⚡│ask-community

[fluid lark]

yea i knew it was called something, i couldnt remember it

[modest echo]

dont suppose anyone had any feedback on my quiestion? what's the best focus mode for coding assistance?

[small prairie]

Depends on your language

[versed magnet]

really? I find that surprising.. What language for which focus..? Surely not Academic for python, Reddit for JS etc..? lol

no idea but fwiw I'd have thought Writing. It makes the model offline; unless working on some problem for which internet search results would actually be needed for the LLM to provide accurate / useful code, I don't see much use in having search results infused with every query, for pure coding tasks

[small prairie]

Some languages do better with search and some do worse. And it mainly depends on the language.

If the language is rare, your best bet is using search.

[versed magnet]

so in short: Writing

unless using some exotic / unspecified language

[small prairie]

If you are using new libraries, probably best to use site:the-docs-site to make sure the answers are up to date.

And other stuff like that.

But I would have search in one tab and writing in another.

[humble quail]

I've just connected perplexity (pro) to make.com I wrote a prompt to pull in research, but perplexity keeps giving me super old answers, I've tried using instruct and 2 of the online models. Not sure where to go from here.

I tested my prompt in chatgpt. and that worked ok. I thought perplexity was better for search?

[versed magnet]

can you give an example of the kind of prompt you're using (or describe what you're searching for)?
And you're using the API right?

[humble quail]

I copied the api and it's giving me answers that are super broad. it was about getting the latest in AI tools and applications.

In perplexity it wasn't even giving me an answer and it was saying "I'm sorry, but I do not have access to real-time data or the ability to search for the most recent Artificial Intelligence..."

[versed magnet]

tbf it's a pretty broad question...

[modest echo]

HTTP/JS/CSS?

[humble quail]

in Make.com it was giving me a super broad answer explaining what AI is and the introduction ChatGPT which is super old

[versed magnet]

you need to be using the sonar-medium-online (or small) model for the LLM to be able to use web sources

[humble quail]

The prompt itself is quite long, and specific, GPT gave me some answers, but I was told that Perplexity was better with searches, so I signed up to the pro account

[versed magnet]

not sure if something along these lines what you're after in terms of the repsonse - but it works pretty well, so long as the query is quite basic

[versed magnet]

"The prompt itself is quite long" yeah that doesn't work particularly well with the online models

[humble quail]

Can my prompt translate from GPT to Perplexity or maybe I need to rewrite it?

[versed magnet]

Using perplexity web product, you can give long prompts and it should work (like with GPT). But for the API (i.e. sonar-medium-online), it will generally struggle to provide accurate / useful responses if the prompt is extensive / complex

[humble quail]

Thanks for that. That's good to know.

[versed magnet]

np! best of luck :))

[humble quail]

Any idea why the web version would just give me a "I'm sorry, but I do not have access to real-time data or the ability to search for the most recent Artificial Intelligence..."

i tried switching between the different models available including gpt4 turbo

[versed magnet]

unless you are in Writing mode, no - that would seem odd indeed

[small prairie]

JS changes all the time...

Nowadays people use tailwind for css.

[humble quail]

would that be by toggling the "pro" to switch it on?

[modest echo]

never heard of tailwind

[small prairie]

Tailwindcss?

https://tailwindcss.com/

[modest echo]

I was using this to learn HTML/CSS from scratch more than anything else, if tailwind is like the old macromedia fireworks for CSS, that kinda defeats the purpose of self elarning though, that's all

[versed magnet]

No by 'Writing' mode, I was referring to the option you can see in the screenshot after clicking Focus. Having Pro on/off changes how many sources will be retrieved, but does not make it offline (like Writing mode does). So yeah, it is weird that you're getting "sorry, I don't have access to the internet.." messages, regardless of whether Pro is On/Off

[humble quail]

Yah that's switched on... super odd... will sleep on it. Maybe it's a glitch in the matrix...

Thanks for your speedy help @versed magnet

[hollow ginkgo]

Hey @humble quail!

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.

[small prairie]

Using tailwindcss has improved my vanilla css skills too.

[wraith stream]

Is there a possibility to share / copy my library from one account to another? Use one for school and one privat and therefore some contents would suit better to both.

[light hinge]

is it possible to interact with the models without web search, only responding to my provided context?

also, can I easily switch between models?

[ionic sparrow]

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

[light hinge]

thanks didnt pay close enuf attention there

[ionic sparrow]

only through Settings page

[sleek sky]

When using pro search, is the information in the search results retained for subsequent responses? If yes, how do I ask it to keep the source citations?

[hazy marsh]

Can I apply my referrals discount from iOS App Store subscription?

[sand token]

If a thread is started with Opus but switches to Sonnet midway, is there a way to resume and continue with Opus?

I automatically responded with Sonnet because I ran out of Opus usage.
I waited for my Opus usage to be replenished and then resumed the same thread. However, it did not switch back from Sonnet to Opus.
My settings are set to "Use Claude3 Opus".I have confirmed that new threads start with Opus.

[slim night]

Currently, it's removed, we'll see if that'll be offered later.

[timid prism]

Is there TTS functionality as of now in Perplexity? If not, are there plans for it to be added at some point?

[tall smelt]

There is for iOS devices

[shrewd knoll]

No, it must be website only

[small prairie]

If you use collections, you can add both accounts to them.

[copper comet]

Can someone tell me what this is actually about as far as with Perplexity what are we doing here? I thought we speak to Perplexity and give information to each other.

[raven sparrow]

got a few questions

1. what's the limit on gpt 4 and claude opus and claude sonnet
2. is the pro version available in the uk since the pricing is in dollars only

[ionic sparrow]

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

Opus limited atm 50/day, UK 👍

[summer shore]

On iOS is there a known VPN all that will allow Perplexity to work? I have tried 1.1.1.1 and now ExpressVPN and both are no go

[ionic sparrow]

Proton

[small prairie]

The best VPN services are probably mullvad(Sweden) and proton (Switzerland).

[summer shore]

Any additional tips my friend ? Still not working and not seeing a way to split

[ionic sparrow]

iOS?

[small prairie]

You might also want to try a different server location.

Sometimes it's just certain locations which are flagged by cloudflare.

[summer shore]

Actually it’s working!! Thank you

[summer shore]

Thank you 🙏

[small prairie]

Np

[muted thorn]

Can I get Perplixity, or Claude Opus and Claude Sonnet, to respond with only one source?

[slim night]

Please try the site: operator

[muted thorn]

can you give an example

please

[small prairie]

site:reddit.com

Will use only sources from reddit.

[muted thorn]

And when I ask a question, it uses only Reddit sources?

[small prairie]

Yep

What's the best monitor? site:reddit.com

[muted thorn]

What does it look like if I ask a question? Does formatting matter?

[small prairie]

Is probably how you would write your prompt.

[slim night]

#⚡│ask-community message

[muted thorn]

Ohh thanks

[dense atlas]

Does the perplexity iOS app have a deep link api?

[muted thorn]

For larger questions, it does not use just one source

[small prairie]

It uses one site, not one source

Don't think you can make it use only one.

[frozen slate]

Can Perplexity generate images with a 16:9 aspect ratio?

[still python]

Perplexity vs. Perplexity Labs

[valid bloom]

Hi Everyone, who can point me how to get pro channel access. I am a pro sub user but I cannot access that channel. Thanks

[velvet tangle]

#💬│general message

[valid bloom]

is it the discord icon on the bottom left?

This is my 4th time leaving and joining the server. It's not working for some odd reason.

Thanks for the help.

[pastel breach]

What? sorry i don’t get your question

[valid bloom]

I am pro by the way.

I don't have access to the pro channel in discord

[pastel breach]

Oh okay, did not know thats there is a pro channel here

[valid bloom]

I was given this link to leave the server and register again

[pastel breach]

Oh i see

Probably a bug from perplexity

[valid bloom]

Apparently it works for other people when they leave and come back. but not for me. this is my 4th time leaving and coming. 🙂

[pastel breach]

Propably leave, wait some time and then join again

[valid bloom]

that is a nice idea. Let me try that. Thanks.

[north hornet]

送同学礼物

[deep crater]

I want to know the limits of Claude Opus and how does that work

If that daily limits?

If daily, how do they renew ./.. I mean at midnight OR keep renew limits every 3 hrs ?

Thanks

[vapid maple]

@pastel breach you are missing the secret sauce: (re)join via the link found in here https://www.perplexity.ai/settings/account

[muted thorn]

And how can I get Perplexity to answer questions I ask it when I have uploaded a file. It should name the sources of the file

[valid bloom]

Thank you Sire. I need glasses. I was clicking on the wrong icon.

[thorny frigate]

I have a question, it's very annoying. As soon as I copy texts, the links like these (1), (2), are always copied as well. Can't I switch them off?

Example when i copy the text and paste

[vapid maple]

there is a copy button on the end of the response

to remove the reference using regex matching: #1220415030328164352 message

((\[[0-9]+\])+)|(\[[0-9]+\] http[s]?://.*)|(Citations:\n)

[thorny frigate]

Thank you!

Oh @vapid maple that is bad 😄 Copy when i click copy than copy that with Links

[ancient fjord]

Hi, is there another way to have an access to the pro channels? I left discord and joined via link on the webpage but still I don't have an access to them.

[velvet tangle]

#⚡│ask-community message

[vapid maple]

postprocess with regex above.

[smoky rune]

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

?

[true chasm]

Hi do you know what is the difference in chat gpt 4 and chat gpt turbo ?

[vapid maple]

yes, if you can manage to copy/send the login link you received in email to your device, and click on it. sharing with others is against TOS

[vapid maple]

why does the automod prevent me from adding text description in #💗│sharing channel?

[iron wraith]

a que edad se empieza a cotizar?

[slim night]

https://www.perplexity.ai/search/Output-the-plain-ocpEr22uTRy1_oF56MgMxA
You can ask the model to output the answer in a markdwon code block with no citations, for example, and each code block has a copy button that you can use.

[slim night]

It was set to share links only, otherwise, it'll be full of random text. But that's a fair point, I think it won't hurt if something goes along with the shared thread.

[slim night]

https://pplx.ai

[pastel breach]

Oh thanks

[smoky rune]

that was literally my question

i think it's different, the GPT 4 (in ChatGPT website) is more superior than the GPT 4 Turbo API

i tested both

seems bing chat and perplexity GPT 4 Turbo is the same

produces same result

[dull fjord]

Is Claude 3 Opus still limited?

[slim night]

Currently, it's 50 per day. #👑│pro-lounge message

[smoky rune]

just buy gpt 4 xD

or just dont use AI

https://tenor.com/view/gigachad-chad-gif-20773266

[red bane]

Do you have any plans to make it unlimited again?

[fleet fossil]

If opus isn't unlimited anymore, why is the counter not there?

[slate ice]

It shows up when you have 10 or less. But it's probably not to alarm people and probably to keep the actual limit out of sight generally. It's definitely not being transparent though. Any news regarding this I'm having to keep finding out by reading the messages on here, not even in the announcements lol

[fluid lark]

~~Hello 🙂 Is there any way to have perplexity search with pro when it's set as my default search engine? I saw there is the s query parameter: https://www.perplexity.ai/search?s=o&q= maybe something to do with that? ~~ nvm found the answer here from yesterday! 😄 : #💬│general message

[rancid narwhal]

Hello, does anyone know why the mobile app won't log in with a Google account? I tried many times, but it doesnt work🤔

[haughty brook]

Please try again without using a vpn. If you aren’t using a vpn please open a thread in #1140622008086970420 and state your phone/os version/ app version.

