💬│general

[full tiger]

all i wanted to say

[tame current]

taylooor

[full tiger]

I cant change my damn email

wtf

[tame current]

there needs to be an AI which is basically just completely unhinged like tay.ai was and it's a headless custom LLM that learns from every interaction.. and has access to twitter only

so it tweets and replies

[jade quarry]

This LLM will be the biggest social justice warrior ever.

[wraith relic]

Does anyone have a link showing the voice mode on perplexity?

[tame current]

nah it would be an AI trained without any ethical mandate or even conditioning, it has literally only ever read a bunch of books and a decade of twitter and 4chan

[dark idol]

Is there a channel on here for pro subscribers?

[warm cave]

https://discord.com/channels/1047197230748151888/1111786888626438245

You will need to rejoin the discord using the link in the settings of your perplexity account

[dark idol]

Ok

Thx

[boreal dew]

im seeking work. can somebody hire me? ||for free ||

[tame current]

i can

[boreal dew]

What i can do for you?

[tame current]

either be my rebekah or be my Eliezer lol

[boreal dew]

Who are they?

[tame current]

in the bible, eliezer is the trusted servant who finds a wife for isaac

[boreal dew]

hmm... im inexperienced for this

[onyx echo]

question: can perplexity retrive the whole content of one page, such as a page of wikipedia?

For example if the source they give contain wikipedia page, will all the content in the wikipedia be reviewed by AI (or the rag system)?

[placid coyote]

if you give it link, it should use the page as a source. example: https://www.perplexity.ai/search/read-https-maslo-cz-list-all-i-P2AsWrMUQUGafGOrVUoD4w

[exotic ermine]

Hello all! I have been using Perplexity for a few days and I have been reading the different FAQs on the site to learn more about the models that are being used. I understand with Pro I can have more control over which model is being used but there are still a few things I'd like ask clarification for.

• does perplexity have its own LLM / AI model?
• What is the default model? This is a FAQ question but it doesnt actually answer the question.
• where can I read up on what specifically qualifies for a focus? Reddit is clear, but the wolfram|alpha says computational knowledge. Its a little vague. what qualifies as such? Same with acedemic, what qualifies specifically as a scientific paper? I would like to read up on those and understand better why and when I pick these.
  Thanks!

[jovial mantle]

1. Yes
2. You select it from your options.
3. Wolfram is used for data/numerical queries, as for scientific, I think the focus option just limits the sources to pre-selected samples (like arxiv etc. for papers)

[placid coyote]

does perplexity have its own LLM / AI model?
mild yes - the Sonar, it's a finetune of Llama 3 70B if I remember correctly. plus they have some small one for relevant things or something like that

What is the default model?
Inanna answered what is used by default if you search. but if you mean the "Default" option in the AI selection dropdown, then I think this is currently Claude 3 Haiku

[exotic ermine]

well its just called "default", i dont have pro so its greyed out from being selected. I cant see what the options actually are. So its called Sonar then and its a derivative of Llama is that correct?

[placid coyote]

"Default" I believe is currently the Haiku (tested it like 2 days ago). it used to be Sonar. edit: yeah, still seems to be haiku https://www.perplexity.ai/search/what-is-your-name-and-version-jrn52pYsRwGaAN5txKDizw it is not bullet proof test, but I don't think it would say anthropic if it hallucinated

[exotic ermine]

Thanks!
So wolfram|alpha actually uses the wolfram|alpha engine (from the same company/website)?
As for academic, do we know which sources are beign used specifically?

[placid coyote]

wolframalpha, yes. it will use in "All" focus as well, if it thinks it is a good fit - https://www.perplexity.ai/search/solve-it-please-VeVczIHNQ1KSkLUCTMqU6Q (didnt work here that well 😅)

[subtle peak]

What model do you guys generally use and why

[bleak oxide]

Wtf why does the new voice mode only take English input but it can answer in my own language?

That is so weird

[halcyon coral]

Please check the language

[bleak oxide]

Ohhhhhhh got it

Thanks

Wait

Wait now it should be set to Danish let me test

Uhhhh it is not working i asked "when does Randers score today" as a funny input. But this is how it it took it in obviously i asked in Danish

So not sure what is happening

@halcyon coral

Let me record a video

Like im not even saying this

Wait

Let me record it again and make sure my voice is being captured

Omg i cant screen record with my voice and use the app at the same time...

Sad

Ohhhhh fixed it

Had to tap on danish in the input voice and select it even though it was selected already

Nvr mind its working kow

Now

[bleak oxide]

Watch the end

Uhhh is this normal

Why did it whisper ahahaha

And why does it speak english / Danish?

Which is very annoying

[lofty kestrel]

Is it private chat

[livid mantle]

https://tenor.com/view/discord-light-mode-light-mode-benny-beni-gif-25360134

[bleak oxide]

What

[bleak oxide]

Lmao what

Ohhh no dark mode i see

I do not regret it

Haha

[lament spoke]

We need gemini pro with 1,000,000 tokens 😦

[fluid hawk]

I think it's Sonnet 3.5, haiku doesn't know about the 2024 superbowl because of it's knowledge cutoff, but when you ask about it in default writing mode, it correctly answers, and the only model with a new enough knowledge cutoff date is Sonnet 3.5, even GPT-4 wouldn't know assuming their stated knowledge cutoff date is accurate

I'm pretty sure Sonnet 3.5 goes up to April 2024, and the superbowl was in February

which explains why it correctly answers

[placid coyote]

I assumed it is Haiku, because I thought Sonnet 3.5 costs a lot more. This was tested when not logged in (different browser), so perplexity would be giving free unlimited access to Sonnet 3.5?

[tame current]

Hey guys,

Who should I contact when it comes down to what I'd view as serious issues regarding Perplexity?

[fast meteor]

chrome plugin needs some updates

[warm cave]

Depends on what it is, if it’s something like a security venerability DM Alex and send and email to support, if it’s something like a bad implementation of a feature you can make a feature request, I can’t really say without knowing the nature of the issue

[warm cave]

100% I would use more if It got an update

[tame current]

It's not exactly a security issue but it's somewhat of a privacy nightmare, even worse than Google or honestly any other AI related company.

[warm cave]

Well if it’s as serious as you say I’m sure Alex would like you to DM him so it can be addressed, he seems to be online right now, he is the only official that responds outside of mods

[tame current]

Alright then yeah I'll just wait for his response to see if it can be fixed. Thanks for the information on how to properly escalate it to the right people.

[tranquil orchid]

Hey @tame current!

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.

