the idea of a more well rounded breakdown of the latest in my industry

[velvet tangle]

This is very difficult to achieve and does not work for every language. All system prompts are in English, so in most cases the AI will try to do everything in English.
You are welcome to share your exact prompt and also threads where it did not work and describe again exactly what the result should look like, maybe something can be improved ☺️

[solemn tangle]

Yes, I did some testing, maybe I found a result, I'll post it here as soon as I'm sure. Thanks 🙂

[odd harbor]

Hello, I subscribed to Pro but I don't see how to use image generation? hmm, it seems you have to search first? Huh? Why?

[velvet tangle]

#💭│ideas message

[fluid lark]

Is there a way to get PPLX on Arc Browser search box instead of ChatGPT? I've heard people are doing it somehow

[slim night]

Hey, @fluid lark! It should be one of the default options.

[fluid lark]

I've just found it, thanks. Can it be set to search with Pro mode? The default one misses the point of query often

[hexed moss]

Hi everyone. When using Focus, is it possible to use a Collection of Threads? I see it’s not an option currently but maybe a more experienced user has a way to do it? Thanks!

[verbal pollen]

Hi! I'm looking to invest in a yearly subscription. However I need to know one thing before. Is it possible to lets say use a github repo or manually upload files as embeddings to a model and/or fine-tune it with data?

[fluid lark]

No, you can just upload up to 4 files per query

Not sure if it's what you mean

[verbal pollen]

alright, i basically want to create a model that knows everything about my coding project

[fluid lark]

You're better off using Cursor IDE or Github Copilot

[verbal pollen]

they dont do that either :/ ill keep looking

[fluid lark]

Aren't they answering based on the current codebase? Did you try GH Copilot with @workspace?

[verbal pollen]

Well, actually both works for my needs but to an extent. Never tried @workspace though I'll read up on it.

I'd like to fine-tune the model however to only answer based on docs I've provided it and it should always know that it's for a specific language etc

Copilot gives answers based on other peoples code, which makes its solution very inconsistent

[fluid lark]

@workspace is just a command in Copilot Chat in vscode, and it seems to be much better than the inline chat/completion

[fluid lark]

Try finetuning gpt4 turbo yourself with api, if you can spare $10. Maybe it's not really feasible to do that

[verbal pollen]

that's what its leaning towards, would be cool if I could do this with Perplexity though thats why I came here to find out. As Perplexity as a whole offers much more than openai

[terse fern]

How can I find my mobile browser threads on my PC browser?

both of them are Chrome

[latent topaz]

Is Claude 3 Opus still limited to 5 queries per day? If not, what is the new limit?

[vapid maple]

no, same limit as gpt4

[latent topaz]

I want to know whether I should cancel my Claude subscription or not because if Perplexity doesn't have a limit for Opus, then I will cancel

Oh so it's 40 messages per 3 hours?

[vapid maple]

a limited context (32k) on uploaded file

[latent topaz]

Okay so that's the only difference?

[vapid maple]

i meant the same limit as gpt turbo on pplx, 600 every 24 hours in total

[latent topaz]

Ah okay I see thanks!

[vapid maple]

the offical statement

[shrewd knoll]

Once Gemini Ultra API is released perplexity will finally get Gemini ultra

I want Gemini ultra

@velvet tangle Gemini 1.5 Pro API is now public is perplexity gonna add it or they will get Gemini Ultra only?

Is Gemini ultra token set to 32k too? Like opus?

[shrewd knoll]

Gemini ultra is best for creative writing

Opus Ultra is Best for information writing

Get both and it will be interesting to craft a novel

I want it so badly right now

[modern grail]

I don't know what to think about this Perplexity. The search results are often cool, but it lacks other features that are present elsewhere. Often when searching for something, one would like to generate an image normally (it’s often needed) but it's not possible here; you have to click on something separate. When it comes to voice communication, it's also not implemented the same way as in ChatGPT. I think OpenAI will catch up with this search functionality. Google will too, eventually. We'll see how Perplexity develops. And the context Window in Claude 3 Opus is a mystery. Have no idea why keeping it secret.

[shrewd knoll]

But it write 32k token of claude3 opus

[modern grail]

I must’ve missed those info. I havent seen any official info here - only user’s assumptions 😉 Please correct me if I’m mistaken

[small prairie]

Minimum 32K was confirmed.

[hollow jasper]

is there a way i can change my email for my account in the dashboard?

[steady vortex]

it’s in settings I think

[slim night]

There's no option to change the email for the account in the settings. Please send us a request with all the relevant details to [e-mail], we'll help

[lofty pilot]

https://x.com/OpenAI/status/1777772582680301665

Do we know if we are getting the new GPT4Turbo starting today in pplx?

OpenAI (@OpenAI) on X

Majorly improved GPT-4 Turbo model available now in the API and rolling out in ChatGPT.

Twitter · Apr 9, 2024, 6:56 PM

[formal oriole]

I am thinking so since I saw something where it just routes to the new one anyways. Not sure though.

[formal oriole]

Nvm it’s not lol https://www.perplexity.ai/search/What-is-the-DI.rc2EBQPK7phj6xFnwtA

[swift geyser]

In IOS app after pasting a bit larger text, I cannot click on send. Even after struggling to hide the keyboard. there is no sending button. Does someone know how to use the app properly? 😂

[swift geyser]

After an amount of text the blue send button just goes lower and lower, out of the picture

[severe fable]

is it possible to create some sort of user friendly thing so whenever I type in a prompt into perplexity, itll automatically utilize claude 3 opus using my prompt improver prompt to improve my prompt?
and could opendevin do that?

[haughty brook]

its a known bug and will probably be fixed next release #1227312488526385212 message

[crude perch]

How to force mixtral-8x7b-instruct to answer in certain language, I ask him to answer in specific language on both system and user prompt but sometimes he keep answering in English. Is there a workaround ?

[shrewd knoll]

@velvet tangle @velvet tangle Gemini 1.5 Pro API is now public is perplexity gonna add it or they will get Gemini Ultra only? https://x.com/OpenAI/status/1777772582680301665

Do we know if we are getting the new GPT4Turbo in pplx?

OpenAI (@OpenAI) on X

Majorly improved GPT-4 Turbo model available now in the API and rolling out in ChatGPT.

Twitter · Apr 9, 2024, 6:56 PM

[shrewd knoll]

Will we even ever get a response?

[vestal shadow]

is perplexity using the new gpt-turbo model?

[slim night]

Hey @crude perch! The only ways to instruct it are by prompting or the system prompt.

[fluid lark]

anyone have a prompt for perplexity labs to make it always give citations ?

[fluid lark]

🥺

[formal oriole]

I’m assuming you already checked #1124060874785685554 ?