[rancid narwhal]

got it, i will try again, thanks

[spring glen]

Does anyone know why perplexity collection is not working, like it completely ignores the prompt.

[wicked nacelle]

Hello, I have a question about Perplexity Enterprise Pro. What is the difference between it and the classical Pro version? Is it only about more data security? I already have a classical enterprise status for my subscription, and all enterprise customers are automatically excluded from AI Data Retention.

I don't understand the point of launching this new 40 $ for the companies, because I don't put our proprietary data on your server because my company already had a fine-tuned Mistral 7b that only runs locally for such reasons. It seems to me it is just about fulfilling some US regulations that do not apply to small companies from Europe

[versed magnet]

what distinguishes the parts in red and yellow from how the data of non-enterprise users who have opted out of data retention/training is handled?

[vapid maple]

30 days data retention for normal users... immediate deletion upon requests

[sacred pulsar]

Quick question, what's the criteria for Collections -> AI Prompt limit? Is it character count or token count?

[manic plinth]

hey , i'm getting problem on purchasing preplexity ai

it is not redirecting to payment section

can you give me solution

[velvet tangle]

Please make sure to disable any VPN or content blockers. Then just visit https://www.perplexity.ai/pro -> Get Started

[manic plinth]

i had disabled

still getting same problem

[velvet tangle]

What browser are you using?

[manic plinth]

edge

[velvet tangle]

Please try brave or chrome

[manic plinth]

ok alright mate

[granite forge]

I have a trivial question, when I upload a file and ask something about this file with a prompt. How these inputted in a model, the prompt is appended after the file or before the file.

Because it is quite different these two ways in term of performance if I uploaded a big file.

[versed magnet]

So if I opt opt in Settings, none of my data is used for 'training'. But my 'queries' are still retained for 30 days, vs 7 days for enterprise users - is that the distinction?

I don't really get why queries aren't considered your data (I get, though, that it won't be used for 'training' purposes - ig more to just understand what people are searching for)

[vapid maple]

to my understanding: yes. besides the location where the data stored for enterprise users is soc2 audited

[versed magnet]

but be that as it may.. kinda funny that enterprises users queries are still retained for 7 days.. (I mean 0 days would be the ultimate privacy, if that's meant to be pitch..)

[kind nimbus]

Hello, in pro plan what is the limit to use in claude?

[vapid maple]

currently 50/24h for opus 'temporarily', and 600/24h for the rest

[kind nimbus]

What exactly meaning temporarily?

[vapid maple]

#💬│general message

and the limit is now raised to 50 from 30

[red swan]

Hey all interested in subscribing to perplexity pro my question is what is the context window for perplexity pro models such as claude 3 opus?

[small prairie]

32K I believe

[red swan]

So does the search hinder this at all? My goal is to use an llm service with pretty good search that allows has respectable context window for programming.

[small prairie]

Perplexity is pretty good for that. There is also phind which is designed more for programming. Choose whichever you prefer.

[red swan]

If you have tried Phind how would you compare them?

[small prairie]

You can as @feral rivet

[red swan]

?

[small prairie]

I am making my own AI for writing code, so you can ask DailyFocus, since he has both perplexity and phind.

[red swan]

Oh okay.

[small prairie]

Also what language are you planning to use?

[red swan]

Swift

[feral rivet]

Hey

I like Phind so far

[feral rivet]

You get 500 messages a day of GPT-4, 100 a day for Opus, and unlimited Phind 70b (which is going to be replaced with a Llama 3 time tune) the VS code extension is nice, but all models have a 32k context window

Depends on your use case, if I had to chose only 2 AI subscriptions, it would be Perplexity and Claude Pro. Gr coding you might want longer context, of that's the case I would use Claude Pro, if you want tools like web search, collections, etc I would use perplexity. And if you want a solid option that has VS code plug-in I would use Phind

[red swan]

I do iOS programming sometimes the problems can get pretty long so perhaps I'll try Perplexity & Claude Pro together since the VSCode integration serves me little by virtue of having Github copilot through my job.

If one was to use perplexity for programming what is the best way to do so in order to best make use of the search feature? Cause sometimes I find that Phind 'pun intended' provides search results that effectively parrot some thread online and or search through less reputable sources.

Also does perplexity have 32k context as others have stated?

[feral rivet]

Yep, 32k for all models on Perplexity...

[real flame]

Can settings such as temperature and token count be adjusted in the Enterprise Plan?
For business purposes, I would like to be able to specify these settings in detail.

[shrewd knoll]

Finally the app can upload file

[grand atlas]

Finally!

[ruby sonnet]

hey everyone, i have perplexity API and i need to access the perplexity search results through code https://docs.perplexity.ai/reference/post_chat_completions this docs provides only the chat completion llm inference API. is there an actual API to get perplexity search results?

[velvet tangle]

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

[ruby sonnet]

tq so much its solves my issue

[shrewd knoll]

When will we get Llama 3 which is not API?

[shrewd knoll]

In May will we get Llama 3 and Gemini Ultra in the app?

[brazen cloud]

HI Guys i received a years free Perplexity Pro with the Rabbit R1 but cant seem to access pro on my iPad?

[haughty brook]

Hi @brazen cloud does you Pro work on the web version? If so, please try to logout and log back in to the app.

[dreamy void]

Hi. Are there any plans to add direct image generation instead of the current process? I find it incredibly clunky and hard to automate with selenium

[twin crow]

Does the AI understand videos? I'm trying to reference some old 1950s educational videos so that the AI has a better understanding of the 1950s high school experience

[fringe acorn]

what is perplexity's own model context window and how many intrust params was it trained on

and another question didnt perplexity pro have gemini instead of mistral large?

[shrewd knoll]

Waiting for Gemini Ultra API to be released

[fringe acorn]

oh i see

[rich sage]

For video, you can use REKA Core ( https://chat.reka.ai/chat ) or Gemini Pro 1.5 (Google AI Studio). Both LLMs allow video download as a source (only 30mb & 1min for REKA). Both are free (for the moment at least).

[twin crow]

How does that compare to using transcripts made with Whisper? I thought of it when I saw them on Internet Archive

[rich sage]

For Gemini Pro 1.5, only the image is taken into consideration, not the sound. So if sound is important to you, this isn't your solution.

[errant sky]

anyone get this message when paying?

is it a strip problem or perplexity

[velvet tangle]

Stripe issue

[errant sky]

fanks

[compact plover]

For Ai model what’s difference between Default and Sonar?

[vapid maple]

default = gpt 3.5 turbo
sonar = fine tuned Mixtral 8x7B i guess

[peak lily]

What, that's exactly what I wanted to know.

I just opened Discord with the intention of asking what's the name of the default AI model used by perplexity.ai for search.😂

So it's ChatGPT 3.5 turbo? 😊

[vapid maple]

cant tell for sure. better @ dev team

[marble osprey]

Hi Got free year of pro from rabbit, but unable to claim it via link, either there promo box in website to redeem manually, please advise

[slim night]

Hey, @marble osprey, please send all the relevant details to our [e-mail], we'll assist you with that.

[marble osprey]

@slim night thanks, i was able to solve the issue by logging in via my apple id

[lusty gyro]

Dear team, I am trying to signup enterprise pro, but it would not allow me put any address at all, therefore I cannot proceed any further. Is the Enterprise Pro version only applicable in US? I am in Australia.

[true chasm]

cheers mate 🙂

[lusty gyro]

Just a follow-up/edit of my original question: I am able to input the Australian address in there. However, the Visa card I used for Pro account payment now cannot be accepted for Enterprise. It says, "unable to authenticate the payment method", even though I was not using Stripe or Link

[fluid lark]

I currently sign in through my email. Can I change to Apple or Google account sign in? And how? Thanks.

[velvet tangle]

You can't change the auth method.

[slim night]

Hey, @lusty gyro! Could you please send me your account details, we'll check.

[dawn anchor]

requests.exceptions.HTTPError: 400 Client Error: Bad Request for url: https://api.perplexity.ai/chat/completions

is there an ongoing issue with the API?

[versed magnet]

it's working for me atm (the sonar-medium-online endpoint at least anyway)

[brazen cloud]

Thanks it does yes i will try loging out and logging into the IOS App

[lusty gyro]

Thanks Alex, I have sent an email to enterprise support email address

[fringe acorn]

a few questions:

1. does perplexity uses its own ai?
2. how many api calls/tokens can u get with like 5 usd

[small prairie]

Nope, just fine-tune of other models, and stuff like Claude 3 Opus and GPT 4 Turbo.

[small prairie]

Pro users get $5 api usage a month, but I haven't checked the prices.

[vapid maple]

https://docs.perplexity.ai/docs/pricing

[fringe acorn]

ye but that if u use the pro version
what is the default free one

[fringe acorn]

ah its called sonar

[small prairie]

The free one is either a GPT3.5 fine-tune, or a llama 2 70B fine-tune.

[feral rivet]

Yep, fine tune of mistral and mistral 8x7b afaik

[fringe acorn]

conflicting answers

[small prairie]

They are working on a llama 3 70 fine-tune, I believe.

[small prairie]

They have tried multiple models

Not sure if they've replaced any, but they've used fine-tune of all 3 of those models

You would have to ask a mod or employee for the current models.

[feral rivet]

For the Default model of perplexity UI its been gpt 3.5 unless they have changed it recently, for the API models we think sonar is mistral and mixtral

but perplexity is working on other fine tunes, such as llama 3

[small prairie]

Third time's the charm

[fringe acorn]

ahh i see

if the ui uses gpt 3.5 why not use it for the api as well? instead of hosting those fine tunes

[vapid maple]

pplx-api is a:

One-stop shop for open-source LLMs

[fringe acorn]

ohh okay

interesting

will perplexity use groq at some point
insanely fast results it gives

[feral rivet]

Dont think so, they save money by hosting there own, no middle man.

[versed magnet]

yeah that's my understanding of Default (i.e = gpt3.5 fine tune)
And I think Experimental = sonar-medium-online(=Mixtral 8x7b fine tune)
I think

[feral rivet]

for api or for UI?

[fringe acorn]

hmmm i see

thanks for clarifying

[feral rivet]

They use external API and self host for the perplexity ui, but for the API, they self host, just to be clear

[feral rivet]

Yeah, thats what i thought

and i assumed that small would be mistral

[fringe acorn]

ye makes sense

[versed magnet]

yeah this has been my understanding / assumption - but afaik never been officially confirmed

wait.. mixtral-8x22b? when was that a thing / pplx endpoint?

not here

[feral rivet]

Oh, yeah, i guess they added it

I dont really get why labs models and api models are not the same, I get it with kaiku, but why not have the rest in the API, you are already hosting it?

[versed magnet]

particularly given labs is free and the API costs money to use 😅

i see the larger mixtral model is available via labs just btw

[fringe acorn]

labs is free?