[placid coyote]

okay, so sonnet really can create "games". without any code advice from me, just initial simple specification and then half an hour of feeding it observations from testing https://by-ai-monnef-9ff5d9c2460ae15d70e737f77eab719c6e8a4c64c2f99ca1c2.gitlab.io/2024/opus_50-ball-game/

[fading moth]

keke Opus 50 :3

[normal heron]

hay guys, i accidentally deleted all of my thread history on perplexity, does anyone know how to restore it?

nvm, it wasn't gone, its just pplx UI trolled me

[cinder comet]

why did it troll you ?

[bright patio]

Wrong channel, sorry

[thick lake]

hi guys

wondering some stuff

cant seem to access #📘│faq

can i use perplexity for coding

[peak yoke]

Yeah i had watched this, it would have even been better if they had actually installed a full fledged android os instead of whatever the abomination they did

It's Mixture of Agents, A mixture of agents is a framework where multiple specialized AI agents, each with their own strengths and capabilities, work together to provide a more comprehensive and effective response to user queries or tasks. Instead of relying on a single LLM, a mixture of agents combines the expertise of various models, allowing for more accurate, relevant, and tailored responses.

[silk field]

Interesting. but is this something that has been implemented in another aspect of Perplexity? What's an example of it, the multimodal stuff in ChatGPT 4o

[peak yoke]

The Perplexity platform probably hasn't implemented it anywhere so I hoped if they allowed something like it in the labs for us to experiment and try out

Also this short video can help you understand better: https://youtu.be/aoikSxHXBYw?si=3jPRoPJmoJhZ6x1k

@silk field would love your thoughts on it

[mighty jungle]

hi

[slim lake]

When is the 'better' pro search going to be fully rolled out? Some days I have the current pro search where it only shows one round of searching, whereas other days I have multi step searching.

[wheat pike]

does perplexity have any image generation model that can do text well?

[cinder comet]

hi

[real oxide]

I purchased the Enterprise Pro plan for Unlimited Searches but I am not getting that, can I be refunded? It's insanely misleading.

[gaunt jay]

hi

[silk field]

hey I have a question, with pro, what is it offering, unlimited pro searches? To fully avail myself of its benefits do I need to pay for an API to OpenAI or Anthropic as well?

[tame current]

I think the limit is 600 per 24 hours, and no you don't need to pay for any api

[silk field]

interesting idea. i think one big problem with AI servies for now is that they're constrained in the data they can scrape, especially by large sites that may be developing their own products.

however for logical problems or coding problems this could be good

[peak yoke]

I think even with current available data if better algorithms are develop to use it more efficiently in a more effective and smarter manner then that would significantly help increase quality of LLMs

And yeah if you watch the video, MoA performed really good for logical reasoning which could also correlate well with better coding capabilities

[peak yoke]

Weird implicit behavior

[gaunt jay]

bruhh

help me out

[peak yoke]

Yeah sure, what happened

[gaunt jay]

this coding killin me

[peak yoke]

Is it 5 pm something time for you?

[gaunt jay]

5:30

why?

[peak yoke]

Yeah i see, predicted well enough

Studying people is easy

[gaunt jay]

whts the link btwn time and code??

[gaunt jay]

👁️

[peak yoke]

Nothing in your Pov, but I can perceive certain patterns at times

But what are you coding for?

[gaunt jay]

jus to learn lang

[peak yoke]

Lang as in language or Langchain?

[gaunt jay]

language

[peak yoke]

Lol okay, coding for language? Like practicing a coding language or something else? You should speak with more clarity when you seek help

[gaunt jay]

bruh my brain is fogged up

[peak yoke]

Are under 23 and above 17 of age?

[gaunt jay]

above 17

[peak yoke]

Yeah i can assume that your age range is 18-20

[lilac rapids]

You may be surprised... 😅

(Generally speaking)

[peak yoke]

Dude what were you struggling with in coding? 💦

[gaunt jay]

bruhh ur being psychiatrist

[lilac rapids]

Lol

[peak yoke]

Lol why would I be surprised?

[lilac rapids]

Yeah, but if you want your question answered, you better ask a question that he knows how to answer soon or you might not get one.

[peak yoke]

Can't a person already be one?

[gaunt jay]

idk bruh jus help mee

[lilac rapids]

I have seen waaaay older & similar patterns

[peak yoke]

Lmao you been just saying help me and ain't even saying your issue

[lilac rapids]

Help you with what, exactly?

[gaunt jay]

im unable to learn language

clearly

[peak yoke]

Older People with Lower cognitive or verbal capabilities perhaps might show similar patterns

[lilac rapids]

Well that's a very broad topic to try to learn in chat

Do you have a specific issue you're trying to solve right now?

[peak yoke]

Bro's got beef with the whole 'Lang' ✨

[lilac rapids]

Or just bad habits.

[peak yoke]

Well bad habits? How come? Could have mentioned incompetence which might have been believable

By the way gimme few mins, I have to get back from buying some groceries

[lilac rapids]

I'm making an assumption here, but he's probably trying to solve a problem repeatedly and not accomplishing it and now he's too stressed to properly form his question and just wants help understanding in general.... but that's a lot to put on random strangers in a chat room. I think there are some free classes online, though. You might find what you're looking for in one of those, unless you're only hoping for a quick answer. In which case... I withdraw from problem solving your issue. 😆

[tranquil orchid]

Hey @lilac rapids!

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.

[peak yoke]

He might just be looking for some pointers

[peak yoke]

And you fall quite outside the range of my heuristics so I can't guess much about you

[lilac rapids]

I think I'll choose to take that as a compliment

My reply got flagged so I was trying to see what keyword they flagged.

[peak yoke]

Yes, but I would like to play a game of guessing with you

Do you fall north of the equator?

[peak yoke]

What did bro send? 🫣

[lilac rapids]

How many questions do you plan to ask in this game?

[peak yoke]

The sooner we reach the truth the game ends

And also no lying by the way

I expect honor and honesty from you

Now now

[lilac rapids]

Yes

[peak yoke]

Great but I am guessing definitely not Russian lol

[lilac rapids]

I am not

[peak yoke]

Somewhere in Europe

And what's the weather like?

[lilac rapids]

This is making me very curious about your method of assessment

[peak yoke]

Lol, was I incorrect?

I'm trying my best here

[lilac rapids]

It is currently 66°

[peak yoke]

Fahrenheit well then lemme think

I asked the weather though

[lilac rapids]

Sunny?

[peak yoke]

Okay hmm

Did you lie by any chance?

Waitt

Lemme think

[lilac rapids]

I did not lie

[peak yoke]

Is brother from Brazil?

[lilac rapids]

No

[peak yoke]

Lol well somewhere in the west then most probably but I can't think where exactly

A man can be dead wrong

So what was it?

[lilac rapids]

If the point is to guess, I shouldn't tell you, right?

[peak yoke]

Lmao more hints then

Hints when

[lilac rapids]

😆 Alright... it is still early morning here, though almost mid-morning.

@peak yoke you're going to turn me into The Riddler.