[fluid lark]

isn't that only for AI Profile ? and I just checked it and couldn't find anything like that

[formal oriole]

Can be used for anything 🙂 some use for collections or AI Profile. But yeah I don’t have one specific for citations. I just have prompts that also ask for citations

[fluid lark]

hmm still couldn't find one that would give me citations

[versed magnet]

something like this might work:

SYSTEM: """Append a list of URL citations to each response."""
ASSISTANT: Understood
USER: {your_query}
ASSISTANT:

But I wouldn't imagine it is particularly reliable.. and when it does provide URLs/citations at the end, the quality of the answer wouldn't be as strong compared to if it was just the query

[fluid lark]

thanks, I'll try it

[versed magnet]

np 👍

[fluid lark]

well, I would still prefer citations

[versed magnet]

ha yeah I'm with you

impressive as they are, still can't just rely on whatever an LLM spits out for anything of consequence/importance ha

[sinful veldt]

How to change that default Focus of new conversation to "Writing"?

[velvet tangle]

The default focus is always “All”.

[small prairie]

You can do so by adding &focus=writing

To the url

[sinful veldt]

GREAT!

thx

[small prairie]

Np

[brazen star]

Hi, I currently cannot log into my pro account by using the email link. My email account is not able to receive any login link today, and I tested that my other email accounts can still receive the login link. It is so weird. Can you take a look into this? thank you

[slim night]

Hey, @brazen star! Please send me your account details, we'll check

[shrewd knoll]

Why still no Gemini? 😭

[shrewd knoll]

Will we get Gemini in May at least?

[formal oriole]

Maybe waiting for Ultra API. Current choices for Gemini are not that great imo but I’m sure it’s possible

[shrewd knoll]

I hope so otherwise I'll be very upset 😔

[north moon]

Isn’t Gemini supposed to be the kinda crappy ai from google

[shrewd knoll]

Gemini is best for creative writing

[north moon]

Hmmmmm

I could possibly use that then

[shrewd knoll]

You can use this for creative inspiration

[north moon]

I could use it for helping me write YouTube scripts

Cool

[shrewd knoll]

Use Claude 3 opus for writing information and Gemini ultra for creative writing

Imagine with this ai You can even create a novel

I want to write fanfic using this two ai model

And gpt-4 turbo in case of Claude 3 censorship

Total 3 ai

[short oxide]

Hello, @slim night, I signed up for Perplexity Pro's 7-day free trial yesterday but it was canceled today, perhaps by mistake. It'll be awesome to get the remaining days back. Any thoughts? Much appreciated.

[glacial arch]

@slim night Hey Alex,I wanted to experience the 7-day trial, but I inadvertently subscribed to the annual membership, which was not my intention. I do not wish to continue with the yearly plan.
Could you please assist me in canceling this yearly membership and processing a refund? I did not mean to commit to the annual plan, and I would greatly appreciate your help in rectifying this mistake.
I have already sent an email to your support team regarding this issue, but I have not received a response yet. My email address is [e-mail].

[velvet tangle]

Hey @glacial arch!
Please be patient, the response time is several business days 🙂

[glacial arch]

Thank you for your response. I understand it may take a few days.

[crude perch]

Is it possible to go in production with perplexity API? Wanted to switch from openAI api to perplexity and just noticed the rate limits are extremely low. Is it possible to reach like 10k requests and 20M tokens/min ?

[velvet tangle]

Please apply here

[pure bay]

Напиши историю Серафима Саровского, на русском языке

[velvet tangle]

-> https://pplx.ai

[spare ivy]

Seems i'm having problems getting the link from my Pro account in perplexity to trigger the role in Discord. No urgency, but I've tried leaving and coming back without any change.

[vapid maple]

#📘│faq message

[spare ivy]

maybe that link expired again ^^ @slim night

[slim night]

The link is in the Perplexity Pro settings. I see that users are joining and getting the role

[spare ivy]

@slim night ok i'll give it another go now

@slim night worked that time. Thank you sir

[onyx surge]

Guys, hello! I have such a general question - where can you now read up-to-date information about LLM models, see their comparison in speed, in tokens, in their intelligence, etc. ?

[short oxide]

The issue has been solved. Many thanks to @slim night, Lilia, and the whole team at Perplexity.

[dapper jay]

I've found Claude 3 Opus to be best

[shrewd knoll]

Claude 3 opus is Only for writing information And not creative inspiration

[dapper jay]

Well it works damn well at creative writing

[shrewd knoll]

https://www.perplexity.ai/search/Based-on-Ironman-yI.f0q9jS.63gJQEo9na9w

Scene: A dimly lit, high-tech workshop tucked away in the labyrinth of an undisclosed location.
Action:

• Bruce Wayne, in civilian clothes, stands hunched, meticulously studying an advanced piece of technology.
• A sudden whoosh of air signals the arrival of Tony Stark, Iron Man armor partially retracted, a smirk plastered on his face.
  Dialogue:
  Tony: "Well, well, well. If it isn't the brooding billionaire. Fancy meeting you here, Bats."
  Bruce (without looking up): "Stark. Your reputation for theatrics precedes you."
  Tony: "Can't blame a guy for having a little flair. So, what's got the world's greatest detective slumming it amidst my toys?"
  Bruce (finally turning, eyes narrowed): "Your tech's been popping up in places it shouldn't. Trafficking rings, black ops... not exactly your usual brand of philanthropy."
  Tony (defensive): "Someone's gotta clean up the mess even genius inventors leave behind. You wouldn't understand."
  Emotional Detail:
• Bruce: Suspicion lingers in his furrowed brow, a flicker of disapproval in his eyes.
• Tony: A mix of arrogance and genuine hurt mask the concern brewing beneath.
  Sensory Detail
• The workshop hums with the quiet energy of dormant technology.
• The faint scent of metal and ozone hangs in the air.
• Stark's repulsor tech glows with a soft blue light, contrasting the shadows draping Bruce.
  The tension builds as their clash of ideologies and methods begins to surface...

Gemini ultra ☝️

[dapper jay]

I'm glad you're happy with Gemini

[shrewd knoll]

@dapper jay I tried a comparison Claude 3 opus and Gemini Ultra is still the best for crafting stories

[shrewd knoll]

No I'm not because perplexity.ai haven't release the Gemini ultra yet 😭

I hope we got in May at least

[fluid lark]

Is the latest version (as of April) of GPT4 turbo being added to Perplexity Pro?

[gilded cedar]

i just entered the chat to ask the same thing 😂 . Hope someone knows the answer

[keen forge]

I believe not

[gilded cedar]

if the bot is not trolling, it is the latest gpt-4 turbo. it gives the same answer as in chatgpt pro according to what i saw in twitter

[gilded cedar]

@fluid lark

[gilded cedar]

hm. i tried replicating the question several times and it did answer 2023 everytime

[opal pivot]

/gay fweet

/gay feet

[half swallow]

Perplexity iOS 全代理了也登录不了，新加坡ip、美国ip都显示连接超时。网页版能上，这是为什么？

[vivid violet]

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

[floral wigeon]

How do I summarize Youtube Videos in perplexity?

[mighty tiger]

Hi, I work for a fairly small company, and we are slowly starting to implement Perplexity in our work. Please tell me if it is possible to organize a team subscription for us (a small number of people, for example - 4-5 people to begin with)

[main shuttle]

I have a Perplexity Pro account, but forgot my login info. It isn't saved in my password manager. How can I recover my user ID and Password?

[velvet tangle]

Perplexity uses login via Google/Apple or via Email - no password!