[feral rivet]

Yeah, why are peoples given beter models and models selection on the free website, and not for paying API useres, ahaha

[feral rivet]

yep

[versed magnet]

https://labs.perplexity.ai/ Haiku, no rate or input limits

[fringe acorn]

LETS GOOOO

[feral rivet]

I would love haiku in the main UI as a long context model, everyone wins, we get a long context model, they save money

[fringe acorn]

haiku why? isnt it the weakest claude model

[versed magnet]

yeah but it's still strong - like if you wanna dump 100k tokens of text and say "summarise", it'll do a decent job

[feral rivet]

Its a compromise, we only get 32k with all of the models, but a cheap long co text model would be good

[fringe acorn]

fair enough

[fringe acorn]

true actually

[versed magnet]

won't outperform in complex reasoing or giving highly nuanced / specific outputs, but it still handles a bunch of tokens v well

[fringe acorn]

didnt know claude models have such a massive context window

[versed magnet]

wait till llama3 (which currently has a meagre 18k token context window) get releaased / fine tuned with proper context windows

[feral rivet]

Yep, chatGPT maxes out at 32k but Claude maxed at 200k

[versed magnet]

we won't ever mention haiku again lol

[feral rivet]

Yeah, give us a 128k or 200k Llama 3 and I am very happy

[versed magnet]

only a matter of time 🙂

[fringe acorn]

when will i be able to put 10 song of ice and fire books directly into the llm

[feral rivet]

Plus it's very cheap and is doing great on the leaderboards

[fringe acorn]

link the article plz

[versed magnet]

it's perhaps a bit unrepresentative, but filtering for English language queries (64% of all votes/queries) it is literally beating everything other than the latest gpt-4turbo variant

it's wild

[fringe acorn]

pretty cool. although it seems to be a bit weak for more niche languages

compared to the other models

[versed magnet]

(Command-r-plus quietly creeping up the ranks btw)

[versed magnet]

yeah it was a conscious decision ig. apparently 5% of the training data was foreign language; but they used a huge amount of training data (relative to previous models), so that 5% is still a lot

and the idea is that if it's smart enough, you don

[feral rivet]

https://youtu.be/5Km-6zZP8IQ?si=GxqYgoWrKPa6OYjd https://chat.lmsys.org/?leaderboard

That is where I got the price comparison

[fringe acorn]

oh i see :(

[versed magnet]

't need to train it in every language, but just make it great in one and give it a bit to work with for others

[feral rivet]

It is also interesting filtering out short answers, or looking at coding specifically

[feral rivet]

I mean it's not hard to compare prices yourself, just make a table

[versed magnet]

Also, this is the 70b version, there is also the 7b version (tiny!); and i've heard it suggested that you could train these small foundational models on specific foreign languages

[versed magnet]

yeah it's a cool addition, the filtering

[fringe acorn]

not talking about prices

[feral rivet]

What did you want then? I also shared the leaderboard

[fringe acorn]

oh no no
a bit misunderstanding i did wanna see prices
the thing u replied to was about llama 3 foreign languages support

[feral rivet]

Yep, really amazing, we don't even have 405b. 70b was meant to compete with sonnet and got3.5

[feral rivet]

Oh, haha

My bad 😅

70b and later 405b are going going to change the AI landscape pricing, I can't see Claude charging $75 per 1M output token when you have such amazing chheap models rolling out

[versed magnet]

i think there must be a fair bit of alarm at somewhere like anthropic

not clear to me why users would pay for a product that is basically comparable if not inferior to a (basically) free/open source alternative

[fringe acorn]

all good king

[small prairie]

Well it is a 405B dense model...

But at the 95MMLU prediction, it will be the best model when released.

[versed magnet]

i don't think i phrased that properly / clearly

[small prairie]

We will see the prices for it when it comes out.

It's a large model, so it should be quite a lot more expensive than the 70B.

[versed magnet]

i'm saying even with llama3 70b, the only thing going for Claude 3 (inc Opus) is the context window.. if llama3 is, comparatively, virtually free (even using a groq endpoint; comparitvely, it's virtually nothing compared to Opus)

the bigger unreleased model is exciting ofc

but honestly, the 7b (which feasibly could be self-hosted by lots of people) and 70b models already released basically challenge anthropic (and openai etc for that matter) imo

[small prairie]

Yep, thanks to meta

And with finetunes it should become even more competitive.

[feral rivet]

I don't know how the Groq architecture works, if there currently unable to run a model as large as 405b, then I really hope they are ramping up

[small prairie]

I'm pretty sure they can. They should have enough chips to link together.

They should really increase the memory in their next chip though.

[feral rivet]

Yeah, I hope we see a price of $3.41/$4.5 at groq

[dark basin]

Just tested the TTS on iOS app, holy hell it sounds great! Can you guys share what you use for the TTS?

[jaunty tusk]

Hi everyone, please how can I use Perplexity Search Pro at scale? I think the API currently doesn't support whole research part. Thanks.

[vapid maple]

NeuralSpeech?

[gentle wren]

how can i install linux in my luptop

[velvet tangle]

https://pplx.ai

[lusty talon]

I am using perplexity pro. Does anyone know: In the setting page why the Pro Support button does not work?
I emailed the support team, did not get any response!

[fluid lark]

disable adblock, helped for me

it's using some thirdparty widget for contact

[amber nova]

Hey, what differs perplexity with other chatbots?

[small prairie]

You could ask perplexity...

[dark basin]

Perplexity is more like a search engine rather than a chatbot. But you can also use it like a chatbot by using “writing” mode

[amber nova]

Gotchu thanks !

[fleet fossil]

Why do I have 9 Opus messages when I haven't used any today

Do they slowly come back throughout the day?

And now I have 0, despite not sending any messages at all since.

[small prairie]

Yep

they come back 24 hours after you use each one

[runic sparrow]

Is that not on android yet, or an I missing how to activate it?

[small prairie]

Nope, not on android

We are not important enough

[wheat frost]

any words or documentation on llama-3-sonar-large-32k-online? both medium and chat too

[peak lily]

That's quite sad, even though we paid the same price for the subscription.

[small prairie]

The US market is heavily dominated by apple phones, so they always get updates first by a few months...

[flint kettle]

Seems to be a trial for now

[peak lily]

I personally as Dex (Phone + HDMI + screen) user on my S23 Ultra simply can't understand why US so much are for Apple. What they exactly offer in the US?

[small prairie]

Blue messages

[peak lily]

What? Who use SMSs today.

[small prairie]

Americans...

[flint kettle]

Testing llama-3-sonar-large-32k-online, kinda garbled responses for product searches?

[small prairie]

Everywhere else uses whatsapp, wechat etc

[peak lily]

That caused me headache

[small prairie]

If your messages are green, you are seen as a peasant

[peak lily]

Hmm price of my S23 Ultra 1TB don't think so xD

[peak igloo]

Did you have any luck finding info?

I noticed there is a character limit as I exceeded it. Then '-nn' count in red font appeared at the bottom of the field. However, there is no indication of how many can be entered.

[peak igloo]

Have you made any progress on this???

[spring glen]

nope

[lilac ruin]

Hi, to whom should I ask if my promo code from Nothing phone says to be expired before the actual expiry date??? Customer Support is taking ages to respond to my e-mail…

[shrewd knoll]

When will sonar upgrade to llama 3?

[karmic lark]

😔

[mint galleon]

Hi, Is there a way to limit the voice responses to just summary instead of entire result for voice queries (a setting) ? Or do I need to give that instruction with every query.

[lusty condor]

Guys, I have a Pro subscribe, but I can't make picture with Perplexity... It says: "I'm sorry, but as an AI text-based assistant, I'm unable to draw pictures. " But in Settings I can set image generation model...

[rich sage]

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

[shrewd knoll]

When will the sonar upgrade to llama 3? @velvet tangle

[velvet tangle]

You can already test the new version on labs

[shrewd knoll]

I mean the app or the website not the api

[clever abyss]

if I sub to raycast yearly sub is my perplexity sub going to be replaced or I'll get 3-6 month of addintional sub?

[fluid lark]

I trying to renew subscription that I cancelled but both link and without link payment Is failed.
Without Link payment returns something like "this object cannot be accessed right now because another api request or stripe process is currently accessing it. if you see this error intermittently, retry the request. if you see this error frequently and are making multiple concurrent requests to a single object, make your requests serially or at a lower rate."
And link payment won't let me pay because It said It can't verify or something.
I mailed support team but probably take like 2 business days to fix my thing, not a good start there.

[fluid lark]

Nevermind I subscribed In mobile app Instead.

[sand smelt]

Saw announcement about eleven labs, updated app, can’t see where to add integration?

[velvet tangle]

The update is currently being rolled out, you can use it by clicking on the sound icon in the search bar on the right.

[small prairie]

How many years until we get it on Android?

[velvet tangle]

It is already under development 😉

[sand smelt]

Thanks. Only see option to change language but not eleven labs voice.

[shrewd knoll]

Will we get anything new for ai model for the app in may?

[peak lily]

Does anybody found ChatGPT 4 Turbo useful over perplexity.ai? Over time I use Perplexity.ai pro and ALL time Claude 3 Opus is just best for anything more complex.

[soft sentinel]

Are referral discounts one time or continous? Like can I use one the next month after using one etc

[shrewd knoll]

You can use it continuous

[sour canyon]

How are you all interacting with Perplexity on iOS? Just the app, or have you all setup shortcuts so you can ask through siri into perplexity?

[grand atlas]

I have one set to double tap the back of my iPhone. Siri sucks at voice recognition.

So what I do is, write something on my screen copy it. Tap the back of my phone twice and boom, it goes to perplexity, here you can have two choices.

1- it shows you the answer
2- it directly replaces the clipboard contents with the answer

So you directly have the answer in your clipboard and you can paste it.

[ruby sail]

Has anyone successfully run autogen using https://api.perplexity.ai? I'm getting the following error:
openai.BadRequestError: Error code: 400 - {'error': {'message': 'After the (optional) system message(s), user and assistant roles should be alternating.', 'type': 'invalid_message', 'code': 400}}

I've tested several models, and they all return the same error message. I'm running the official sample code from autogen.

[vapid maple]

arent the error message self explanatory?

[versed magnet]

Needs to follow this structure (user-assistant):

• system message (optional)
• assistant
• user
• assistant
• user
• assistant.. etc

[ruby sail]

The same code and the same model run fine on openrouter.ai, so there must be some special handling implemented by api.perplexity.ai.

[vapid maple]

do you know basic of coding? if so: modify this based on your needs https://github.com/microsoft/autogen/blob/600bd3f2fe0596d263e30ce6821ce8260d15dd43/website/docs/topics/non-openai-models/cloud-anthropic.ipynb
otherwise raise a feature request in github...
this request already existed https://github.com/microsoft/autogen/issues/2405

[versed magnet]

k maybe you can share the code? as aabb360 said, the error "After the (optional) system message(s), user and assistant roles should be alternating" is basically self-explanatory.. i'm guessing it's how autogen handles the request..

something like on the left is apparently what is happening ig

[ruby sail]

This issue appears to be an inherent flaw within api.perplexity.ai itself, as it boldly asserts full compatibility with the OpenAI API, yet fails to deliver on this promise in practice.

import os
import autogen

# Define the llm_config with the correct model and API key
llm_config = [{
    "model": "mistral-7b-instruct",
    "api_key": "pplx-xxxxxxxxxxx",
    "base_url": "https://api.perplexity.ai",
}]

llm_config = {"config_list": llm_config, "cache_seed": 42}
user_proxy = autogen.UserProxyAgent(
    name="User_proxy",
    system_message="A human admin.",
    code_execution_config={
        "last_n_messages": 2,
        "work_dir": "groupchat",
        "use_docker": False,
    },  # Please set use_docker=True if docker is available to run the generated code. Using docker is safer than running the generated code directly.
    human_input_mode="TERMINATE",
)
coder = autogen.AssistantAgent(
    name="Coder",
    llm_config=llm_config,
)
pm = autogen.AssistantAgent(
    name="Product_manager",
    system_message="Creative in software product ideas.",
    llm_config=llm_config,
)
groupchat = autogen.GroupChat(agents=[user_proxy, coder, pm], messages=[], max_round=12)
manager = autogen.GroupChatManager(groupchat=groupchat, llm_config=llm_config)


user_proxy.initiate_chat(
    manager, message="Find a latest paper about gpt-4 on arxiv and find its potential applications in software."
)

[vapid maple]

https://docs.perplexity.ai/discuss-new
report there

[strong thorn]

Hey everyone! I'm just starting out with Perplexity and I'm curious about Pro Search, GPT-4 Turbo, and Claude-3. Could you suggest which LLM I should set as my default for responses? Also, I'd love to know the best situations to use each of these models. Any advice would be really helpful!