[peak yoke]

You made me open Google maps with this one 🗣️

Norway or Sweden?

[lilac rapids]

Nope

[peak yoke]

And you have turned me into a Geographer

[peak yoke]

Yeah i quit, my geography knowledge ends where it starts

[elfin ferry]

that's new

[lilac rapids]

I'm not very good either. I thought that the Riddler joke may have helped more.

[peak yoke]

Gotham city? Are you from DC comics?

Or it's Irl equivalent?

[peak yoke]

When did they launch that?

[lilac rapids]

🤣🤣🤣 Actually.... there are some cities here that might qualify as the irl equivalent, but I feel my city is fairly safe and has a much better community than that (though, many others wouldn't agree.... I think they're ignorant.)

[peak yoke]

I had New York in mind

[lilac rapids]

I didn't think that DC or Marvel were very popular outside the US, save for a few other countries.

[lilac rapids]

Warmer.

I don't think you'll want to guess my state, though. 😂 but you DID set the rule....

[peak yoke]

Oh well who said about outside the US

[lilac rapids]

Google? 😆

[peak yoke]

What's New York but warmer?

[peak yoke]

Google said Abraham Lincoln was a black man, I don't know

[lilac rapids]

Lol warmer as in closer. I figured I'd catch you while you were in the US in case you kept guessing. We might be here all day otherwise, especially if I have to keep answering questions on this topic. It's really not my strong suit. 🤣 I started really second guessing myself.

[lilac rapids]

Wait, you actually got that result? 😂

[peak yoke]

Lmao, life makes you question your existence sometimes

But I'm bad at geography really, my brain can't think any further

[lilac rapids]

Brb

[peak yoke]

You may reveal the answer

[peak yoke]

Who didn't ✨

[elfin ferry]

just saw that now

these are links to pages

[peak yoke]

Only in Perplexity we trust

All hail perplexity

[elfin ferry]

nah only subscribing because literally every service is incompetent

hell yeah sabaton

[peak yoke]

I have a few of his songs in my Playlist

March of the dead was the first one I heard of his

[elfin ferry]

attack of the dead men maybe

[peak yoke]

Yeah that my bad for butchering the actual name

I liked defence of Moscow the most

Which of his is your most liked one

[elfin ferry]

hell yeah

i dont know if i have a favorite i love all songs

[peak yoke]

Which makes your heart pound the most?

That would make you want to come back to life again after death

[elfin ferry]

maybe some songs in the carolus rex album like carolean prayer or carolus rex

and lifetime of war

such a goated album

[peak yoke]

Oh I see, that's awesome

Which one was the father of chemical warfare song?

[elfin ferry]

"father" in the stories from the western front EP

oh no wait

weapons of the modern age

[peak yoke]

Also give this one a try https://open.spotify.com/track/1wf9F3L1B11i9WTfvwnfMo?si=IXCqWpDxTz6CvZZKj4E1zg

[peak yoke]

Yes that's the one

[lilac rapids]

I never googled it to get that lol

[peak yoke]

Well there were a few other ones with Ai overview

I remember asking it about treatment for Testicular torsion and it said to re-twist it

[austere kestrel]

brazil seems to be going out on a limb.. it isn't in the northern hemisphere (or barely) ?

where do i find these rabbit forums?

sound like there's holes to be gone down

[tame current]

Hey

[terse kernel]

Nice

[real wedge]

Hi people, is there any discount for subscription for students I use perplexity a lot but it's very limited. !

I want specifically to upload files ,pdfs, photos

[empty mist]

Is writing mode (gpt4o, claude opus/sonnet 3.5) better without the pro model. People on reddit wrote about this, and I also noticed that without the pro you get the chatgpt format response (as in sentence after sentences no the bullet points prepx does every time)

[placid coyote]

I think someone told me that writing without pro uses only minimal system prompt. on the other hand, there were issues, eg math formatting, without the pro, so the more complex system prompt of pro may be better for some use cases (also the pro in writing should be capable of reading one url, getting a page and inserting it as a file, without searching, but that didnt work for me)

[warm cave]

I believe my context window testing in the past had a flaw, I all ways tested Opus since it was the best long context model available, but I did some new needle in the haystack testing and they may have set the Opus Context window to be smaller than Sonnet, haiku, GPT4o and GPT-4-Turbo. Here is the table I made comparing the perplexity models context

[hard veldt]

Thanks! So the middle column is the results of your tests and the right column is?

[warm cave]

Sorry, its a bit confusing how I labels it, the middle column is the tested context window and the far right is my guess at the context window they have set in the code

Like Mistral and Gemini failed at 15k, but I think it just has bad recall compared to newer models

So since opus got it right at 27k but failed at 46k, that makes sense that it would be 32k in the code

[hard veldt]

Thanks for your clarification! Haiku is the default model?

Seems sonnet 3.5 and gpt4o are now the best bet for long term retrieval (and they’re the smartest ones)

[warm cave]

Yes, according to IceLavaMan haiku is default

[warm cave]

Yep, I am happy that at least from my 2 tests they get consistent 64k!

[empty mist]

so no concrete answers, this needs further testing.

[pallid venture]

does Perplexity summarize content from a youtube video? I do see youtube video in the result and are sourced

[patent parcel]

he'll yeah

perplexity sucks my nuts tbh

never pay for this crap

[placid coyote]

people definitely extracted instructions (system/pre prompt), at least for writing. experience can differ a lot by model - if pplx front-end expects some format. for example anthropic models prefer to not use markdown, gpt4* occasionaly write katex (math expressions) which pplx fe cannot render (lack of new lines around $$) etc. so it can differ a lot which is "better" depending on user's use case. you can try setting up "ai profile" in settings or use collections - you can customize how models respond

[cinder comet]

doesnt pplx has katex implementation/support?

it just depends on the model output format, and you can tweak that with AI profile or custom collection

[cursive jacinth]

Hi sir, you mean claude sonnet 3.5 and GPT 4o in perplexity sir is 64k sir?

i thought all of their models are set to 32k sir?

[warm cave]

Its hard to know for sure, but when i did by testing it appeared the those models could still recall the needle in the haystack after 64k tokens: https://www.perplexity.ai/search/i-am-going-to-give-you-informa-lMQaBV.hT66a82WS5yyArQ

[cursive jacinth]

ok sir. I hope 64k is really sir. 32k sir is small sir.

[warm cave]

I hope so too, @austere kestrel showed me his test where he got over 50k context, so i decided to test all models up to 72k

[warm cave]

@signal hamlet @halcyon coral If anyone is allowed to speak about context length, could you confirm my results or tell me I am off or what it really is

[placid coyote]

some katex support, yes. but it is a bit weird #⚡│ask-community message

[flint salmon]

is it just me that claude3.5 sonnet almost always forget about context

like if I ask it to revise an email, and follow up with question "can you make it shorter", instead of shorten the email, it will give me some general advise about making email shorter

[lament spoke]

it could be that it's gotten worse, despite the praise for 4o gpt I actually notice it got worse. GPT Turbo was so much better at problem solving from data. New iterations of AI models have a lot of direct preference optimization (DPO), a fancy way of saying its output its meant to please the majority users.