[main shuttle]

But I don't remember which option I used when I created my account. How can I recover that?

[rain wadi]

I want to gift my brother perplexity pro. where can I buy his sub?

[velvet tangle]

Just search in your mail programs for "Perplexity" or try out all options.

[velvet tangle]

It's currently not possible to gift subs.

[rain wadi]

crazy. ok

[vapid maple]

max 4 per query, 8 files per thread. max 1000 files every 24hours

[wicked lynx]

Can we pause the subscription? I still have 1 referral, will it be used when I resume the subscription?

[shrewd knoll]

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

Just use this code in case your referral is finish, the code can be used for new and old user

[shrewd knoll]

If you cancel your subscription, you will still have the old referral but you won't get a new referral if people use your code until you resubscription

So Don't worry

I love using Claude 3 opus to interpret the thriller movie

So confusing and the ending made me even more confused

https://www.perplexity.ai/search/Interpreter-gain-insights-JN9krgGoQ36OrPAJsUrrCg#6

I need Gemini Ultra and Claude 3 opus To be satisfied for inspiration and information

I want to write fanfic

[warm grotto]

Is it possible to get rid of an image from a chat? It keeps trying to reference something completely irrelevant

[shrewd knoll]

You can try remove the website source

I hope perplexity.ai gives us Google like they used to

[warm grotto]

fixed it, thanks

[hollow ginkgo]

Hey @warm grotto!

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

[versed magnet]

gpt-4-turbo of some variant confirmed then ig ha

[shrewd knoll]

https://x.com/OpenAI/status/1777772582680301665

Recently released not sure whether will get or not

OpenAI (@OpenAI) on X

Majorly improved GPT-4 Turbo model available now in the API and rolling out in ChatGPT.

Twitter · Apr 9, 2024, 6:56 PM

[versed magnet]

yeah but that's the same model in the top row of that screenshot

Dec 2023 is its knowledge cutoff

[shrewd knoll]

That is gpt-4 vision you are showing that just released api not to long ago

They won't release early

[versed magnet]

that is their 'majorly improved' gpt4-turbo

unless we're talking about different things

[shrewd knoll]

Prove

[shrewd knoll]

They didn't write gpt-4 turbo vision no update because API recently released

[violet hornet]

Anyone know why Perplexity sometimes doesn't seem to know the information I included in my profile in settings? Is it only for certain focuses or prompts?

[slim night]

Hey @violet hornet! Please ensure that the Profile is not paused. Which model are you using? Could you share some examples where it doesn't follow your instructions.

[violet hornet]

Most recent example was with Claude 3 Opus. I did double check that it isn't paused as well.

In this example, I ask it for something relevant to my work, but it doesn't seem to know my name or where I work, despite me including those in my profile.
https://www.perplexity.ai/search/Please-help-me-TOwxp.MpRkKTf9xN7ph0Ww

[shrewd knoll]

I wonder next month will we finally get Gemini Ultra?

[shrewd knoll]

So many new models that publicly release API but no new model yet

[glacial horizon]

Is there a way to change the font size of the answer to a question?

[slim night]

Hey @violet hornet, thanks for sharing your example, I see. I also see that when you prompted the model to use your profile data, it put your name in the example email. The model reads the Profile, too, you can check it by asking to provide some details, but we've also got reports earlier that profile data is used in many cases where it's not necessary, so we have to balance it. And I'd suggest referring to your profile by prompting when needed, just like you did.

[slim night]

Hey @glacial horizon! Are you referring to one of the apps or the web version?

[violet hornet]

Thanks, that's totally fair! Indeed, I've had better luck when asking it to refer to my profile, but I wasn't thinking that was needed, but I suppose it doesn't hurt if I definitely want it to recall that info

[austere dove]

Hello,
How can I check how many messages left for today? I've checked every single part of dashboard on web and iOS and I can't find it.
I've got a pro version and I wonder how good that will work with opus3, because I was kinda tired that on claude chat I am out of messages each hour 😂

[shrewd knoll]

Claude 3 opus best for finding information

[austere dove]

I am using it mostly for code cleanup and refactoring and some code-related fixes. I am too lazy to check documentation or search for stackoverflow 😅

[shrewd knoll]

You can try it then

[jagged fiber]

Hi guys, a quick question - is Claude opus here and at claude.ai chat exactly the same in terms of knowledge, reasoning, coding capabilities?
(I don't care about speed, message limits or cost)

[austere dove]

I would like to know this either from a reliable source.
What I can say after using it for about an hour (while I've been using opus 3 claude chat for the last 3 weeks, at least 100 messages a day) is that I feel it's even smarter than claude.ai opus3 chat. I am not sure why.
I've sent a complex task related to fixing the problem with a RESTful golang API to both - claude.ai opus3 chat and opus3 on perplexity. I am not sure if it was just luck or something, but perplexity fixed the problem in one reply, while claude.ai opus3 chat needed another fix after the first reply.
Maybe that was just luck. I don't know. But from now on I probably like it even more than claude.ai chat. The fact that you are asked for some additional information before the answer is generated is just amazing. The articles/forums that are provided with the answer also seem to be a great way.

But I would also be interested to hear from a reliable source if the opus3 chat on claude.ai and the opus3 on perplexity are the same / which one is better?

[ocean saddle]

https://docs.anthropic.com/claude/docs/long-context-window-tips

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

#⚡│ask-community message

[cyan fractal]

used this referral code, thank you.

[vivid violet]

got it , thanks!

[versed magnet]

oh I see.. sorry wasn't saying that perplexity's gpt-4 is the latest version.. was trying to say that all versions of gpt-4-t have the same knowledge cut off date.. so getting "December 2023" as the response (as in the screenshot I was replying to) doesn't show which variant it is

[hollow ginkgo]

Hey @versed magnet!

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

[shrewd knoll]

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

Yup the code Can be used for new And old user

[versed magnet]

hmm not sure what word in my response triggered this lol

[shrewd knoll]

Perplexity.ai won't be released early because API recently public

[shrewd knoll]

Maybe the word I'm sorry triggered it

[versed magnet]

ha maybe. though "sorry" doesn't naturally align with finding something "helpful" 🤷‍♂️
perhaps it was "I see"

[shrewd knoll]

Could be

I'll be happy if Perplexity gave us Gemini Ultra but we must wait for next month because could be maybe Google will release it to the public the API

[vapid maple]

sry made a mistake. 4 per query, 8 per thread

[indigo tide]

It seems that the hover-over limit usage information has disappeared in the browser version. Has it been removed?

[vapid maple]

only shows when less than 100

[abstract kraken]

what temperature is used on claude sonnet and opus?

[severe fable]

does anyone having tips for it to not hallucinate all the time, its being lying like 90% of time recently

[small prairie]

90% of the time? What were the prompts on?