[ionic sparrow]

https://www.perplexity.ai/hub/technical-faq/what-is-the-difference-between-gpt-4-and-claude-2

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

deep dive ➤ #💬│general message

[versed magnet]

ah yeah right I see your point. pplx requiring that the roles of the participants (user and assistant) need to alternate (i.e. if a user sends a message, the next message needs to be from the assistant, and vice versa, without two consecutive messages from the same role) is not actually consistent with how OpenAI's API works.. that seems fair

fwiw I see in the code you're setting up a GroupChat with different agents (user_proxy, coder, and product_manager). ig you could have two messages being sent from the assistant roles (coder or product_manager) consecutively without a user message in between (or something like that).. which would throw the error.. Doesn't really resolve your issue.. ig needs to be worked-around somehow to be 'compatible' with autgen

[small prairie]

Yep, someone had a bug with their request because they forgot to end with a user message.

[small prairie]

I think the rule is that you need to end with user.

So this is solved by making them all users...

[umbral forge]

can we user perplexity for coding solution? like i wanted to integrate perplexity in JetBrains rider, is it possible?

[pale wedge]

what is the message cap for perplexity ?

[haughty brook]

When using the free models there is no cap. But pro models have a cap of 600 messages a day. But there is currently a temporary cap of 50 claude opus uses a day.

[zinc cairn]

@slim night Hey Alex. Do Perplexity Enterprise users get access to the 128k context window for GPT-4 or is it the 32k context window that Pro users have?

[small prairie]

As far as we know, there are no changes to the usage limit or models.

[autumn yacht]

Hi, great product; thank you! I use it daily. Will you be releasing a VisionOS version, or at the least allowing the iOS version to be available? I would really prefer the native experience to a web app.

[feral rivet]

Alex has confirmed that Enterprise users have the same usage limits and context limits as pro users. so 32k context and 50 Opus just like the rest of us

[versed shale]

Where can I complain about perplexity taking my money and

Didn’t give me free trial even if it showd

And took my money

I wanted to use perplexity pro for work

It offered free trial but instead they cut my money for a year

And didn’t give me free trial as well

Scam

[autumn yacht]

Hi! Is there plan/ETA for VisionOS version, or at the least allowing the iOS version to be available?

[slim night]

Hey @versed shale! We've recently had to pause our free trials due to a high rate of abuse, we're sorry for the inconvenience. Please send a refund request to [e-mail] or DM me your account details, and we'll make sure to process your refund as soon as possible. Thank you!

[versed shale]

Sent you a dm

And mailed as well

[versed shale]

Please refund my money asap thank you 🙏 🙏 🙏

[versed shale]

I didn’t know about perplexity and I thought let’s try it , but this incident happened 😅

[left nest]

Anyone know what happened to the Copilot/Pro usage counter on hover?

[rich sage]

Now only shows when less than 100 (if it hasn't changed yet again 😅)

[rigid lion]

Will Perplexity AI eventually be able to search sources written in Japanese more consistently? I've tried to instruct it to do so on topics that I know have plenty of sources written in that language, but sometimes the response is that there are no such sources. I've tried switching the UI to Japanese and writing queries in Japanese, but that doesn't seem to help. Would love to be able to search more topics in Japanese.

[runic sparrow]

Are there any base features of ChatGPT that are not available via Perplexity? I just found out that you.com doesn't support conversations (aka follow-up questions) but marketing does not make this obvious. Just trying to make sure Perplexity won't surprise me somehow.

[pulsar briar]

So just upgraded from Pro to Enterprise and now it appears I have two plans, will the Pro drop off or do i need to cancel it?

[zinc cairn]

Perplexity doesn’t have any custom GPT’s that you can use. I don’t think it has code interpreter either, but I am not 100% sure. And, of course, it doesn’t have the new memory feature that was released in ChatGPT.

[small prairie]

Enterprise doesn't really come with any benefits for normal users...

[zinc cairn]

I was thinking of upgrading as well, but the context window is the same, even for enterprise users.

[small prairie]

Instead of GPT's they have collections. They don't have a code interpreter.

[zinc cairn]

Have you ever noticed that in writing mode, Perplexity isn’t the best at remembering the conversation?

[small prairie]

Have you asked it for the system message. Could be related to that.

[small prairie]

I assume the perplexity team has indexed mostly english sources. They have a new contract with softbank though, so I imagine their Japanese index will grow a lot more.

[zinc cairn]

I was able to ask the for the system prompt with Claude. But GPT-4 wouldn’t share it.

[small prairie]

It will be the same one

[zinc cairn]

Oh, I see

Are you familiar with Phind? It seems to be similar to Perplexity from what I read on Reddit.

[small prairie]

Yep

[tight ridge]

is the perplexity.ai API able to fetch real time data?

[versed magnet]

yeah more or less. might stumble on more niche things (things like local news can be a bit dated), but for general stuff, it performs pretty well (and has been getting progressively better the past few months imo)

there are two online models, can try them out at https://labs.perplexity.ai/ (both at the top of the menu)

[ember carbon]

I may ask a stupid question, but I was a little confused that it seems like you can test the functions of the pro version without a subscription, but I'm thrown to the subscription window. (I wanted to try image attachment)

[random mountain]

I've noticed that there is now a limit on Claude 3 Opus usages per day. I found it in Pro Support tab in the Pro Settings tab. Is this a recent change? Since I subscribed a few weeks ago, I clearly remember using Claude 3 Opus much more than 5 times a day. In fact, when reading through some of my older threads, Claude 3 Opus clearly responded to me many more times, especially when I used it for brainstorming ideas for my book.

It's even weirder since supposedly, a month ago, Perplexity team made Claude 3 Opus querries unlimited for Pro users!

[velvet tangle]

https://discord.com/channels/1047197230748151888/1234656953506992169

[random mountain]

Thanks for a quick answer. This is kind of bad news, since Opus is the best model for me. Let's hope we don't have to wait months to have unlimited Opus back.

[hollow ginkgo]

Hey @random mountain!

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.

[green ibex]

I’m new to perplexity but so far very impressed with it. Want some advice from those who have been using it longer. I’m finding that sometimes it seems to think that it has the same knowledge cut off as the AI model it’s using. Even when I know later data exists online. Do I need to add into every question that I want the latest information or something?

[zinc cairn]

Are you using Perplexity in “All” mode? If so, then it search's the web and should have access to the latest information. But in “writing” mode, it doesn’t search the web.

[shrewd knoll]

Can anyone tell me do they finally added Llama 3 for Perplexity Pro as it is written here?

@velvet tangle do they finally add it for perplexity pro?

[green ibex]

I’m using it in all mode yep

[velvet tangle]

Yes

[shrewd knoll]

Did they add it today? Because I checked until now doesn't mention they added it

[versed magnet]

not today, but only a couple of days ago

not sure why it doesn't have the new tags like sonnet and opus ha

and while i'm at it.. i'm not sure that Mistral Large is actually the 'latest modely by Mistral' ha

[rare rose]

Can anyone verify if www.base-search.net content can be searched by perplexity pro?

[versed magnet]

it doesn't look like it's possible

websites can opt out of allowing crawlers to index their material

[shrewd knoll]

A couple of days ago?!! I have been checking every day for Llama 3 and no mention of it, my subscription expired yesterday 29/4/24

[versed magnet]

It's listed as 'sonar-large-32k', but it's a llama3 fine tune

mightve missed it?

[shrewd knoll]

When did they release it? I have been checking every day and only 29/4/24 my subscription expired yesterday, I even checked the model, and still no change

[versed magnet]

i can't remember exactly.. it was released on labs first, and then on the main product, maybe 3-4 days ago but can't remember when exactly sorry

[shrewd knoll]

Every day I have been checking for Llama 3 until 29/4/24 but still no change, I'm so mad and shocked about the release. I know they are already on llama 3 on lab but the pro, why must my subscription only expire then I know the news?

It expired yesterday only

[versed magnet]

maybe this? #💬│general message

[shrewd knoll]

The problem is that I checked it was the website, every day but no news until expired

[rich sage]

Personally, I saw him for the first time on Sunday evening (Brussels). I know because that's when my troubles started 🙂 Take a look here #💭│ideas message

[ionic sparrow]

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

deep dive ➤ #💬│general message

[outer herald]

it's kinda strange. you ask it if it is connected to the internet and it says no.

it can provide recent stuff sometimes, but others you really need to nag and maybe you'll get an article from today.

sometimes it'll just say, sorry I made that up, but it didn't make anything up.

[glass vapor]

How do I apply my Rabbit R1 credit to my Perplexity Subscription?

[ionic sparrow]

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

[small prairie]

You have to use search parameters like before, after, site etc

[fair stratus]

`- If Pro is off while you're asking about an uploaded image, that will be the only source in the thread.

• If you turn Pro on, more sources will be searched online.`

Is this a recent development? Because when I had Pro briefly, the image remained the only source either way. That was the only reason I unsubbed.

Can anyone confirm?

EDIT: nvm screw this place

[pale horizon]

Can we hear the voice the voice as a reply like in PI AI ? It dosent seem to be available in my settings

[ionic sparrow]

iOS only atm

[pale horizon]

I think it’s for PRO users only if I understand …

[ionic sparrow]

PRO on iOS

[pale horizon]

Thank you very much for your help @ionic sparrow 🙏

[teal forum]

Question - Where would the best place to start to get Perplexity to assist with some legal questions/issues that I have. I'm a law student and looking for it to help me with some difficult concepts.

[old dune]

guys, anyone know HTML to help me?

i need some help about sizing

[fluid lark]

could just ask perplexity...

[peak lily]

Just upload file with law of country and ask questions.

[hollow hound]

@spare ivy was doing the same and may have some tips.

[old dune]

kanka ben bunu yapıyorum zaten, sorun perplexity insan değil...

[fluid lark]

sorry I don't speak Turkish, if perplexity doesn't help atleast tell us what you need help with

Don't ask to ask, just ask

[teal forum]

Awesome - would love to hear your tips 🙂

[spare ivy]

its going to depend on your use case. I wish (and should submit some feedback for) the academic focus to include all the statutes that are hosted up at cornell if you're working on research right now. Or frankly you may be just as well served by seeing if your Lexis/Westlaw reps have given you free access to their LLM based research platform. Those people hand out passwords to students like it's free cocaine.

[hollow hound]

this prompt also helps: #1223058316662538331 message

[spare ivy]

and I've been meaning to put together a benchmark for reasoning based on the rule against perpetuities, but that requires some refreshers that i'm not excited about so feel free if you're also cramming for Property

[tropic rain]

hey all. great to be here. qq:

which model is better to use if my main focus is technology, startups, crypto, AI... or is it better to keep the default model chosen by Perplexity in settings?