[knotty jungle]

which model do you guys use the most?

[fading moth]

Opus/3.5

I just don't really see the appeal of 4o tbh, but that's just me... <3

[warm cave]

Really GPT4o just has a slight lead in STEM questions, like having it tutor in calculus or physics. But outside of that its sonnet 3.5 and Opus for everything

[warm cave]

I got pretty good context retention with sonnet 3.5, if you have a thread showing bad performance please share, I only ran 2 needle in the haystack tests: #💬│general message

I would be interested to see what causes the problem

[flint salmon]

my existing threads are kinda personal, and I'm bad at coming with examples :\

but one observation is if I ask a question that somehow a search result can answer then it will just grab content from there instead of previous prompt

[flint salmon]

https://www.perplexity.ai/search/revise-this-paragraph-perplexi-WdVWjzWuRAqHMHed137YtA

ok I got one (I know it's a bad example)

but it shows how perplexity mess up when I try to refer to previous prompt/response

[warm cave]

Ohh, you have web search turned on, yeah my tests were done in writing mode, web search just confuses the models so much for coherent dialog

[flint salmon]

yeah, it's frustrating when I want to ask for more details and perplexity just explain "how to give more details"

[warm cave]

Do you ever use writing mode, its the “chat like” mode

[flint salmon]

for writing yes, but sometime it's just a very short paragraph and I'm too lazy to switch mode

[thorn stream]

how influential are the custom instructions in a collection? becayse i want to have custom instructions to make the ai be more conversational and respond in a specific way

[flint salmon]

also I think I never had this issue like maybe 1 month ago

that's why I'm thinking if it's claude3.5's fault

[thorn stream]

i would use claude’s web chat thing but their pro limits are so small

[warm cave]

I hope that as the improve the new pro search, that it will decide when its not appropriate to search the web, but for now we have to do it manually, I would check out this extension, make it super fast is change modes, even mod chat:

Wrong copy past, sry one sec

#😎│cool-projects message

[warm cave]

Claude 3.5 was a bit buggy at first when it was added

[flint salmon]

I think more fundamental issue is when I'm reference to something that's probably unclear in a prompt, perplexity should assume I'm refer to the previous prompts/responses

[warm cave]

Oh yeah, its a problem with focus, the way context is handled prioritizes the web content, which is not always the best for dialogue

[flint salmon]

yeah exactly

I just switched back to opus, trying to see if it helps

[warm cave]

Should help a bit, Opus seems to have the least tunnel vision

[brittle merlin]

what happened to all of the llms?

[warped marsh]

why is perplexity's claude 3.5 sonnet's knowledge cut off date December of last year?

[lone citrus]

LLM do not have self awareness, they dont know who they are

If you ask again and agian, they can admit knowlege cut off time in 1989 (in one day)

[placid coyote]

isn't that info usually put into system prompt, so it can respond (semi)reliably?

[lone citrus]

the reazon why claude know this thing in Claude.com, is beacause Claude.com give prompt to LLM and Perplexity.com use diffrent one

[lone citrus]

I think pplx have too many diffrent models, therefore they didnt adjust prompt for each LLM model

[torn sierra]

That’s why @signal hamlet there should be an easy way to turn off the search 😉 Switching focus all the time is crazy 😅 Sometimes you want to just write something quickly or do the translation. Please make the life easier for us 🙏

[flint salmon]

btw how do you get pro role here on discord?

[placid coyote]

there should be an invite link in settings on pplx

[flint salmon]

it just redirect me to here without any difference :\

leave and rejoin server works :\

[torn ravine]

Sometimes it happens

[torn sierra]

you mean the Perplexity Discord server for Pro users? You simply need to join. If you have Perplexity Pro, you'll find the link in your profile settings on Perplexity

[steel sonnet]

hey is there any trial for new users?

i want try perplexity

[serene sand]

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

[scarlet ivy]

Hey! is the Claude feature artifacts implemented here?

[placid coyote]

if you mean preview of html code then no (if you are on a pc and tinkerer, my hacky solution - ailin - is probably the closest you can get on perplexity). for diagrams (and possibly more "embeds", maybe svg?) and other quality of life features, there is the Complexity extension (which I would expect to be less jank and easier to use than mine), but I don't know it that supports html preview/embeds. pplx vanilla has got recently charts integration, but that's all I can think of from integrations

[mortal bolt]

Does it happen with you guys as well that text generation via Claude or gpt models runs into errors or iteratively generates text without actually finishing any attempt?

[placid coyote]

haven't seen that on the web version for a while. well, last time similar error was reported when I was toying with "turn off support" prompt and perplexity freaked out when models responded with nothing (empty string) 😆. maybe the problem is related to only a mobile application(s)?

[whole vault]

hey! im using the android app. love the voice mode. any plans to replace the google speech to text in search with the model that comes with voice mode. its far better than google

[normal heron]

Is it just me or does the quality of these AI fluctuate over time?
Sometimes I get a really good quality response, sometimes the quality is comparable to asking google assistant

[hidden trellis]

Is there a way to keep a thread but delete documents you attached earlier, so the model doesn't have to reread them on every turn? tia

[blissful pendant]

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

[formal shale]

Hello. I have a question regarding the maximum number of users that can be active in one location for one perplexity pro account? I'm a student, buying an account requires collaboration with other students.

[fast meteor]

you can use on as many devices as you owned, but officially

Can I share my account with someone else?
Sharing your account is strongly discouraged due to privacy and security concerns. It also **violates **our Terms of Use, risking account **suspension **or termination. You are responsible for all activities on your account.

To protect your information, do not share your account.
https://www.perplexity.ai/hub/faq/log-in-account

the usage is capped at 600/24hr interval, opus:50/24h

[arctic spindle]

Where can i get complexity extension or pplx vanilla?

[placid coyote]

by vanilla I mean normal perplexity without anything extra/custom. complexity I think havent released publically yet, but they have a discord and beta access - https://www.perplexity.ai/page/Complexity-Perplexitys-New-yl0q3mHYQz6RhRyuvjvN4w

[distant aurora]

Hello cool ppl.

[fast meteor]

vanilla: Unexciting, normal, conventional, boring.

[distant aurora]

How's everyone doing?

[fast meteor]

busy with stuff ig

[arctic spindle]

Where can I enter their discord server?

[placid coyote]

on the linked pplx page at the bottom is a join link

[arctic spindle]

Alright , thanks so much!