[slim night]

The sources are attached and you can check the information the answers are based on. Could you please share some threads, @severe fable, we'll check 🙏

[severe fable]

i tell it how I want it to respond and it uses that as an excuse to make things up to conform to me

https://www.perplexity.ai/search/statistics-on-media-GevGip1LRAaheLcUQCORRw

for example, the first result says, as of 2022, but i checked and the publishing date is 2021 from that source

also, during academic focus search, it makes up results from studies whenever it can only access the abstract like most studies are in that focus mode

i dont think it can note the publication date of sources and figure out that when a source says "today" it means day of publication or around there

what are some thigns i can prompt it so it wont conform to that too mcuh?

or note publication date better?

https://www.perplexity.ai/search/Prompt-Conduct-a-zvPh3sJeSGCaDXYVa_Epdg

here it just makes up things without providing sources

[glacial horizon]

Yes, one of the apps. I'm sorry. I'm using the Android app.

[glacial horizon]

And as a reminder, the question was how to reduce the font size of the answer to a question?

[velvet tangle]

You can only change the font size in your global phone settings

[small prairie]

I can probably edit the app to change the font and font size, but then I'd have to do it on every update.

[reef igloo]

People why that I need maximum of it, Only used apr.. 50 times this pro function, limit is 600, how to fix this bug?

@slim night

[small prairie]

???

They changed it so it only shows the number when you get to 100 left

Unless you know how to change it...

[reef igloo]

Well yesterday I used pro search 10 times. I am still pro status.

I guess it s a bug, not first time

[small prairie]

Why is it a bug?

You have 600 messages a day, so using 10 would not make it go below 100

[reef igloo]

What does mean CTRL + . so?

like on the screenshot

[small prairie]

It toggles the pro mode

It's just a shortcut for it

[reef igloo]

but usually it shows how much is left for example 567

but I have a .

[small prairie]

yep, to the right of that

[reef igloo]

this is the question

[small prairie]

Now it's changed, it won't appear until you only have 100 left

[reef igloo]

Oka

Understand

[small prairie]

Over 100 messages

[reef igloo]

But Honestly, This change affect all my threads

Where I can read about Update, because OPUS start working worse

They cut of limits

[small prairie]

What you saw in the past

Yep, I don't like the change either

[reef igloo]

new Q> why OPUS ignoring instruction, but SONNET not.

What happen with models

[small prairie]

Maybe they did more RLHF on Opus because it's more dangerous

[warm grotto]

For the raycast/pplx partnership, if I'm already subscribed to pplx and get raycast pro, would I still benefit from the 3 months or no?

Apologies if this has already been answered

[cyan fractal]

If I want Perplexity to extract the text from an image that I paste in to the thread, is there a way to skip the web/tool search activity? I expect the multi-modal LLM to extract the text itself and it does not need to do a search for external tools, documents, resources etc.

[fluid lark]

Just use writing mode

You have to click on the "focus" button and it will be there

[cyan fractal]

Ah OK, this makes sense. Thank you.

[alpine zealot]

Hello

Did anyone was able to get 6 month Perplexity Pro using Copy AI

I contacted Copy Ai Team

They gave me their own unique code

But i am now wondering where to use it in perplexity site

I contacted Perplexity Support, But they have not contacted me in 3 days

THanks For the Support By the Way

[velvet tangle]

Please be patient, the response time is several business days

[ionic sparrow]

#1227595427373776906 message

#💬│general message

[feral rivet]

If it’s the same as with the R1 promotion then the code has no expiration date and you can activate it a few months from now or a year from now.

[warm grotto]

Thank you!

[feral rivet]

No problem! Hopefully it’s exactly the same for raycast

[novel breach]

Jemand deutsches hier? Muss eine facharbeit für die Schule schreiben und bräuchte einen Vorgang wie dir ki mir behilflich sein kann 🙂

[vapid maple]

ehh, subscribe to pro?

[solemn nymph]

Hello, I am a subscriber to Pro and have been making API calls to Perplexity using your online pplx model. The query I send is asking for news stories from today and wants the URL source link. However, what is returned are the articles without the news URL/links to the source. I have tested the query/prompt in the Perplexity web interface, and it works perfectly well. Any ideas what I may be doing wrong?

[velvet tangle]

#🧪│api-general message

[summer hornet]

Thanks

[near nymph]

Hi, chat gpt has just released a new gpt 4 turbo version yesterday. Could we use it from perplexity? Thks.

[supple orchid]

Data protection / privacy question: Perplexity allows opt out of use of prompt and uploaded data for training, but does this carry through to downsteam models like ChatGPT? Does Perplexity pass through that preference? For direct users Chat GPT has a scorched earth option of deleting history to avoid training on prompts, or an opt out by making a privacy request via this link, but it is only for direct account holders https://privacy.openai.com/policies, what happens via the API? What's possible, and what's done by Perplexity?

[velvet tangle]

https://openai.com/enterprise-privacy

[obtuse star]

I am a pro, cantt figure out how to generate images. also would like to know if its possible to compare naswers side by side for different engines to become familiar on what kind of questions should be used on engines.

engines =models.

[hexed flicker]

Did you actually get up to date news articles? Did you provide context as to what sites to source from in your prompt?

[versed magnet]

you can try using the site: operator, it might work/help, though only if it is simple query. the API isn't the same as the web product, you can't really give it complex/multi-part queries or tasks and expect it to work particularly well

but for simple queries about recent/current events, it does work well and increasingly consistently

[unreal violet]

你好

[shrewd knoll]

Soon

Gemini ultra soon

[stark idol]

Can I have
Image prompt>Image
and not
Image Prompt>Search Results>Click Generate>Image
?

[small prairie]

Where's the fun in that?

[stark idol]

Who needs search results when you want to generate image 😦

[small prairie]

You can ask it how to generate an image...

[stark idol]

Yea, but I have to wait for search results to be dispplayed first, wait then click generate image

[small prairie]

The dev team are just lazy. Probably because they see it as more of a search engine, and the extra stuff is just a gimmick.

[stark idol]

okay

[spark pier]

How accurate would you say PPX is with looking up data on a table?

[small prairie]

Depends on the model

[spark pier]

I fed it a PDF with offset tabular data and it was quite accurate, compared to standard GPT models, I want to know how reliable it is

when I tried that a year ago with other apps, there was nothing but halulu, even when fed a very organized CSV or XLS

[small prairie]

If you pick GPT4Turbo, it should be slightly better than it since it's the API version.

[viscid sorrel]

Hello! I was wondering if there a group account options for Perplexity? is sometrhing like that offered for say schools or educational companies?

[frail tulip]

Hey guys I recently upgraded to Pro and set GPT 4 turbo as my LLM. How come when I asked it about its capabilities it returned something like " as an AI developed by Anthropic, I am not ChatGPT 4 Turbo, and I do not have the capability..."? Found this pretty funny.

[small prairie]

Lol, are you sure you're not on Opus?

[frail tulip]

Yeah I checked multiple times 🤣

[small prairie]

Could be the default system prompt then

[exotic warren]

I am getting very different answers when I ask a query on Perplexity Pro vs. when I do the same via the API. Is there a way to get the same answers on the web (pro) chat, that I will get through API? I will need to test it before I start using the API. Thanks

I am assuming there is no way to constraint the API based responses to certain original sources? I am asking the query to be limited to a website source but am still getting some hallucinating answers along with wrong source attribution (e.g. it says that the answers are from the source website but I cannot find those answers on that website)

[crimson crag]

Why do you choose a specific GPT model and for what purpose? Within a thread so you adjust accordingly and turn pro on and off?