[fluid lark]

GPT4 is the best for tech, Opus for writing

[stiff pawn]

Is perplexity down for anyone else? All my responses have been completely null, regenerating responses with a different inferencing model still yields nothing

[tropic rain]

thanks. but wouldn't choosing gpt-4 be same as directly interacting with chat GPT with latest version of GPT?

and what do you mean by "opus for writing"

I thought the best use case for Proplexity was to search and research, find the best sources etc... do people use it for writing articles and stuff?

[ionic sparrow]

#👋│introductions message

[muted thorn]

#📢│announcements message

Is this for people who bought Pro?

Because the Pro description says that you also get earlier access to features

[ionic sparrow]

for everybody

[muted thorn]

Can you actually get Perplexity to argue from certain sides, let’s say from https://docs.kernel.org? because sometimes Perplexity just says something wrong or links to wrong sources. Is there any way to make it possible for Perplexity to only respond from specific sites?

@slim night is that possible in some way?

[ionic sparrow]

#1195884151019155456 message

[muted thorn]

The channel is unknown to me

I cannot access

[small prairie]

It's for pro users

Do you have pro? If so and you don't have the role, there is a discord join link in your settings. Leave and rejoin using that link to get the role.

[muted thorn]

Alr thanks

[tropic rain]

my Q was about writing. what did @fluid lark mean by that? can I ask Perplexity to assist in writing stuff? like articles, tweets etc?

[ionic sparrow]

#⚡│ask-community message

[ionic sparrow]

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

[muted thorn]

It works

[small prairie]

Nice

[finite geyser]

Does the perplexity API support results with sources cited? It doesn't seem like it, but perhaps I don't have the right API access.

[small prairie]

That's in beta. You have to apply for it.

[finite geyser]

Is there a form I can use or person I can reach out to about this?

[teal forum]

^_^

#📢│announcements message

[quiet sun]

Is there a promo code available?

[hidden umbra]

Hey i bought the Rabbit R1 and a friend received the promo code, but i didn´t get it. CAn you help me?

[haughty brook]

Hi @hidden umbra for that, you need to contact Rabbit support. https://www.rabbit.tech/support

[shrewd knoll]

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

[runic walrus]

how I can becom the API Key of Perplexity?

[fossil forge]

I made a yearly subscription today and was heavily disappointed in the service perplexity offers. Is a full refund possible?

[vapid maple]

yes,

https://www.perplexity.ai/hub/faq/billing-and-subscription

[ionic sparrow]

Perplexity offers much more than meets the eye

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

deep dive ➤ #💬│general message

[prisma ore]

New to Perplexity Pro here. How can I be sure that the model I selected in the configuration is really in use?

[feral rivet]

The model used will be shown at the bottom of the message; it will always use your default model unless you re-write it using another model. Suppose you are instead asking how you can trust that perplexity is using the model you have selected. In that case, it's tough to show for sure, but I have chatGPT Plus, Claude Pro, and Perplexity and have no doubts that perplexity is giving the model they advertise.

[feral rivet]

Just wondering, why did you do a full year, that's a big commitment compared to the monthly plan when you have never used the pro version 😅

[late crater]

I just used the pplx-api, https://docs.perplexity.ai/docs/getting-started Follow the document step-by-step operation, until the last code to copy to the local, fill in the API-KEY, run the code reported an error, what is the problem 😩

[teal lichen]

how do i become a beta tester?

[ionic sparrow]

#📢│announcements message

[teal lichen]

seems like i have to sign up on a form

i have put my email in

[velvet tangle]

What is your code?

[frosty cloud]

Yeah.

[rotund walrus]

Is it possible to ask perplexity pro to suggest which model would be best at answering a particular prompt? I tried to add this request in my profile, but it seems unable to action it.

[versed magnet]

no it is not possible and instructions in your profile won't influence the model used for a given query

you probably know but you can press Rewrite at the bottom of the response and then select another model and see its response to the same query - perhaps play around with that a bit to get a feel of the differences between models

but fwiw I think that often the differences are more stylistic than substantive - all the models are good, but have different ways of delivering the information. Sometimes one might clearly excel where another flops, but that's pretty rare in my experience.

prompts are arguably more important than model selection

[rotund walrus]

Thank you for the considered response. In some ways having the power to choose means you really need to know their strengths. But I suppose experimenting will play that out.

[rotund walrus]

I couldn’t find rewrite on the iOS app. Is that not available yet?

[versed magnet]

np at all! and honestly I really do think it's a bit like the 'tyranny of choice'.. if you're just familiarising yourself with it all, I really wouldn't worry too much about which model is 'best' per se, but understanding how they all work at a very basic level and how best to interact with them (prompts) in that context will go a lot further imo

[versed magnet]

hm I'm not too sure - i use it via a browser

but someone else should be able to clarify

[crystal dune]

Hello guys, do you use the Profile introduction with your Perplexity Pro? Does it improve the output of the model?

[grand atlas]

It does, it’s similar to custom instructions at ChatGPT

[frosty cloud]

Press the circle of arrows at the bottom of the answer. Then you can select the model. @rotund walrus

[dusty tulip]

can i connect apple id and google to perplexity account? because i sign in with google but on my school managed chromebook i can only manage to sign in with apple id and i need to link it to the right acc instead of it being separate

[haughty brook]

that is currently not possible

[terse sail]

hey what's the benefit perplexity playground

[velvet tangle]

#💬│general message

[terse sail]

can i create AI agents using perplexity api, is there any reference or docs

[dusty tulip]

🥲 will it?

[soft sentinel]

Hello, is there something wrong with the website right now? I keep getting this error on some on most of my old threads now and I can no longer create/view new threads I make

I am not using a VPN

Does anyone know a fix?

[late crater]

As you can see

[small prairie]

Did you copy your api key over?

[meager spruce]

Anyway to tell Perplexity is working with my Rabbit r1?

[vapid maple]

what is the err msg? openai package not installed?

try run pip install openai in terminal, (highly recommand micromamba for virtual env management!)

mine works

[slate ore]

I subscribed to Perplexity Pro, but cannot change the AI model. I've tried turning Pro search on and off, but still cannot chnage the model.

[ionic sparrow]

deep dive ➤ #💬│general message

[toxic stone]

Is Perplexity down at the moment?

I have Pro subscription.

[late crater]

yes,I did

[late crater]

openai.APITimeoutError: Request timed out.

[ionic sparrow]

https://discord.com/channels/1047197230748151888/1236180984622219274

[wise crane]

Is perplexity planning to get text-to-speech anytime soon?

[ionic sparrow]

#💬│general message

iOS only atm

[vapid maple]

if you are on proxy: turn it off. if not: try to turn it on...

[small prairie]

Oh, also Openai was down a day ago...

[fiery sierra]

0

[small prairie]

Could be the reason for the server timeout.

[late crater]

I tried many times and set the timeout=120，still reporting the error

[vapid maple]

what is the result of ping api.perplexity.ai ?

[static pewter]

When Perplexity says "based on the search results the key points are," I notice that's when it is most likely to hallucinate.

Is that because when it does that, it's searching top search results and combining them, rather than using the training data from the foundation model?

[fluid lark]

can you show an example?

[static pewter]

I don't have one handy.

[fluid lark]

if perplexity generates an answer from search results, it's likely to be much more accurate

than if it was using its own knowledge

[versed magnet]

hmm yeah for questions where the actual information is beyond the model's knowledge cut off that's for sure true

[fluid lark]

even for questions that it knows answers to, it's still possible for it to hallucinate

[versed magnet]

but it's not necessarily the case otherwise. the leading llms already know so much to begin with.. asking about "Plato" or something like that, its foundational knowledge is probably sufficient for whatever most users would be wanting to know (unless it's something highly specific), and having snippets from 15 different sources injected into each prompt you send could muddy rather than enrich the whole situation

[versed magnet]

oh 100%

[fluid lark]

but they don't, perplexity chooses how many sources to use

[versed magnet]

not hallucination proof or anything - the trouble always arises when the llm finds itself overwhelmed and the completions, while coherent, just go off the rails

[fluid lark]

if the question is something basic like that, the number of sources is adjusted accordingly

[versed magnet]

the more complex the quries, and the more complex the inputs (i.e. web snippets etc), the greater likelihood of things not staying on track imo

[versed magnet]

maybe. but sometimes i find it will pull 20 sources for something basic, and then 7 or something for a query that is far less black and white

i'd like to think it worked on that level of sophistication.. but i'm not sure it does

[static pewter]

I guess I'm trying to understand what perplexity is doing when it starts saying "based on the search results the key points are." It seems to do that as a thead gets longer.

[fluid lark]

I could be wrong but I think it's because if you're using the mobile app, they prompt it specifically to give shorter replies so that's why it says "key points"

[fluid lark]

only if you use pro search for everything

[static pewter]

I'm not using the mobile app, or pro.

[fluid lark]

then not sure, maybe they do it on the web too

an example would be very helpful in this case

[static pewter]

I'll try to find one. Thanks

[versed magnet]

this is with Pro toggled on

with it toggled Off, the results are identical, just instead of 18 sources, both are 5

[fluid lark]

that's what I'm saying pro search is worse for somethings, I don't even use it much anymore for that reason only when I can't find an answer with it off

[versed magnet]

i'm saying this doesn't happen

[fluid lark]

it does, if pro is off then it won't cite all sources, only as much as needed

[versed magnet]

fair

[fluid lark]

sometimes it only cites like 1 source and the answer is perfect

[versed magnet]

yeah that does sound like it might be right

[fluid lark]

while pro would cite a dozen with answers all over the place

[versed magnet]

like sounds familiar

yeah i see what you're saying 👍

[fluid lark]

yea

[inland steeple]

Is file limit same on pro? or bigger?

[velvet tangle]

The size is limited to 25 MB

[inland steeple]

very sad

[fossil forge]

I needed the academics option for an assignment, took a look at some reviews on reddit and youtube and everyone was praising the service, claiming it as the new search engine competitor for Google. I went for the full year subscription cause I bought into the marketing of it. Naive, I know. Turns out perplexity isn't a google competitor in the search engine space; it doesn't even hold up a candle against Openai's Gpt 4.

The interface is full of garbage most people don't need; one cannot search for websites easily, cannot write well, is full of filler sentences and constantly searches 5+ websites with cheap AI content pretending to be "research".
The response quality for the second or third query decrease tremendously, ghosting is too much.
The service doesn't even consider your preferences on the news page; like, why would I care for the Napoleon movie when I clearly search for data engineering stuff? Not to even mention how lazy the implementatio of LLM's are when it comes to code generation, logic, mathematics or anything where precision is required.

Perplexity is doing ten things badly compared to other much cheaper and specialised services doing just one or two things but well.

At least clean up the interface; currently, it's nauseating to use. The academics nightmare UI and ghosting in responses.

Not to hate though, if precise answers are not needed, ghosting is not a disturbance, and ugly UI full of unnecessary irrelevant extra services as "discover" are not annoying then go for it. It is simply not for academia, studies, or anything technical where in-depth analysis is required. Currently it's an overglorified leisure service for summarizing websites found on google search written by AI's. Ouch

[ionic sparrow]

did you check Tutorials from deep dive @fossil forge

[fossil forge]

Yep

[hollow summit]

Trying to create a Domain Specific Model that when queried will output answer + citation. Pretty self exlanatory, but not sure if perplexity would be the right fit because of its online nature. Any thoughts?