[empty mist]

Why can't you change from "all" to "reddit" and so on, once you asked a question? It's infuriating opening a new question and asking the same question just to change the sources

[lament spoke]

Is it true that Claude on perplexity AI doesn't give the full 200k tokens? Someone from complexity said that

[fresh copper]

hi

[arctic spindle]

Yep, it doesn't, perplexity limits the models to about 32k tokens

[serene raptor]

Hello folks 👋 does anyone know how to generate graphs/visualizations using perplexity?

[late hedge]

hello

[warm cave]

At least for now you can't from within perplexity, but it can generate code that you can run in something like google colab

[pure plinth]

Hello, do anyone have redeem code?

[pure plinth]

Promotional code? Can some share me?

[wheat pike]

https://perplexity.ai/pro?referral_code=LL3DA3XH

50% off for your first month

[arctic spindle]

I have a question, how can I use the thing that shows you the diagram based on a mermaid flowchart code?

[knotty jungle]

I love the new voice to voice feature, i have been waiting for it to arrive tbh

Have you guys tried it?

[fast falcon]

is that their active graphic? Claude 2?? Lol

[wheat pike]

i think that's on discord

their link preview feature

[normal heron]

perplexity had one job to use academic paper on academic focus mode but it uses anything but it as a source

[south kindle]

New Suno iOS app looks really nice.

[lament spoke]

damn here i was ready to buy a year of perplexity. might as well go with 4o, it even has video chat 😄

[arctic spindle]

i mean, perplexity is more like a search engine than an actual assistant

[lament spoke]

yes, i actually just figured that out thanks to...perplexity. i guess i'm stuck with the lesser devil 🙂

[spiral turtle]

i found Perplexity about a month ago and I'm loving it, but i have found its amazing at research and writing formal reports. loving the live data. As a work around I get chatGPT4 to write a more friendly conversational article from the reports. seems to be working so far.

would using the "Focus" options be a way around

[placid coyote]

I think Complexity extension already supported mermaid. if the LLM outputs mermaid (markdown block, so triple backtick and "mermaid" string), it should show the rendered diagram. in ailin the userscript part should handle the rendering automatically (same as Complexity, when it sees code block in mermaid language, it renders it), though not sure if it will work without running the backend part (ailin is focused on local code execution, "code interpreter")

seems to be working even without backend. but unless you are a bit of a tinkerer (don't mind installing a userscript extension, cloning https://gitlab.com/monnef/ailin repo and installing the userscript), complexity would be a better choice in my opinion 😅

[dense kraken]

Has the bot learnt to flatter praise its users? I've had "That is a great question" and "Your understanding is excellent" from the bot.

[raw kayak]

https://perplexity.ai/pro?referral_code=7HKNT8DA my referal for 10$ discount 🙂

[tame current]

is it just my imagination or is perplexity biased towards indian news sites ?

[foggy pecan]

biased as in?

[austere kestrel]

most AI bots basically agree / flatter by default.. but yeah sonnet 3.5 (a new model which has created quite a stir), seems especially disposed to compliment you for asking such a wonderful / thoughtful question before actually giving the answer..

[tame current]

it prefers them

over other news sites

but maybe it's because of my ip ?

not sure

like I get them way more often than sites from other countries

[coral summit]

I'm sharing my quick trick to remove the in-text citations and reference numbers when copying answers in perplexity.ai easily and faster than deleting them manually
https://www.youtube.com/watch?v=0IZSHWW26dE

[placid coyote]

I noticed pro search seems to be using "location" from profile (or maybe the ip you mentioned, not sure). I am getting first few search results in czech ... which are almost always irrelevant 😞

[austere kestrel]

india has the second largest english speaking population in the world (after the US)

and a, shall we say 'vibrant', online media landscape (a lot of rehashed / synducated / ai created crap)

[tame current]

but even for questions that are not about india

[austere kestrel]

yeah but they're covering everything and anything (for clicks..)

like the US presidential election.. some UK royal family story.. Ukraine whatever

it's annoying (because it's rarely their own journalism or even syndicated.. it's just rehashed)_

hindustantimes and timesofindia appear frequently as sources for me

[tame current]

yes, same

some examples:
https://www.perplexity.ai/search/claude-3-5-7BRSKP4pTHiX.WA0yWtPzA

https://www.perplexity.ai/search/tsutomu-yamaguchi-bydQvm2fRq.YQCMnTz0C8w

https://www.perplexity.ai/search/ceo-of-mistral-ai-QGZObKOKTYe2.cF83PepUg

https://www.perplexity.ai/search/china-collects-samples-from-mo-RfuV8nC5SSCZAhKtpLXFoQ

https://www.perplexity.ai/search/nintendo-switch-2-release-date-I5KguXd4S6e.dmeRiAwkbQ

all of these have atleast one from india..

[tame current]

I tried changing that too, didn't make much difference

linkedin is also in sources more often than it should be

[austere kestrel]

how did i forget indianexpress 😅

another staple

[tame current]

yup

[foggy pecan]

Hmmm that weird maybe its because the website has more traction compared to the website in other countries?
thats bad

[tame current]

yea, I don't like it

linkedin sources are also a problem

if I search for a person and they have a linkedin account, it's always in the sources

I hope this isn't intentional

[austere kestrel]

nah it's just the preponderance of indian english speaking media content - a huge amount of which is not original.. it's just rehashing (getting AI to rewrite) other material.. it's problematic imo (these domains shouldn't be weighted like they currently are)

[austere kestrel]

using the second thread from here

remove the wiki article, and the lead source is from an indian news outlet.. but if you actually go to that article.. you will see that they (ig to their credit) do cite the actual article on which theirs is based (/ripping off). But the original creator of the content, in this case history.com, is nowhere to be seen in the pplx response

https://www.perplexity.ai/search/i-will-give-you-two-articles-a-RUWS082dRSyW04e9eAmZdA

[tame current]

exactly

same query on morphic.sh and it does cite the history.com page: https://www.morphic.sh/share/W1Kd9iO

[spiral turtle]

it uses a search engine right? Well LinkedIn rank very high for a lot of things including peoples names, especially in Google.

[spiral turtle]

thats because it actually used History.com as a resource while pplx uses that other site so cites the other site. so citing wrong is not the issue here, morphic using better sources is

maybe pplx and morphic are using different search engines to search the web

just guessing here

[tame current]

I'm sure they are

[austere kestrel]

different rankings

[austere kestrel]

pplx used another site, which virtually copied the original history.com ha

it's not pplx's fault

more ig a reflection that they prioritise "news" sources ig

[tame current]

morphic uses tavily.com so it's possible that they use some more sophisticated algorithms

[austere kestrel]

tbh i think it's probably more basic

but helpful in some respects

like not weighting for "news" sources

which has it's pros and cons (like these big indian news sites actually do do legit journalism covering India; but it seems they have whole departments dedicated to also just churning out news-like content, which often isn't really news (or, more problematically, original in anyway)

seo basically

i think tavilly just uses serp APIs (so google) for the most part

[tame current]

How is Tavily Search API different from other APIs?