[small prairie]

You can choose a specific site etc

You have to use the syntax like site:reddit.com

You can also use other stuff like:

[smoky rune]

Hi, I have a question. Is Perplexitity Pro's GPT 4-Turbo is the same as if I bought one from ChatGPT website?

[grand atlas]

Yes, though I would advise you to check out the technical faq. Since well I mean it would depend on your use case, if you’re looking for simple chat then sure yes, even document uploads work (32k token context window) but if your use case is code interpreter or addons then no. But then you do get the most advanced search engine so… trade offs.

[smoky rune]

ahhh okok, got it

[vapid maple]

no? chatgpt is not the same as gpt turbo api. unless you mean to purchase gpt4 turbo api on openai website

[smoky rune]

Yeah, that's what I thought. I tried programming problem, and original GPT 4 solved it, while the API one doesn't.

[untold jasper]

I m a student.Most of the things I need to focus on is Writing articles, do math and physics problems, look up concepts to complete tasks, etc. Should I buy chatgpt or this?

[shrewd knoll]

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

If you gonna buy here is a code for half price off

Chatgpt can't search websites, while Perplexity can search the website

The reason to choose Perplexity it is more accurate with the search website than Chatgpt without the search website based on my experience

[vapid maple]

chatgpt CAN search the web

[fluid lark]

perplexity is better tho, from what I've heard

[tacit abyss]

Hi, does perplexity have a sales team or a contact to discuss use for a company?

[vapid maple]

[e-mail] https://www.perplexity.ai/hub/faq/billing-and-subscription

[shrewd knoll]

Gemini ultra

Soon

[wispy raptor]

Hey guys do you notice that Opus started working so bad? Do you know why? Because lately I gave it a full book and asked to make an article of it. But now I give it and all I got is a short summary of the book. It seems like it wouldn't follow the rules that I give it

[shrewd knoll]

Coming in may

[shrewd knoll]

32k token only

[wispy raptor]

The books I give are short enough

[shrewd knoll]

Why didn't give Claude 3 opus?

[wispy raptor]

Oh sorry I meant opus of course

[shrewd knoll]

Try gpt-4 turbo

Claude 3 becomes strict when it comes to censorship, they're very sensitive about it

[shrewd knoll]

Gemini ultra

Soon

Coming soon

In may

Hopefully in perplexity

[shrewd knoll]

The amazing Gemini Ultra has the power that give creatively

[slim night]

Hey, @wispy raptor, we have a ticket for this. Please share your examples in this bug report: https://discord.com/channels/1047197230748151888/1228400472151294012

[fiery pendant]

Are there any other keywords, e.g. to only look at items in the last 3 days as an example

other search operators too maybe 🙂

[slim night]

Hey @fiery pendant, please use the 'after' operator, e.g. after:2024-04-14

[fiery pendant]

Thankyou !!!

[small prairie]

Yep, was gonna mention it.

You can also use these same syntax in your actual browser too.

[frank topaz]

Why is there a limit of only 4 files? Is there consideration of increasing it in the future?

[grand atlas]

Are these the search parameters that are usually used in search engines like google? Do they work in perplexity too?

[small prairie]

Yep, a lot of them work.

Not all of them though.

[versed magnet]

should look up 'Google Dorks' - there's a bunch of different operators and different ways they can be combined to make really targeted searches.. Can be fun to see what can be surfaced playing around
(also just tbc.. unlikely that any of the docs in the screenshot are actually 'classified'.. but just to convey the general idea...)

[modern abyss]

Hey,
SOmeone have a refferal link / promo code ? 🙂 Want to try Pro
Thanks

[shrewd knoll]

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

[small prairie]

Here's a referral link if you want:
https://perplexity.ai/pro?referral_code=T5MB8Z4J

Not sure if normal discount codes are currently working

[modern abyss]

Thanks guys, ah it just apply a $10 referral credit

I tought I would get one month or something like that like some other services

[small prairie]

The normal codes are currently disabled I believe.

Or else, you could probably have gotten more.

[modern abyss]

too many abuse I guess

[small prairie]

Yep, it's more that a partner flopped and send codes to anyone who made an account for them...

[shrewd knoll]

Just get refer code then

[alpine stream]

Hi Guys, after using Astrill VPN on my Android phone, I cannot connect to the server when I open the Perplexity app, but when I connect to the Perplexity website through Chrome, I can use Perplexity normally after verifying that I am not a robot. How should I deal with the issue of not being able to use Astrill VPN to connect on an Android phone?

[small prairie]

Try changing the location, other servers might not be flagged.

It's likely cloudflare that is flagging them.

[modern abyss]

What do you mean ?

[small prairie]

Lol, he is just saying to use his code...

[modern abyss]

Ah.

[alpine stream]

Yes, I tested the DNS of my current VPN, and the ISP shows Cloudflare.

why is that? the flagging issue?

[small prairie]

Most likely other users of the VPN have used that IP when scraping/ddosing a site that uses cloudflare protection.

[alpine stream]

Understood, so Perplexity's servers are deployed on Cloudflare? And my current VPN address might have been used for scraping or DDoS attacks on Perplexity's servers before, so Perplexity's servers have blocked the VPN IP address I am currently using.

[small prairie]

No, cloudflare acts as a proxy for loadbalancing. Their servers are probably on AWS.

[alpine stream]

OK, but this seems unable to explain why it can be accessed normally after verifying that I am not a robot through the Chrome page?

[small prairie]

Not perplexity, another website that also uses cloudflare likely got spammed too.

[alpine stream]

Currently, I only encounter this problem when using the Perplexity app.

[small prairie]

Because that checkmark is from cloudflare.

After checking the mark, you get a cloudflare session token which lets you through.

The app doesn't load the cloudflare human verification page though, so you can't get the token there.

[alpine stream]

I see, switching the VPN server seems to be the only option.

Thanks

[shrewd knoll]

Refer code will give you this discount https://perplexity.ai/pro?referral_code=TQRFFK3J

[patent rune]

Is it possible to quickly switch between "ALL" and "Writing" focus using keyboard shortcuts?

[crude perch]

How to active function calling using pplx api?

[vapid maple]

you mean this: https://docs.perplexity.ai/reference/post_chat_completions ?

[crude perch]

no something like this https://platform.openai.com/docs/guides/function-calling

[fleet fern]

Hi,Currently, I am developing a chatbot using the RAG methodology and GPT-4. I'm looking to replace the Language Model with one that measures perplexity. Are there any APIs available similar to GPT?

[subtle thunder]

Hey guys, is it possible to add excel table somehow for analysis?

[versed magnet]

there is no function calling capability with the API afaik

i think i saw something about JSON outputs on the roadmap.. so ig function calling would be after that if at all

[vapid maple]

is there any reasons why the generated images is limited to personal use? is this due to

2. Additional Commercial Terms.
   If, at any time, (a) image generation or image editing is a core business or product of Licensee’s and (b) the total monthly unique users (MUU) of the products or services made available by or for Licensee, or Licensee’s affiliates, for such products or services is greater than 1 million MUUs in the preceding calendar month, then immediately thereafter you must request a license from Playground. Playground may grant this license to you in its sole discretion and you are not authorized to exercise any of the rights under this Agreement unless or until Playground otherwise expressly grants you such rights as a Licensee.