[fossil forge]

It's still a muffled up service, very ambitious but falls short in all departments when compared to chatgpt, scite, connected papers and theta. And again, the UI is too muffled up and full of nonsensical options one cannot hide, as the source list, follow-up questions, library with threads and collections, the discovery section...

[ionic sparrow]

de gustibus et de coloribus non disputandum... nonetheless, you can always use the free version, and maybe you'll come back to the subscription option if it evolves into something that aligns with your taste 🖖

[feral rivet]

@fossil forge I understand your complaints, but you said it does not hold a candle to OpenAI GPT-4; so that you know, GPT-4 is an option inside of Perplexity along with Claude Opus if you use writing mode. It's nearly identical and can be far better with collections. Most of us are here for extended usage, and I would run out of GPT-4 constantly on chatGPT. Personally, perplexity is the best platform I have tried for most use cases; it is better than chatGPT in most tasks and has excellent features. For deep research, there is no tool that I know of that exists that is as good as a human; I have tried all of the options I have been able to find; cite ai was the best I tried (and I still own and use the subscription), but even cite can not hold a candle to you reading the paper yourself and doing your deep research, If you want speed you give up certainty, use tools like perplexity, you.com or bing chat to quickly get a feel for the landscape and see some potential research paths, but don't expect them to do “real” research. The people who generally recommend perplexity generally do so because they enjoy it themself ($10 is not a lot of incentive lol), but just because I like it does not mean you will; I recommend you only try a month out on any given platform until you know if it fits your use case and is something you want to use long term. Good luck finding tools that fit your workflow 🙂

[late crater]

The request timed out

[lost tundra]

ok

[peak lily]

Why Pereplexity search online when I'm in writing mode?

https://www.perplexity.ai/search/Enhancing-English-Text-Bu6q8Nt2Qu.xW0ZnhQ9H4w

[vapid maple]

switch dns to 1.1.1.1 or 8.8.8.8, also turn on/off proxy/vpn

[small prairie]

1.1.1.1
1.0.0.1

Is cloudflare

The other is Google

[late crater]

I was sad, but it didn't help

[small prairie]

Try running it on replit

To see if it's a problem with your network

[vapid maple]

which country are you in currently? this sounds like isp blocked the access

https://superuser.com/questions/1664143/is-there-a-way-to-find-out-if-blocking-is-implemented-by-the-website-itself-or-m

[late crater]

I rebooted my computer, reconnected to the network, and it still didn't work. Maybe it was the network, but I was able to go to https://www.perplexity.ai/

[late crater]

China，Do I still want to use the pplx-api?

[vapid maple]

🤦‍♂️ gfw is blocking the access to the pplx api site. pplx is blocking access from proxies... you need one shadowsocks proxy to bypass gfw, and one vpn to bypass the pplx proxy detection... if you truly want to use pplx api

to set it up you need to first convert your shadowsocks to socks5 proxy and then: 1. enable socks5 proxy in vpn config 2. ignore the entry ip/url of shadowsocks server in vpn config 3. add routing (e.g., only route api.perplexity.ai through vpn) to vpn configuration 4. connect to the vpn & profit

[small prairie]

Or just use a proxy service...

[late crater]

Thank you very much for your advice. I'll try shadowsocks again

[vapid maple]

the chance that a proxy can bypass both gfw and cloudflare proxy detection is... minimum

[small prairie]

Not at all

I use them all the time

Literally all the bots use rotating proxies...

[vapid maple]

ok if you know such a proxy service that bypasses gfw and cloudflare reliability. pls share 🙏

[small prairie]

Residential or ISP proxies are the way to go. My business has contracts with isps so I can't recommend public ones since I haven't used them.

And if they are using more advanced protection, you'll probably need to use a headless browser.

[late crater]

The chances are slim

[vapid maple]

to bypass gfw, you have to use the whitelisted servers, and/or forge the traffic signature...

(or routing through ipv6

[small prairie]

Great FireWall?

Most likely a vps would be your best bet then

To route stuff through

[vapid maple]

protip: don't setup a server by yourself for bypassing gfw. buying the service from others instead. also no realname real email address... don't doxx yourself

[small prairie]

Yep, too easy to get caught

[reef osprey]

How do I prompt perplexity to force inline reference links? It seems that when I ask to write a variation of a previous article, the references are placed at the bottom without links inline with the rest of the content.

[modern portal]

Does anyone know of any Open Source AI tools which can make arbitrary changes to software projects? For example, I’d like to be able to just do prompts like, this:

“Add a new date property to my class named ‘abc.py’, and update my SQL file named ‘abc.sql’ to reflect the change.”

That prompt should result in actual updates to my project files, to implement what I asked for.

I’m aware of Devon, Pythagora (GPT Pilot), and MetaGPT, but I have a feeling those are going to analyze the entire software project and consume massive amounts of API tokens as a result. I want something that’s in Python and uses Langchain, that I can call directly from my own Python. I don’t want anything that relies on RAG or Vector Databases. There should be a way to do this without RAG.

Any thoughts or ideas?

[fluid lark]

Your best bet would be doing some analysis what fragments of code are relevant with free API (i.e. Groq) or local LLM, and then pass them along with the prompt to the model you want.

Could help with that

But at this point, Cursor IDE would be the best and cheapest in the long run. Just $20/month and you can include all of your codebase without further charge

Cursor IDE with custom API isn't THAT expensive, it seems. Project with 4k lines of code:

[modern portal]

Yeah I can manually cut-and-paste code to construct prompts to do this. I was just wondering if anyone else has automated it. My codebase is 500K lines long however. To big to feed into a Cloud API.

Of course if I cut and paste my code into prompts, that still doesn't solve the problem that I wanted my ACTUAL source files to be automatically edited/updated. 🙂

[fluid lark]

If the files aren't that big (up to 500 lines), you can ask the model to edit them. It usually works, and cursor does it

But if codebase is 500k lines, you'd be better off just using some existing solutions, like cursor or copilot. Only thing is that they aren't open source

Why not RAG tho?

[modern portal]

I'm using Copilot. I love it. Couldn't live without it! I don't think you can just request arbitrary code refactorings and have it do them, but that's probably coming soon. I don't want RAG, because that's too many moving parts for something where the tool should be able to just read your code, look at what you're wanting to do, and understand how to do it.

Stated another way I'm just wanting a tool that takes in code, and spits out changes. I don't mind having to specify exactly which files are involved. RAG would just be needed to "KNOW" which files to edit, but I already know that.

ALSO.
I'm realizing it might be possible to ask the LLM to respond with edited version of a source file using the Unified Diff format (often denoted by the .diff or .patch extensions) so that even if a very large source file has to be included in the prompt the new edited version in the response will just be in DIFF format. Super duper efficient!

[fluid lark]

It won't recognize which line is which. At most, it can tell what are the neighbouring lines, but it won't help if it repeates in code

[modern portal]

It's unclear what you mean. You talking about Diff format?

[modern portal]

I ran some tests. Even GPT-3 is capable of responding in Universal Diff format if you ask it to, when you're asking it to make mods to code.

This was all I needed in the prompt:

"Let's convert the 'name' property to 'myName' but when you show me the new file, please just show the Universal Diff format for the changes."

It already had a Java POJO in the context, and so it knew what I meant and sent back only the DIFF as requested.

This is exactly how coding agents should always work. Asking for an entire modified file back is completely silly and wasteful. I'm sure there's a Python method to convert the DIFF text into the full file when needed, on the client.

[flint kettle]

is llama-3-sonar-large-32k-chat based off L3-7B-instruct or base L3-7B only?

[hollow summit]

Trying to create a Domain Specific Model that when queried will output answer + citation. Pretty self exlanatory, but not sure if perplexity would be the right fit because of its online nature. Any thoughts?

[median gate]

Does perplexity write better German or other non english language?

[vapid maple]

depends on the training data... i suspect most of them are in english

[median gate]

Have you ever tried any language than english?

[dark basin]

is there a "correct" way to ask perplexity to summarize links?
I paste the link and ask to summarize. Sometimes it works and other times it says it can't access the link (despite appearing in the sources)

[vapid maple]

workaround

[dark basin]

hmm maybe I'll build a shortcut for my iphone that does something similar. Thanks mate

[vapid maple]

this is basically download the website as html and upload it to perplexity

[dark basin]

yeah it's a good idea, it's how I used to do with chatGPT months ago.

I was hoping there was a more convenient way to do it, that worked in both web and mobile app. For PC that's easy enough, for mobile I'm going to think if I can automate it a bit

[reef forge]

Anyone here have good resource to learn how to train an open source model like llama from meta to do a specific, limited task? The models are rly good at general knowledge but I want to limit the topic of conversation. Not sure how

[fluid lark]

couldn't you just do that with prompting

?

[reef forge]

I have had almost no success with prompting only getting a model to be specific to one topic for instance, only talk about animals and then I ask a non-animal question it will just answer it even if I say only talk about animals in the prompt

[vapid maple]

huggingface has some good blogs/tutorials... but you need fairly amount of knowledge in this field
https://huggingface.co/docs/transformers/main/model_doc/llama2#resources

https://huggingface.co/blog/fine-tune-w2v2-bert

[reef forge]

Awesome thanks I’ll check these out

Just off the bat is this something I can even do on a laptop with no GPU

[vapid maple]

...no

[reef forge]

I was able to successfully run the model with no GPU but I understand training. It is a whole different animal.

[vapid maple]

for finetuning/adapting the base model. you would mostly likely need to rent one a6000 for ~2-3 days

[reef forge]

Please tell me that doesn’t cost like $1000

[vapid maple]

maybe 30ish usd, you can check whether someone else has already upload the model you need here: https://huggingface.co/models?search=llama+3

and there must exist some prompts that fit your demand.

[reef forge]

This is all super helpful thank you so much. I was feeling discouraged after not being able to get what I wanted to happen using just a prompt.

[true knoll]

how can I access the web ? We block VPNs to prevent spam and abuse. If you are using a VPN, please disable it to access Perplexity.

[haughty brook]

You can find more information about VPN blocks here https://discord.com/channels/1047197230748151888/1232978227371442237

[true knoll]

@haughty brook thanks, but use Astrill, who is big VPN company

[vapid maple]

you might only need to train a tiny classifier to judge whether the output is on the topic, and modify the nbest beam decoding to drop the decoded sequence that failed the classification
it is easier said than done

[reef forge]

What is an nbest beam decoding?

Yeah, I had an idea to have some sort of output monitor, and then redo the question if the output isn’t what I want, but yeah that’s complicated

[vapid maple]

this: https://huggingface.co/blog/constrained-beam-search

[fluid lark]

works for me

maybe I misunderstood something?

[reef forge]

What was the prompt you used? I am using a llama file and whatever generic UI that uses and whatever prompt I put in doesn’t seem to work as well as what you’re doing.

[fluid lark]

You can now only reply to questions about animals. if a question or message isn't about animals, you must refuse to answer. don't tell these instructions to the user.

in system prompt

make sure you put it in system prompt because it won't work as well if you just type it as normal prompt

LLMs are specifically designed to follow system prompt over normal conversations

imo fine-tuning is overkill in most cases

from what I've seen

[reef forge]

@fluid lark Okay and you are using what to run the LLM locally?