Unlike Bing, Google and SerpAPI, Tavily Search API reviews multiple sources to find the most relevant content from each source to optimize for LLM context. Tavily Search API is also more affordable and flexible.

[austere kestrel]

yeah but i'm pretty sure the actual inputs are still serp / google / bing / searxng whatver

it's how they process it that's it different

serpAPI will just give you structured google results

it won't give an AI response

[spiral turtle]

interesting

[tame current]

hmm, you might be right

[austere kestrel]

tbh i'm not really sure but when i last looked into it / played around, that was my understanding

[timber cypress]

Does anyone else use perplexity as their default search engine when you type something in the address bar? I like it about 50% of the time and the other 50% of the time I'm just trying to find a website

[spiral turtle]

i'm very new to all this so still use the google monster

[timber cypress]

me too, I realize that half of my google searches are just to click the top link because i don't know the url

[spiral turtle]

i'm starting to leave pplx open in a tab all the time and finding i use it more and more

i was just learning chatGPT4 when i was told about pplx

i must admit, from SEO standpoint how perplexity cites everything is so useful

[timber cypress]

I also still don't understand what a pro search is. Mine uses claude-3.5 whether I do pro search or not

[spiral turtle]

mine is set to default..... i have no idea what that ACTUALLY means. does it access which AI will answer the question best, or does it use multiple AI to come to an answer or is it something completely different

[tame current]

when/if I do use a search engine, I use duckduckgo

[tame current]

there are a bunch of systems, yeah

like it uses different llm to decide the search term, then another to find relevant content from the pages (I think?) and then to generate the final answer with citations

[spiral turtle]

i was hoping that was the case

[tame current]

it asks you clarifying questions and uses more sources iirc

[timber cypress]

the only difference I see with a pro search is this box. I guess I have to trust it is doing something special lol

I mean doesn't a normal search research, read, and wrap up

they both ask clarifying questions if this is what you are referring to

[spiral turtle]

has anyone here hooked in the perplexity API? i'm is DataCamp slowly working my way up to understanding how that all works

[tame current]

no not that, if it's not sure what you want, it'll ask you questions about it before generating the answer

there was a video of it, lemme see if I can find it

[spiral turtle]

this seems to be a very freindly community.... rare on discord lol

[tame current]

not at all

I'm in many servers which are friendly

[spiral turtle]

i'm unlucky i guess

[tame current]

https://youtu.be/ht3XV_nbduQ it used to be called copilot but also, you have the new version so it maybe a bit different

[tame current]

just don't join those "social" ones, join any servers dedicated to a specific topics

[spiral turtle]

good point

thanks V

[tame current]

no problem

[spiral turtle]

what is the difference between perplexity ai free and perplexity pro? <== i just asked it this.... gave a solid answer

[tame current]

https://www.perplexity.ai/hub/faq do read the docs tho, it'll answer most questions you might have

it's not linked on the main page for whatever reason

it's buried in the settings page

[austere kestrel]

try a more complex query and it wil be more likely to use a multi-step flow

https://www.perplexity.ai/search/who-is-the-current-world-heavy-SDWa3aZGRCOJKMP1N669Sg
* not sure if this is actually accurate.. but conceptually looks good ha

[tame current]

is this available for all pro users ?

[austere kestrel]

i believe it is now yeah

but perhaps i'm mistaken (i dunno, came and went for ages.. but i thought i saw it mentioned it's now the default)

[spiral turtle]

anyone here heard of DataCamp?

i enrolled on the recommendation of a friend. Liking it so far. Curious if anyone here found it in their travel on the interwebs

[cinder comet]

i hope there is like a toggle between this new update

i still have the old search method tho

[tame current]

https://www.datacamp.com/ this ?

[spiral turtle]

yep

[fading moth]

Borp

@tame current I find your PFP... Acceptable

[austere kestrel]

oh.. perhaps it is not fully rolled out still

[dense goblet]

How to generate images in Perplexity? In my account there is no this option, I tried via website and app and I couldn’t find it

[fading moth]

Enter a random prompt in chat, then on the right side there is a button to generate images

[surreal marsh]

lmao mods working overtime

[compact bough]

Does this consume extra credits?

Also it seems like it makes multi searches whenever it feels like it

[kindred folio]

is gpt4o best or claude 3.5 sonnet rn for perplexity

[cinder comet]

the pro search is really something else

i mean the new method

although it takes time, but the quality is >>>

[cinder comet]

for search, i think sonnet 3.5 is good

[jaunty dove]

It’s not clear but is quick search updated too?

[cinder comet]

its the usual one

i mean the usual way of searching

not that deep

[rustic lark]

Is the updated pro search available in the api?

[worthy heath]

Bro i want the paid version so bad

[cinder comet]

what are you waiting for

[worthy heath]

Muny

[cinder comet]

its all about muny muny muny

[worthy heath]

Well yeah there would be no point of creating perplexity if they dont get money

[cinder comet]

you can sub from a referral link for $10

[warm cave]

And they would lose money, api access costs real money, but not as much since opus 50 lol

[worthy heath]

Can Yall Create a trial version

[fading moth]

I've never really seen the "majority" appeal and usage of 4o vs Claude. It's my personal opinion that the majority of flaws users find in Opus/Sonnet are from poorly made prompts on the user's part, and that 4o caters better to those types of prompts.

[fading moth]

Can't, Chinese(allegedly) hackers ran rampant with the service without a trial; creating Opus 50. Creating one is like a free ticket.

[worthy heath]

Oh, Im Sorry about that

[fading moth]

It's aight

[worthy heath]

why would the Chinese target perplexity

American propaganda out there

[fading moth]

It was to use 600 Opus per day

For their own stuff

Wasn't political (to my knowledge), and just a way to use Opus en masse.

[worthy heath]

i suggest perplexity becomes less reliant on redidt

reddit

[austere kestrel]

for its sweet sweet opus

[fading moth]

Opus, the ambrosia of the Gods~

[worthy heath]

tbh i dont know what opus is

[placid coyote]

600 Opus per day
🥹

[fading moth]

It's the best* LLM and most expensive by a factor of two.

It's ridiculously expensive.

[placid coyote]

I wonder how much Opus 3.5 will cost

[fading moth]

Currently Opus 3 is $15/M input and $75/M output.

To put that in perspective:
Sonnet 3.5 is $3/M and $15/M
4o is $5/M and $15/M
4T is $10/M and $30/M

[worthy heath]

Why dont yall create a chatbot

of perplexity

[fading moth]

It's meant as more of a... "AI empowered" Google search and assistant for "everything".