[olive topaz]

Hi 👋,

My name is Mark Salden, graphic designer and social media marketer from Belgium. I have some questions about Perplexity.ai and hope someone can help me.

Is it true that I can ask up to 300 questions/prompts per day? (At least that is more than Chatgpt.)

Example: What if Chatgpt launches model gpt 5 tomorrow. How fast can Perplexity integrate this model?

So i can generate images with Dalle 3, SDXL, etc. Is there a limit per day on this too? What about dimensions, quality (px), negative prompts...

Does Perplexity have access to the web? Can I ask questions that Perplexity looks for on the web? Can I also give url links as a prompt? Can it scrape websites?

Can i also create chatbots from files (pdf, cxv..) i upload? Also, can Perplexity filter things away and re-download the file?

I read in comparisons with Chatgpt that Perplexity is 'less creative' in terms of content creation. Is this true?

A lot of questions, i know but i hope you guys can help me out. Thanks!!

[vapid maple]

1: currently: 600 questions with pro enabled per 24hours interval. unlimited questions with default sonar model & without pro toggled.
2: 🤷‍♂️ idk, not the fastest but will eventually shows up
3. 50 per 24hours interval. personal use only. 1024x1024
4. yes, esp. with pro mode toggled on. site: would work like 95% of the time (only in user query/question not in ai profile). but no live scraping
5. the uploaded file is kept upto 32k tokens with RAG for web/mobile users, you can not redownload the file. no native file uploading support for api users.
6.

[olive topaz]

Thanks for the fast reply. I appreciate it!

[hollow ginkgo]

Hey @olive topaz!

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

[vapid maple]

bad bot

[feral rivet]

oh ,hey you are here now

[covert vessel]

Is there any login issues with Perplexity currently? My app hangs and never logs me in

[slim night]

Hey @covert vessel! Which app/device version are you using? Which login option? Are you facing any issues signing in on the web version?

[covert vessel]

Thanks for reaching out! On iPhone 15 Pro, iOS 17.4.1. Logging in with my own email via link. Can’t validate web cuz I’m not home atm..

[slim night]

Are you also using a VPN service? Please try to log in without it, if you do.

[covert vessel]

Oh, I do! Trying rn

I’m in! Thank you so much

[lofty sphinx]

I see a setting to generate images but when I ask for it to generate one it says it can’t do it.

Playground v2.5 is selected

[ionic sparrow]

https://discord.com/channels/1047197230748151888/1194794305362071552
#⚡│ask-community message
Create image using precise wording provided in the prompt

[lofty sphinx]

TY. I’m assuming this only works on web for now. Don’t see option for iOS

[ionic sparrow]

correct, you can also print them from any browser on your iOS

[lofty sphinx]

Ty

[chrome magnet]

Hi is there a way to access through api to obtain references/summary in my response?
I've noticed eg sonar-medium-online returns a different response than the browser app

[woeful saffron]

For citations through the api? Applying is discussed here -

#🧪│api-general message

[shrewd knoll]

What happened? To opus? Now they have usage limit but how many?

Why limit now???

[small prairie]

What do you mean?

[shrewd knoll]

No more unlimited opus

They are going back but is 5 the daily usage now? And is it still 32 tokens or is it 200k tokens now?

@velvet tangle

[small prairie]

Looks like they added an opus limit.

It's more than 5 though

Not sure of the actual number yet

Mine says opus limit 10. Not sure if it's counting the times I've already used it today.

[shrewd knoll]

Is buying Gemini Ultra that expensive until they decide to limit Opus?

[small prairie]

No idea. They have next to no transparency with the pro users.

[shrewd knoll]

I know they will add Gemini ultra soon but didn't think they would sacrifice to limited opus

[small prairie]

I would rather have Opus over Gemini, any day.

[shrewd knoll]

Gemini ultra is for creative option

It's still important

[small prairie]

Yep, but it's terrible at everything else.

I have all the major models, and Opus has the best overall performance.

Also why would they have to limit opus for that?

[shrewd knoll]

It's for inspiration and creative choices like creating a novel

[shrewd knoll]

Opus is for information choice not for creative choice

[small prairie]

it's for both. You simply need to update the system message.

[shrewd knoll]

Because they want to include every model otherwise is pointless

[small prairie]

They already added Gemini in the past and took it down because of bugs.

[shrewd knoll]

Opus too many censorship unlike Gemini ultra

[small prairie]

...

It's pretty much the other way around.

I'm guessing you haven't tried Gemini ultra then.

[shrewd knoll]

Now they recently release Gemini pro 1.5 api

[small prairie]

1.5 pro

There's no ultra yet.

[shrewd knoll]

I did for writing fanfic

[shrewd knoll]

They are saving money for next month's release otherwise why would they limit opus?

[small prairie]

Saving money?

Why would they need to do that?

They are using the API, and Gemini is likely cheaper than opus.

[shrewd knoll]

No, I try Gemini Ultra not as strict as Opus when comes to censorship

[shrewd knoll]

Gemini ultra unlimited token

[small prairie]

It is a lot stricter from my usage.

[small prairie]

Unlimited? That is definitely false.

They are probably gonna cap it to 32K tokens.

Like the other models.

[shrewd knoll]

It's not, I try writing fanfic and they can write much gores than opus

[shrewd knoll]

We don't know yet because when ask @velvet tangle @velvet tangle replied depends the price

[shrewd knoll]

Opus is prune when it comes to gores

We don't know whether they will upload video option now for Gemini ultra

I try to ask but again they said depend

[cyan arrow]

If we could increase the OPUS limit to 60 times a month, I think everyone would be more than happy to accept the limit instead of the limit of 30 times a day

[shrewd knoll]

Opus is too expensive the problem is

[small prairie]

I think we've already mentioned this multiple times... it's only 2.5x more than GPT4Turbo for output

which would make it around 240/day

[shrewd knoll]

The problem is that they haven't updated the gpt-4 turbo to gpt-4turbo v yet

[woeful saffron]

This is my experience too. I often hit censorship for seemingly benign prompts.

[shrewd knoll]

Which model are you referring to?

[small prairie]

Yep, with claude I can atleast persuade it.

[shrewd knoll]

Too many censorship the problem

[woeful saffron]

They were discussing Gemini Ultra, one member was discussing how it was more lenient than Opus, Code was expressing experiencing the opposite, I was echoing that sentiment.

[shrewd knoll]

That user don't use for creative reason and dare to said opus is better than Gemini ultra for creative reason

Opus is best for information reasons and Gemini Ultra is best for creative reasons

We need both for everything like crafting a novel or a video

The censorship is strict as f when asking opus question

Gemini ultra is strict but not that strict, they can even write gore unlike opus

[woeful saffron]

I work on a lot of projects that are centered around substance use harm reduction and education, and I’ve tried to use Gemini Ultra in the past to brainstorm some ideas for outlines of an infographic - but quickly get shutdown due to the content pertaining to substances, even if the focus is harm reduction. Fortunately I don’t have this problem with some other models. But everybody’s use case is different, so we all experience different barriers.