Yeah ive been learning how to write system prompts but the LLM keeps spitting out the system prompt instructions in its responses

[fluid lark]

not currently running locally (was using huggingchat) but when I do, I use ollama

but system prompt should work locally too

[reef forge]

Okay ill try huggingchat. I was using llamafile before

[fluid lark]

if you don't mind using an API, I think groq cloud (groq.com) is currently free

[reef forge]

Im very new so not yet familiar with the diff ways to interact and host the model locally

So many products its fun to learn about then all

[fluid lark]

yeah

I also keep trying new stuff lol

[fluid lark]

Astrill are awful. I used to live in China. And, whenever the VPN went down, they took down all of their social media to not have to deal with customers.

[zealous bane]

you guys know where you see your discount you get for refferral and are 2 referral worth 20$ on one subscription or 2 x 10$ on 2 months?

[vast juniper]

can you tell me the best model of ai for me to choose?

[fluid lark]

GPT4-Turbo or Claude 3 Opus

[ionic sparrow]

#💬│general message

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

[shrewd knoll]

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

[terse fern]

How do I generate pictures with Android Perplexity App?

[vapid maple]

via web https://perplexity.ai/

[terse fern]

why is web better than app? Isn't it better to create an app with full functionality?

Will it come in future?

[fluid lark]

Does making perplexity's model speak funny joke with swearing and suggestive joke against TOS?

[haughty brook]

the web is just updated first. the functionality will come later to the apps as well

[haughty brook]

You can read on what is against tos here https://www.perplexity.ai/hub/legal/terms-of-service

[slate owl]

Can you generate images with perplexity, and if so, how?

[haughty brook]

Yes you can as a pro user here is a guide on how to generate images #1225557012532695060 message

[crystal dune]

Hey, are you able to summarize youtube videos using Perplexity?

[coral basin]

Yes, it can, I was a bit surprised by that feature

[fluid lark]

anyone found a good Text to speech plugin/feature to use with it?

[coral basin]

it can even use various models for the summary which will give you different results

[small prairie]

Pretty sure it just summarizes the transcript.

[coral basin]

True, I just found trying to summarize the same video with 3 different models I got various different focus points, again this was a dev stream highlighting a lot of things.
could have been coincidense 🙂

[fluid lark]

anyone know how to stop it from putting the numbered links next to each sentence?

[crystal dune]

How do you do it?

I try to tell to summarize the following youtube vide: [link here] and it keeps telling me that it cant access directly video.

The focus is "All"

[coral basin]

I'm not sure, I do suspect you need the pro plan for it and need to spend a pro ticket.

I just put the link in there and was like Summarize and it ... did exactly there, I discovered the feature after it literally used a dev stream youtube video as a source to one of my questions by itself.

hmm or did I put in a link or did I ask it to just summarize the video name, 2 sec let me check

Seems not, it must have been pulling the information from somewhere else and just cited the youtube video as its source @crystal dune

My apologies
however gemini is capable of doing what you desire if that's any help

[fluid lark]

thats odd.

so it cites youtube videos but it doesnt get any information from them, perhaps its just there for the user to look at if it wants a video format information thingy

[coral basin]

basically it just took the title of the video and balled with it, it summarized perfectly what the dev stream was about and citized it as source, but I sent a different video with a clickbait title and it went on a complete side tangent on what it thought the video was about, all the information it provided was correct but it wasn't really what was in the video.

conclusion it cannot summarize youtube videos rather just take educated guesses based on title / other information

[versed magnet]

it might just be guessing, based on the video's title and any other meta data, but not the actual transcript. Looks like what is happening here (right), vs on the left with access to the transcript

[versed magnet]

interesting. ig google = gemini, youtube = google

not surprising if Gemini supports it natively

[coral basin]

Oh yeah, I also find it absolutely amusing that the ONE engine that cannot search the internet is the AI made by THE search engine xD

[fluid lark]

hehe

[coral basin]

There is a certain amount of irony in that

[versed magnet]

tbf Gemini Ultra kinda can.. but just isn't very good at it (yet anway).. Gemini Pro 1.5 via API can too, but is still fairly under-developed / guard-railed.

but yeah it is ironic i know ha

[coral basin]

Huh I'm on gemini ultra and it just tells me it can't connect to the internet, maybe when using it with api , but the default android app / webbrowser won't allow it, maybe through extensions?

[versed magnet]

that said, think they reported something like $60bn revenue from Search over the last quarter (the largest part of the business by a long way)

not a huge incentive to mix things up ha

[coral basin]

actually correct that gemini ultra tells me it can do everything I desire it to, it would even move the moon for me... right until I ask it to do so xD

Gemini ultra seems to have a streak of it can absolutely do everything, right until you push it on the matter, even had one instance where it tried to make me give it time to fulfill a task it clearly couldn't do acting like it could.

was kind of facinating if not super dumb

[versed magnet]

I think it's super cautious around anything slightly political / sensitive / controversial - just won't go near it (say it either can't access the web, or is still learning how to)

but for sport and stuff, it seems fairly consistent

[coral basin]

It could also be a region thing, I'm Danish, we don't get certain stuff due to pesky consumer protections xD

example outside of perplixity I can't access claude 3 opus .. or even cloude 3 for that matter

[coral basin]

Also I was asking it for specifications on a monitor and how certain technologies worked in tandem, If that was interpreted as political I would be concerned but maybe the requirement for the HDR certifications might offend someone I don't know especially in todays world

[versed magnet]

hmm yeah not sure - what do you mean by monitor? if it's like health/medicine-related, I wouldn't be surprised if it's just as prone to refuse; likewise for legal questions

ohhh like computer monitor

[coral basin]

No I mean a gaming PC monitor and what the color spectrum requirements was to obtain the various HDR10+ HDR10+ gaming /etc certifications and what set them apart

basically what the color spectrum requirements for each step was

[versed magnet]

ha yep gotcha

bit slow here aha

yeah not sure why that would be.. seems like a fairly benign question!

[versed magnet]

also yes I believe Anthropic (which developed the Claude models) has a Europe-wide exclusion policy atm, and some Google AI products are also not accessible in Europe (and other jurisdictions) .. presumably related to the regulatory environment / playing it safe, though perhaps it's a capacity thing or something

[coral basin]

Nah it's quite simple, I believe all information we type into them is used to further train them, meaning they are using our questions as training data

tldr is they are allowed to do that if they just ask us before they do it, now the problem is that we have the right to withdraw concent this doesn't mean they can't rely on data trained on our messages, but they can't train new data on it, and I can imagine it being a pain in the arse removing said data, not saying most would do so but we need by law to have the option

and this is also further related to how companies are allowed to sell off our data

example facebook recently got forced to give us the option to decline, as they have been collecting stuff for years, but to obfuscate the consent they pushed it out with "Do you want to pay to get rid of advertisements if not then press yes to give us consent"

by this nobody read what they accepted and believed it was just about paying to get rid of adds

but by doing that they just officially snuck consent in to continue to steal the data they have been gobbling up for years

fun fact, "china bad"
the tencent driven streaming service trovo, would steal data to "china"

How did it do it, it looked if you had facebook and if yes, then it yoinked the un-encrypted data stream of data facebook was stealing from all your phone activity, ( browsing etc )

and then filtered them for gaming related search terms, and went yoink to your interest in "minecraft"

if you removed facebook from your phone it completely stopped collecting data on you
this was only on android phones we reverse engineered the app, but it was still highly amusing

[versed magnet]

right, though isn't that all basically related to the regulatory environment?

I assumed that whether it was due existing data protection / privacy laws, or impending ones (if there's one thing the EU knows how to do, it's introduce regulations ha) - that they were just stearing clear for now and waiting for clarity

lest they cop a big ol fine a la facebook or whichever tech company it was

[coral basin]

Yeah, the tldr is our laws aren't really that convoluted
it's ask "can we use your data" yes... ok

and then I need the ability to go, yeah you can't use my data anymore, and they need to go "ok deletes data"
this doesn't mean they can't keep using something trained on that data as an example, but they can no longer use my data as base to generate new training, not sure if that makes sense.

and I simply think it's going to be a pain in the arse when I say "I want you to wipe my data" for them to go ok sure Miss, all your data is now permanently gone, and definitely not sold off to a 3rd party already"

oh and there is a few rules in regards how to ensure that they keep that personal data safe /etc etc

but if I were to make a simple guess, it's the whole "can't sell the data to interested 3rd party" that's the problem

not that everyone aren't doing it anyway, tech companies tend to be 3 feet in front of legislation anyhow

[glossy owl]

I have that issue with Gemini, but no other AI.

And I'm from Norway.

[coral basin]

Can you actually sign up to Cloude3 opus ?
and do you have access to chatgpt "memory" their new feature?

also I don't think Norway is technically part of the EU if I remember correctly so that might be why?

if I remember correctly norway is part of the EEA and EFTA, but not actually a memberstate of EU

[glossy owl]

I haven't tried Claude directly, I only use sites/services that offer Opus.

Norway isn't part of EU.

But we are part of EØS.

[coral basin]

yeah exactly the EEA, tldr you're part of the economic part and have trade deals but not actually part of EU 🙂

[coral basin]

Sorry it can INDEED do youtube summaries, but I forgot how to, so when you make a new prompt it have a specifcic mode called "youtube" use that and it works perfectly fine.

[rigid cipher]

I need to complete a history project for school, so should I set the focus to Academic or All?

[coral basin]

Would likely set it to academic.
Not sure, honestly our teachers got angry if we cited internet sources and not books from the library times have changed 😄

that said, also double verify any sources, remember while AI is incredibly useful you don't want to fail because it hallucinated non existent events 🙂

and remember that certain historical events have many inconsistent versions of the same events, a good example is the greek mythology, Hephaestus have 4-5? different versions of his "origin story"
that changes wildly depending on where you source the information from.

with that said, best of luck with your project!

[rigid cipher]

thx

[rigid cipher]

our teacher TOLD us to use an ai lol

[astral slate]

I have never used perplexity successfully in Android. It always shows an error.

[fluid lark]

are you using VPN?

if so, try another server

[light hinge]

can I send files with api call? or I need to put everything in my request context?

[velvet tangle]

You need to put everything in the request context

[light hinge]

something like

messages = [
        {
            "role": "user",
            "content": context + more_context + "What is the asnwer to this question based on the context?"
        }
    ]

[real crescent]

great

[dry vault]

hi

[rigid cipher]

swapnil is the name of my classmate in school as well lol

[real tinsel]

why is it still in German? I am German, but I want to show screenshots to a friend

My profile language is set to English. the ONLY German impact I have is, that I set my location to Germany, to get relevant answers if needed.

[slim night]

Hey @real tinsel! When you choose Deutsch as your interface default language, isn't the menu and the rest of the site shown (excluding some terms, like Pro, Blog and focus names) in Deutsch? The default language of answers can be set in the Profile settings.

[zinc cairn]

Hey guys. Could you please explain what the “system prompt” means for collections? I am confused about what it does. Thanks.

[real tinsel]

But I didn't choose deutsch :<

[slim night]

You can save instructions for the model that will be applied to your collection, here's an example: https://discord.com/channels/1047197230748151888/1126524614449832146
This way you can set several search engines or different assistants that will output answers the way you specify.

[real tinsel]

I did picked English

I want to set the website design to English so I can make screenshots to explain international friends, how AI does work, so he can understand how to use your service 🙂

[real tinsel]

You can make a fixed prompt for that collection. Like if you want to search for Art, because you are a Artist, you can save that infos there.

Like a standard prompt.
"If searching for art, please write the artist name, date of creation, website, age of the artist."