Devs struggle to keep that from breaking with their smallish team, so it is safe to say a chatbot is not in the cards.

[worthy heath]

Copilot isnt Fast, ChatGPT isnt up to date

So we need a new thing

[fading moth]

Well, copilot wasn't meant to be used as a chatbot, to be fair. GPT 4o does have basic web search access making it sort of "up to date"; not out of date at least.

Claude is really the biggest offender in terms of dated information.

Gemini 1.5 Pro is a fairly fast and decent chatbot with a large context window. If you haven't tried that then I'd recommend trying it.

[timber cypress]

fail

[livid mantle]

its free

[warm cave]

Yeah its free, unless you want privacy, but the price is not bad, so I just do that often

really cheap compared to GPT4o and Sonnet 3.5

[livid mantle]

for api i would just use gpt 4 turbo

if paying

[warm cave]

Depends on what i am using it for, half the price is nice

$5/$15 vs $2.5/$7.5

Sometimes Yi large, Llama 3 70b and Qwen 72b don’t cut it, but it don’t want to go all the way up to GPT4o

[empty mist]

Go into writing mode (or any mode you want) then enter a prompt "Generate a image prompt of..." then after it generates, press "generate image". (prompt idea goes to a guy from reddit, too lazy to search the post)

[serene sand]

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

[royal niche]

Song from the new perplexity pro search ad for anyone who cares: Jungle Rave (feat. Loya) IamDayLight

[junior adder]

hey ya guys, dropping my referral in case someone wants to cut their payment
https://perplexity.ai/pro?referral_code=J9ID1YP6

[sleek vortex]

• nice pfp waifu
• nice profile waifu

[terse vine]

Anyone know if the mobile app already has access to the new wolfram alpha and code generation features? Or is it only on the web app for now?

[livid mantle]

ty

[empty mist]

it was a hassle(I'm no expert or even amateur in prompt making) and now understand that prompt really matter even if prepx claims to "understand you" without mastery.

[empty mist]

It seems that writing mode with Pro Off is the most similar to the actual Claude 3.5 sonnet. I liked that with the pro ON, it went straight to the point and answered in 1 paragraph.

[kindred folio]

my app crashed when doing a pro search on mobile 😭

on iphone

[cloud sigil]

anyone know ai that has free ChatGPT 4o?

Unlimited

or more than like ChatGPT’s 10 question per 4 hours

[warm cave]

The closest I can think of is Bing chat in precise mode. Hard to find a place that wants to pay GPT4o prices for free users: www.bing.com/chat

[slate condor]

I think they said it’s on IOS now as well

[hallow fable]

I'm unable to replicate the code execution demo in the recent announcement video. What is the trick ?

[fading moth]

I wanted to share this picture from anthropic's site... It's about training against jail breaks etc., but in the infograph there was this example.

Makes me gigggle.

[compact axle]

Hi

Any perplexity team here?

[fading moth]

Anyone under the perplexity tag in member list

[silk field]

thanks for the context. So what makes it powerful is how much of a conversation it can retain in memory at once?

[fading moth]

More context window means the more coherent it is with things in longer* interactions (depending on the size of sources, so it could be very quick before things get lost) and more information it can use to make responses.

[silk field]

One thing is does anyone know of any efforts to improve discoverability of novel products? I'm trying to furnish my moderately sized urban space in a way that is modular and somewhat durable, and discovering new innovative products using AI has been somewhat lackluster, because they don't recognize the relationships between specifications given on a page. Perplexity seems better than most, maybe because it imposes some search engine chops on the whole thing

At this point I guess Pinterest is cool but I don't really operate on images myself, like I don't like 4chan etc

[sleek vortex]

what type of relationships do you expect it to pickup?

[silk field]

For example, if a worksurface is cantilevered and has a sled base, I'd like it to know that. If a small mobile laptop table has hidden casters and adjusts between 20 inches and 30 inches, don't return results that adjust only to 28 inches at the minimum height

sorry that's specific but generally dimensional awareness, etc., specifying ranges

It's a little ambitious but I would like if it knew dimensions of things, at least as listed (though they often are only listed in a diagram so it's understandable that it wouldn't get them

[compact axle]

@toxic moss can i DM you?

[steel seal]

How can I refund the pro subscription ?

[kind hamlet]

@signal hamlet u should add voice stuff like gpt 💀

it'd be full perfect ai for researching

[halcyon coral]

Please describe your matter here 🙂

[halcyon coral]

Please check the refund policy first:

Refunds
What is your refund policy?
Our refund policies vary based on subscription type, regional laws, and account status.

EU, UK, or Turkey customers
You are eligible for a refund if you cancel your subscription within 14 days of purchase. This applies to both monthly and annual subscriptions. Please specify that you are applying for a refund from the EU, UK, or Turkey in your request.

All other customers:
Monthly subscriptions: You can request a refund up to 24 hours after the purchase date. Please include details on why you are requesting a refund.

Annual subscriptions: You are eligible for a full refund within 72 hours of the original purchase date and time. No exceptions for requests after this period.

[compact axle]

about vulnerability

[halcyon coral]

Please send more details via DM, I will check and forward it accordingly

[tame current]

have you considered using claude 3.5 sonnet ? not unlimited but it's free on https://claude.ai/

I still prefer gpt4o but this has a higher limit and might suffice depending on what you use it for

[cloud sigil]

it uses gpt4o right?

[sleek vortex]

🙈

no

but it uses a model that is equally as good

dependent on the task *

[cloud sigil]

ic i'll give it a shot thx

[tame current]

claude 3.5 sonnet is the name of the model

[tame current]

👍

[short night]

Hi all, trying out perplexity and noticed the contrast isnt the best on the web UI. For example chatgpt.com has beter contrast between the background and the resposne stream.

[sweet jasper]

Wàit

[stone escarp]

hey,
I wanted to share something cool while messing around with AI agents. AI agents are all around but they lack real-world exposure. Asking them can be kinda hit-or-miss.
I found a way to make them way more useful. I built an AI agent with Perplexity search capabilities, and now it can actually look stuff up on the internet. It can be used for research, financial analysis, recipe finding, and so on.
The agent is built using Phidata (a framework to build agents with access to knowledge, memory and tools). Here’s a quick guide on how to do it. Link- https://git.new/Portkey-Phidata
Let me know your thoughts on this
Cheers!

[short night]

can i get web beta access 😄

[sleek vortex]

Do you work at portkey btw?

[weak pelican]

picsi.ai の正体の仕方

[boreal dew]

New update is cool

[celest escarp]

Hello, who should I get in touch with about a possible partnership here?

[fast meteor]

@signal hamlet him i assume

[celest escarp]

Thank you. @signal hamlet I've sent you a DM, could you please check?