[shrewd knoll]

They are both strict but opus more strict when asking about the shipping they said they are uncomfortable discussing it and we have to go to Gemini Ultra to discuss

The problem is that Gemini ultra don't have any search source so we can't ask it

And only can wait for the release then

[hollow kestrel]

So I am looking to run a specific query with a sizable list of websites for a research project. How can I best automate my way through those ~400 requests. (Even better if you have a way to have said answers archived into a parsed database. I’d settle for google drive or excel online 😬)

[dull fjord]

Did a 30-per-day limit get applied to Opus recently? Any idea if it will be unlimited again?

[small prairie]

Elons son, what are you doing here?

[dull fjord]

Thanks.

[modest echo]

I've written a lot of stuff over the years, does anyone have a recommended workflow for inputting my material and getting the AI to copy my style? and if so, which pro model is the better for it? gemini? claude? gpt4?

[peak lily]

Guys? Who also stopped pay OpenAI ChatGPT and only uses Perplexity?

[shrewd knoll]

Gemini for creative, Claude for information and gpt4 in case of Claude's censorship

[shrewd knoll]

I didn't pay Openai but I stopped using it and bought Perplexity pro

[sonic dew]

Newbie here, to use perplexity API : The doc is written in python can we use it with TS/Javascript?

[velvet tangle]

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

[sonic dew]

thamks IceLavaMan !!

Where is the doc to install the perplexity API, I follow a video where it showing part of the installation. But I can't find any information about it in the doc

Screenshot from the video

[velvet tangle]

npm install api, it's in the code block

[sonic dew]

No this is from the video

the code block is only showing:

const options = {
method: 'POST',
headers: {accept: 'application/json', 'content-type': 'application/json'},
body: JSON.stringify({
model: 'mistral-7b-instruct',
messages: [
{role: 'system', content: 'Be precise and concise.'},
{role: 'user', content: 'How many stars are there in our galaxy?'}
]
})
};

fetch('https://api.perplexity.ai/chat/completions', options)
.then(response => response.json())
.then(response => console.log(response))
.catch(err => console.error(err));

Why I can't find this information in the doc?

[velvet tangle]

You can find this in the docs, you just need to choose your language

[sonic dew]

I saw that we can change language but for JavaScript there is no installation cmd specified

[velvet tangle]

You can just fetch?

[fluid lark]

what font does pplx use to display codeblocks

[sonic dew]

then when/where do I use the API key ?
Hum.. Usually I find all the informations needed in the doc

[haughty brook]

when you fill in the bearer input at the top it will show you where you should put it

[pulsar briar]

Should we expect some response via email regarding filling out the form for beta access to citations or should it just "appear" in results?

[peak lily]

Guys ChatGPT 4 turbo over Perplexity is 32K or 128K?

[fleet token]

is there any month trial?

[brazen quarry]

You get a free week to try it if you're new

[fleet token]

no month trial?

[brazen quarry]

Nah

You could use a referral code and get it for $10 for a month

Instead of $20

[vagrant glacier]

Hi there. Just subscribed to Pro. I was under the impression that perplexity can generate images, or am I mistaken?

[small prairie]

Search in #⭐│starred for image

[slim night]

Yes, you can generate images: https://discord.com/channels/1047197230748151888/1194794305362071552

[slim night]

Yes, there should be an email regarding the citations access.

[pulsar briar]

How long should it take from when form was submitted?

[pulsar briar]

would there be a confirmation email of the form submitted?

[vagrant glacier]

Thanks. I managed to figure it out on a web interface, but not in the app yet.

[winged moth]

Is it true, that you cannot generate images and use it for commercial projects? OpenAI allows it for DALLE3, PPLX not?

[vapid maple]

personal use only.

[velvet tangle]

The response time is 1-2 weeks, maybe also 3.

[amber berry]

is there any promo code for one month

[slim night]

No, but only $10 for your first month with a ref code Pro users can share.

[opal quartz]

is illama 3 going to be added as a model in a perplexity to use?

[velvet tangle]

On labs and API for now #🦙│feedback-playground message

[shrewd knoll]

The refer code is both for old and new user

[prime moon]

does perplexity offer a trial for the pro features?

[cyan arrow]

My opus is not updated, I used it 3 times yesterday, and now it is only 27 times??? doesn't it say that it will be updated every day? Or do I only use it 30 times a month?

[small prairie]

It is refreshed in 24 hours after you use it.

For each use.

[cyan arrow]

Thank you for your answer, which solved my doubts

[small prairie]

Np

[peak lily]

For me, it's still not refreshed... it's already more than 48h

[turbid gorge]

me too

[ruby dawn]

me too

[alpine stream]

I got refresh but I have 3 times remain which is ...

[peak lily]

lol

[alpine stream]

yeah which is lol..

[jagged dune]

Anyone can't use the app without VPN?

[olive topaz]

Hi @slim night Could you - when you can - answer my question I sent you privately on X? Thanks!

[fluid lark]

that's normal

it's to prevent abuse

just try another server

oh nvm, you said without vpn

[jagged dune]

Yeah, i have to connect to vpn to use the app

[fluid lark]

it's possible that your IP got falsely flagged or something

[jagged dune]

Sure, i contacted support but no reply yet

[fluid lark]

are you using mobile data or wifi ?

[jagged dune]

Im on wifi

[fluid lark]

sometimes turning off your router then on again can change your IP (or so I've heard)

if you have dynamic ip

[jagged dune]

Using the same wifi for the laptop but it can connect

[fluid lark]

oh.. that's odd can you check if the IP on your laptop and phone is the same ?

just search "what is my ip" on ddg or google

[jagged dune]

It’s different

[jagged dune]

try to catch the http request and got a 403 forbidden

[fluid lark]

well that explains why it works on your laptop but not on the app

[jagged dune]

I contacted support but they haven't repiled yet

[cinder coral]

When Llama 3 70B 🥴

[versed magnet]

if you just want to use it (without internet connectivity) https://labs.perplexity.ai/
but also, there was a tweet say they would be integrating / offering it is an online model down the track, which will be cool

[tall cosmos]

is there a way to prompt directly for images?

[fluid lark]

???

[vapid maple]

--font-family-mono: "SF Mono", SFMono-Regular, ui-monospace, "DejaVu Sans Mono", Menlo, Consolas, monospace;

[small prairie]

On on Groq for 350/Ts

[versed magnet]

40 tokens/sec is already way faster than i can read ha so it's fine ig

but yeah wow that's fast af

haiku is 70/sec, and dbrx 80/s on labs (for the same query)

llama3 seems on the slower side comparatively

[feral rivet]

they have it now! thats great

[lime sphinx]

why is it that when i run my discord bot code i get a 401 error in return when trying to run the script?

is the api down?

[small prairie]

401 means you aren't authorized

Maybe check your API token, or that you have credits in your account.

[lime sphinx]

okay so how do i get that?

[small prairie]

is it your bot?

[lime sphinx]

YES

[small prairie]

If so, there is likely a .env file or something where the key is.