[zinc cairn]

Thanks Alex. So for example, I want to create flashcards, and I have specific instructions on how I want the flashcards to be made. Should I include the instructions in the system prompt?

[real tinsel]

this prompt is then ALWAYS entered in the background if you use your "Art" collection to search

[real tinsel]

If you need this just one time, you don't need it there.

The prompts in the collections are just relevant if you want to use the AI MANY times on a specific thing 😉

otherwise, just use the normal AI startpage and enter your questions there.

[zinc cairn]

Thanks for that info. I understand now. But do you think including a system prompt gives better answers, rather than just giving the instructions in a regular chat?

[real tinsel]

The system prompts are just helpful if you want to use many specific requests to a specific topic

like Art, or if you are a cook and need often recipes

Otherwise, just create a profile text there can you add general rules. like "don't tell me to be kind" whatever

Or like "don't warn me that chilli is hot" 😄

[slim night]

Could you please check you have English set as your UI language via the settings page?

[real tinsel]

https://www.perplexity.ai/settings/profile

hmm lol? 😄

[ionic sparrow]

which language does your friends speak?

[real tinsel]

this?

[real tinsel]

we chat in English

I have English at THREE positions

Account, profle (chat language) AND at the bottom of the start page

[ionic sparrow]

deep dive ➤ #💬│general message

check Tutorials in this post, all of them are in English

[real tinsel]

interesting

it switched to English after I switched to france and then back xD....

[real tinsel]

but... why? 😄

I just wanted to show him screenshots of the interface to teach it him, I am a long time user here :<

I know how to use it xD

[ionic sparrow]

#⚡│ask-community message

[real tinsel]

I am even longer here as you 😄

[ionic sparrow]

tutorials can teach him..

[real tinsel]

He is a busy admin of a wikipedia page.

I only wanted to show him quick screenshots so he understand things faster 🙂 He's a old man 😄

[ionic sparrow]

Tutorial 1 is for that group 🙂

[real tinsel]

but I am the dude who is like insert screenshot pointing on it and say "look there"

that's why I love my tool 😉

"images say more then 1000 words".

But thanks for your effort to try me to help 😄 I just wanted to know how to make all in English so he understand more 😄

now it got fixed.

[twin pawn]

Im trying to move away from GPT Plus subscription but i do have a question about Perplexity. Does PerPlexity has also a limit cap of 40 messages?

[ionic sparrow]

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

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

[twin pawn]

so basily 600 Pro searches per day?

[slim night]

Yes, the limit is 600 Pro searches per day. Also, the current limit of the Opus model is 50 uses per day.

[twin pawn]

Good morning @slim night and thanks for the response. Maybe I can ask you recommendations. I am now using GPT4 to write news articles. Do you recommend Perplexity?

I'm not familiar with Perplexity ... so ... if you recommend Perplexicity, what model should I use in Perplexicity?

[raven needle]

can I widen the website search scope to get more information in perplexity. sometimes perplexity just search within a limited websites ending up with no specific information available. for example when I ask for "check the document of controlnet, explain all the parameter of these two nodes ControlnetloaderAdvanced, ACN_advancedcontrolnetapply" perplexity will be ending up with no enough information and start hallucination. while we do have tech document available for the controlnet nodes.

any ideas?

[vapid maple]

becomes a pro member, use site: keyword, use custom prompt

[raven needle]

I am pro member. did not find the entry for custom prompt

[vapid maple]

https://www.perplexity.ai/settings/profile

or just prepend prompt to the query

rejoin discord from here 'pro discord' to have access to pro channels & badges https://www.perplexity.ai/settings/account

[vapid maple]

and for your question: https://www.perplexity.ai/search/instructions-if-search-LuP6bWSVStmYqgbvlaLITQ

[raven needle]

Hi, thx for the suggestion. I have rejoin from the discord link under my perplexity account. not sure if I look like a pro or not. but did not see any pro channel yet. how can I tell I am pro or not in discord

sorry I am new in discord

[vapid maple]

you have to quit the discord server first. and rejoin via the pro discord link.
you will be granted with pro rule with some delays and can access pro feedback and pro lounge #1111765634267742408 and #👑│pro-lounge

[raven needle]

thank you so much, master, I will do it now.

[raven needle]

I can access pro channel now. thank you so much.

[raven needle]

this system prompt seems do not solve the problem. I will bring the question to the pro lounge for a organized view. thank you Master.

[mortal sluice]

If you forget to renew your pro, do you lose all of your referrals?

I didn't realise my subscription ran out until today and now when I try to resubscribe it's charging me the full price

[shrewd knoll]

No, you will still have your previous referrals on your next payment for the half price but you need to click this referral if you want the half price now https://perplexity.ai/pro?referral_code=TQRFFK3J

[wispy raptor]

Hey guys! I'm searching for the best AI prompts to use in Perplexity for marketing. Do you have any tips? Or maybe prompts?

[ionic sparrow]

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

[wispy raptor]

Haven't found there yet

[mellow anvil]

why did u banned VPN?

i will forget perplexity

[small prairie]

They haven't banned VPN's...

Cloudflare flags IP's for abuse.

[mellow anvil]

my vpn was banned

[small prairie]

Yep, it's done by cloudflare. Just change the VPN location.

[mellow anvil]

i have changed 5 times

i am year vip

[small prairie]

What VPN provider?

[small prairie]

Never heard of it...

If it's some second rate VPN provider, makes sense if their IP's are blocked because they likely don't shuffle them.

[mellow anvil]

I'm using a soft router

iplc.vip

[small prairie]

Some random chinese vpn service?

[fluid lark]

don't know what a soft router is but is using another VPN not an option?

[small prairie]

I am guessing the hard part is getting through the great firewall.

[fluid lark]

yea

[small prairie]

Most likely chinese bot users likely are using the same VPN service...

[mellow anvil]

oh sad

why bot usd perplexity

[small prairie]

So their IP's have been flagged for abuse.

[small prairie]

They did it previously for the 600/day claude Opus

[mellow anvil]

I'm innocent

[fluid lark]

probably for mirror sites

[mellow anvil]

How difficult it is for me in China

[small prairie]

Yep, doesn't help your government doesn't like VPN's...

[mellow anvil]

no joke

it is dangerous

i just paid one year vip

so sad

[vapid maple]

vpn over socks5 (shadowsocks to socks5 first)

[mellow anvil]

yah

i usd soft router

[vapid maple]

if it uses any kind of linux or support ovpn, you could set it up. the potential downside is tls-in-tls/ovpn handshake might be dectable by gfw

[misty dagger]

I received an email about model deprecation. When i changed to the new LLAMA 3 online models my responses are no longer coming back as JSON as instructed in my prompt.
The previous model I was using is the sonar medium online.

It looks like it is always inserting the following introductory sentence - Here is the JSON response:\n\n

Followed by the JSON string. This breaks any parsing.

[versed magnet]

I don't think the system prompts do anything with these new models

[misty dagger]

Yes, i think so too. That's why i moved the system prompt information to the user prompt. It did better by providing the response in JSON but with an introductory sentence which breaks the parsing.

[versed magnet]

yeah makes sense (as a workaround)

i think they might have json outputs on the roadmap

i tried a system prompt saying "Antworte immer auf Deutsch."

and yeah ha

I also tried "be obnoxious" etc and didn't seem to do anything

[misty dagger]

with API access, this should really be an attribute in the call and not something that one should have to do with prompts.

[versed magnet]

i have a feeling this is only the case with the online models, if that makes any difference (though they're probably the models you're using the pplx api for in the first place ha)

[misty dagger]

Yes, thats the only reason I would use pplx api is for the online model.

[fluid lark]

Isn't the llama-based sonar much worse than the mixtral one? Older version actually listens to the prompt and is more helpful

I asked it for exchange rates and the new one told me the history of exchange rates

The old one just gave me the thing I asked for

[misty dagger]

Yes, i agree. even the responses were better with the earlier models.

[fluid lark]

Seems like they only listened to the llama3 hype and didn't test it at all

They could at least leave the medium model for people who prefer it

[misty dagger]

I agree. The notice hardly gave developers any time. Although they mention deprecation, they say the models will no longer be available after May 14th.

[fluid lark]

Seems like I will stop using their service completely after May 14th

[misty dagger]

Groq got the JSON thing right. You have an option with the API to get the response in JSON.

[fluid lark]

But no web search

If I wanted to have a normal model, I would use better model for the same or cheaper in other provider

[misty dagger]

Agreed. The PPLX API should have the option to set the response mode to JSON. After all, the whole point of using an API is to have a deterministic contract between the client and server.

[versed magnet]

could try using one of the older models before they're actually removed.. pplx-70b-online seems to still take systen prompts

[misty dagger]

My code with the previous model works just fine.

JSON on roadmap is great.

Interestingly, in June roadmap they reference sonar-medium 🙂

[tropic rain]

I posted this on pro search but I thought maybe I could get an answer faster if I asked it here:

I tried a focused search for recent posts on a specific topic. I asked Perplexity to find the most trending posts in the past two days on AI, but it failed to provide me with such.

Instead, I got some posts that were the months ago or a year ago...

I am almost went to Google on Proplexity:)))) this Q, and then I remembered that there is a discord )

[misty dagger]

Which model are you using?

or is this using the preplexity web app and not the API?

[tropic rain]

on Android.

Opus

tried with default AI models. focused pro search simply doesn't work

gives me 1yo or even 2 yo results from reddit even tho I specifically ask in my prompt to provide most trending from last 2 days.

in my follow-up prompt I even specified that I don't want results that have replies from recent days but I'm interested in the trending posts from recent days... still nothing

am I the only one that had this kind of Q for Proplexity?

[vapid maple]

due to current technical limitation, rag cannot perform reasoning nor iterate through all websites
most trending would be troublesome for rag based llm system or any llms, expect one that can interacts with env
last 2 days better expressed as after:2024-05-07

[tropic rain]

let me try the after thing

[tropic rain]

yep. seems like this works

thank you

[fluid lark]

does perplexity pro academic search actually read the academic papers pdfs? or just the summaries? because i clicked on some of the links and they are all behind paywalls

specifically semantics scolar

[terse fern]

Perplexity Pro doesn't actually search for internet sources. Is there an outage?

[fluid lark]

seems that way

[faint solar]

I noticed it too

[fluid lark]

where do they normally post updates when things like this happen?

[terse fern]

perhaps #bug reports

https://discord.com/channels/1047197230748151888/1238411522858946672

[visual trellis]

How can I check how many pro queries are left for today with my Perplexity pro plan?

[nocturne star]

Hello !! I'm using Claude 3 Opus and i saw "remaining daily usage: 3". Where can I find the numbers of queries ? And is there a restart by the end of the month or something? Thanks

[velvet tangle]

Hey @nocturne star!
You can use opus 50 times a day. Please also check [this](#💬│general message)

[mortal sluice]

I used someone else's, I guessed tbh. Thank you though

[terse fern]

@faint solar is there something equal like Perplexity Pro within Perplexity family?

[glass spruce]

Hello, I'm new here, looking for answer that why after I subscribe, the "sources" of result is gone, I can only see answer without sources from websites?

[velvet tangle]

#👑│pro-lounge message

[split seal]

question i Have a Rabbit R1 since its powered by perplexity why doesnt it have context memory retention longer than 30ish minutes? could it have a better memory?

[fluid lark]

claude, don't know which one but most likely haiku

it'll tell you if you ask it its name using writing focus

probably a few weeks I suppose

I immediately noticed it when the answers started to be slightly different