[lime sphinx]

yes

i do have credit and i set max token to be 10.k? should be enough?

[small prairie]

You need to replace the key with one in your api dashboard

[lime sphinx]

i have done that

[small prairie]

Have you checked the http requests to see that it has the bearer token header added?

[lime sphinx]

it is added response = requests.post('https://api.perplexity.ai/chat/completions', headers=headers, data=json.dumps(payload))

[small prairie]

Can you log the request header and manually check

just to see that the token is there

[lime sphinx]

headers = {
'Authorization': 'pplxtoken',
"accept": "application/json",
"content-type": "application/json"
}

# Construct the API request payload
payload = {
"model": "llama-3-8b-instruct",
"messages": [
    {
        "content": "string",
        "role": "system",
        "system message" : "turn the positive prompt into a negative prompt word by word seperated by comma in 10 sentences"
    }
],
"max_tokens": 10000,
"temperature": 0.2,
"top_p": 0.9,
"top_k": 0,
"stream": True,
"presence_penalty": 0,
"frequency_penalty": 1

} this is my request body though

[small prairie]

I would hide the token...

I think I know the problem anyway

you haven't added bearer in front of the token.

[lime sphinx]

ohh oki

thx ill look into it

[small prairie]

basically the header should look like:

{
 "Authorization": "Bearer <api-token>"
}

[lime sphinx]

im now getting 400 error

[small prairie]

Okay, better than 401 error

It basically means bad request.

I'll check the url you are passing in

[lime sphinx]

oki thank you

[small prairie]

Firstly, I would remove all the none required parts of the payload. So you can minimize parts that can be causing the problem.

The API says you only need the model and messages.

I think it could be failing because of your messages.

That last line is pretty important.

Really it should be bolded, or something.

[small prairie]

Do you get what you need to do?

[versed magnet]

"max_tokens": 10000 This vlaue is too high / invalid. Needs to be < 8,192 (in practice, it should be a lot less than that, like say ~5,000, otherwise there will be no room left for the tokens in your actual query)

"presence_penalty": 0,
"frequency_penalty": 1

I believe this also invalid. If you are going to include these (optional) parameters, it can only be one or the other, but not both

[versed magnet]

yeah think this is right. it is sensitive to order in which the roles are structured in the request

[small prairie]

Yep, you need a user query last

Since that is what it's supposed to answer

[versed magnet]

@lime sphinx https://docs.perplexity.ai/reference/post_chat_completions it might be worth starting here and testing it, using only the required the parameters (can then add any optional ones later). Add your API key to the field in the top right.

[small prairie]

The stuff there is pretty buggy though, outside of the shell option.

Literally switching to go just return a code example of:

package main

...

[versed magnet]

tbh i've never had an issue with it - but ig i've never done anything more than use it just to play around with different parameters, then maybe copy-paste the curl request into the terminal ha

[terse fern]

How can I find one of my Smartphone App Threads on my Desktop Browser and vice versa?

[lime sphinx]

Well i solved my bot code and it is now working properly thank you

[amber berry]

free trail in not available

[modest echo]

what's the best mode ot get coding assistance, writing? academic? I dont want interent searches, just direct coding feedbackin.

and is Opus better than GPT4 for coding? I hear gpt is kinda lazy and vague

[umbral forge]

I wanted to generate sprite, texture for my game - i have purchased perplexity pro and using DALL-E 3, I am able to generate some image but it is not precise and i am not able to improve image with previous image, any suggestion on how can i improve my prompt?

[olive crypt]

Hey everybody, relatively new to Perplexity and currently trying to use Perplexity, Make and Notion to automate account research and assist in prospecting messaging for my sales role. I've generated prompts that work quite well when in perplexity; however things get quite messy and unhelpful when using Make to automate said prompts... That being said, do In need to use one of the perplexity models sonar online or pplx-7b-online to get accurate information, or should it still be able to generate accurate and up to date insights using the open source models? Have seen folks do videos using Mistral as the model for account research, but that hasn't led to the accuracy I need. Any help provided would be amazing!!!

[velvet tangle]

https://discord.com/channels/1047197230748151888/1223947903941349476

[balmy bough]

Is there any way I can edit my collection on my iPhone similar to how I can on my android device or my windows laptop?

Or are iphone users simply not allowed to do that? I brought it to developer attention at least 5 or 6 iPhone updates ago, but so far nothing has changed on the app at all. It almost feels like they’re doing it on purpose since nobody else cares and it only affects me :/ I wonder if they think it’s because I’ll pay or something even though it seems like a pretty simple yet important thing to address :p I mean, a collection is supposed to be able to hold multiple threads which hold multiple prompts. If you want to edit that collection, you must simply make a new one and try to remember what the other collection had in its prompt. Very frustrating!!

Basically makes the iOS app completely unusable for me :pp

The ability to talk to an ai on a mobile app is NOT very impressive, the features that make this particular AI unique are absolutely necessary

[mortal sluice]

If I get two referrals, does that mean I get the next month free? Or just 10 dollars off for the next two months?

[fluid lark]

I'd assume you can't stack them like that, I mean that's how it is everywhere else

[slim night]

Hey, @mortal sluice, it's the latter

[mortal sluice]

Appreciate it 🫡

[peak lily]

Does anybody know upcoming plans for Perplexity.ai? I started uses that lot of and also recommend to my friends instead of ChatGPT from OpenAI.

I even started use that as replacement for Google search. Does more search results are planned? For example more than 20?

[timid prism]

Are Pro searches better when you are a subscriber? Or is the main difference only the number of Pro searches you can make?

[small prairie]

Doubt more than 20, since they take up more and more of the context

[peak lily]

Well, I'm ready to pay more 😅

[velvet tangle]

More sources do not mean a better result. pro search uses more sources than a normal search, but even more will not improve the final result.

[peak lily]

For most searches, yes, but if you try to search for something written by a news or political source, there are a ton of sources, and I sometimes find the pro search stuck on 20 social media posts instead of using even more sources and then finding an official article on a government website.

That's why I looking for analysis even more sources

[small prairie]

You can specify where you want the sources from

[peak lily]

Yes, that requires knowing where the information is, but what if I don't know where the inflammation is...?

[small prairie]

If you don't know where the inflammation is, you should see a doctor

Also it's using embeddings to match the search, so if you specify something like from credible sources it should filter the result somewhat.

[peak lily]

Thank you for your very constructive answer.

[ocean saddle]

opus 3 uses $ at inline math .
so, say "use $$ instead of $ inline math"

maybe OK. so let's throw out the math practices to LLM.

[teal dock]

I have read a lot about perplexity copilot. How can I activte the feature. I see no copilot button on the web portal. Thanks!

[grand atlas]

Renamed to “pro” toggle that

[teal dock]

💡That was my concern, that it is an older feature which was turned into something else.

[fluid lark]

can gpt4 from pplx run code to do math calc or is that exclusive to chatgpt

[small prairie]

Are you talking about Wulfram alpha?

There is a focus mode for it

[fluid lark]

nope this

[small prairie]

That is code interpreter

And perplexity doesn't have one

[fluid lark]

👍 thanks