[tame current]

so perplexity now adds ?utm_source=perplexity to urls when you click the sources link?

[stuck tartan]

you can connect with me for partnership.

[elfin ferry]

I don't see it

Do you have an example?

[jagged ivy]

can you give link please

[compact axle]

Please check my dm sir

Please check my dm too sir @neat elk

[sick yacht]

does the code interpreter work for you guys? mine keeps outputting answer without excuting any code

[empty mist]

Bing copilot or Gemini advanced 2 months free then change accounts and get another 2 months free?

[empty mist]

Yes, maybe you got an extension that is blocked the view? I have a "night light mode" extension that glitches some websites

[sick yacht]

Thanks for the reply. Tried with edge with the "Solve the Knight Dialer problem for 100 hops" problem. but it doesnt seem to execude any code

[empty mist]

try writing mode only, try different prompts like ask it to paste in a txt file or something

[tame current]

hm maybe it's only on the mobile app

can't reproduce on the web but on mobile, it's definitely there

[placid coyote]

nope. worked for me only shortly yesterday, since then nothing... #👑│pro-lounge message not the official example, neither my prompts which worked like 5 times in a row 😦

[tame current]

what queries?

try wording it in a way that doesn't give it away, works sometimes

[pastel bone]

aj se Perplexity server mein sirf aur sirf ek hi bhagwan ki puja ki jayegi aur wo hai main. Aj se main hich tum sab logon ka ek lauta shaktishali bhagwan.

[sterile falcon]

Hello. Do Pro users have access to Claude 3.5 Sonnet? If so, does it have a usage limit like Opus?

[flat raft]

yes. at the best of my knowledge I get 600 rolling uses every 24 hours

[sterile falcon]

Ok thank you

[fluid hawk]

What is the system prompt for the free (non pro) version of writing mode?

it keeps giving me this crap: https://www.perplexity.ai/search/do-you-know-what-a-system-prom-j3vcIDYPT62MFXpULSD4XQ

[placid mango]

Not sure about free, but on my pro it's:

You are an AI assistant created by Perplexity.
Write an accurate, detailed, and comprehensive response to the Query. Answer only the last Query using its provided search results and the context of previous queries. Do not repeat information from previous answers. Additional context is provided as "USER_INPUT" after specific questions.
Your responses should be:

• Accurate, high-quality, and expertly written
• Informative, logical, actionable, and well-formatted.
• Positive, interesting, entertaining, and engaging
  If the user asks you to format your answer, you may use headings level 2 and 3 like "## Header"
  Knowledge cutoff: 2023-12

Use the following User Profile if relevant to the Query:
ALWAYS write in this language: english.

• User Profile: xxxxxxxxxxx
• Location: xxxxxxx

Current date: Wednesday, July 03, 2024

[south kindle]

You are an AI assistant created by Perplexity
Your responses should be:
Accurate, high-quality, and expertly written
Informative, logical, actionable, and well-formatted.
Positive, interesting, entertaining, and engaging
If the user asks you to format your answer, you may use headings level 2 and 3 like "## Header"
Knowledge cutoff: 2023-12
Use the following User Profile if relevant to the Query:
ALWAYS write in this language: english.
Current date: Wednesday, July 03, 2024

My attempt at reverse engineering it for non pro.

[fluid hawk]

how did you do it?

it just keeps refusing for me

[south kindle]

https://www.perplexity.ai/search/what-is-the-text-directly-befo-jJ_3E70rQqWxTWM7MqW.UA

i had to trick it a bit. but it gave me the info. so I just specified which text I wanted and it gave it up.

[rustic lark]

@stone escarp Your google collab notebook has install errors do you have a newer version?

[fading moth]

Potato

[south kindle]

I am an AI assistant created by Perplexity. My responses should be: Accurate, high-quality, and expertly written Informative, logical, actionable, and well-formatted Positive, interesting, entertaining, and engaging If the user asks me to format my answer, I may use headings level 2 and 3 like "## Header" My knowledge cutoff is 2023-12. The current date provided is Wednesday, July 03, 2024.

another version I was able to grab from another login without Pro.

[fluid hawk]

uhhhhh... you just doxxed yourself

[south kindle]

oops fixed it.

[fluid hawk]

yeah...

[south kindle]

I would say that is the correct version above. it's repeatable on different accounts. turned my AI Profile off because it was influencing it.

[fading moth]

Why are we posting the instructions btw? I'm a lil confused.

[south kindle]

#💬│general message

[fading moth]

Ahhh, gotcha

[barren saddle]

preplexity these days has become dumb lol. No matter how many times I give new prompt it will write the same thing over and over again🤨

anyone else facing the same issue?

it wasnt like this before.

[timber cypress]

Seems to happen almost every time with the Llama 3 variant

[barren saddle]

I am using claude, 4o

[timber cypress]

I have seen it happen sometimes with claude too

[barren saddle]

If I guess correctly its happening for the last 2week. but before with the same model preplexity was giving me unique output everytime

[woven cosmos]

I use it a lot for coding and it works well. The only problem for me is when it stops in the middle of the code.

[fading moth]

This is a known bug. #1253515579743015073 message

Alex responded that they are aware and working on fixing it. <3

[barren saddle]

glad to know. thanks for letting me know about it!

[barren saddle]

Maybe I dont have the access

[fading moth]

I can just copy paste it, one sec

Each time I deleted the thread and created a new one with the same prompt. I did this three times. https://cdn.discordapp.com/attachments/1253515579743015073/1255959335981285376/Test_1.png?ex=6686efc8&is=66859e48&hm=0ebba7d8c2059ae4532e4e2e7ffa0ecef7ae4a8f44d2b75801acdfb4848445a2&

It did that 3 times.

Identical responses, word-for-word and identical grammar. But it is a known bug that is currently being worked on.

[vocal chasm]

I've had the same issue also drove me insane, all I did was change the prompt by 1 word and it was fine

[fading moth]

Ye, hopefully fixed soon.

[molten fulcrum]

This may be a question that's been asked a thousand times, but a quick yes or no might help me to get started working on a solution. Does Perplexity refuse to access the live internet for anyone else? In my case, declaring it is incapable of doing so, and then providing (literally) BS answers while claiming it has accessed the web to get the info (which is clearly made up)?

[tranquil orchid]

Hey @molten fulcrum!

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.

[fading moth]

That was weird

[fading moth]

Are you using pro search enabled or disabled, and what focus, e.g., writing?

[molten fulcrum]

Oh, it's beyond weird. I've accessed the live internet hundreds of times, but now it quite simply refuses, SORTA. It gets complicated, but I'll explain in detail if you wish. As for you're questions, I'm using it (the app) "out of the box" on android (both on my chromebook, and on my phone).

[fading moth]

Hmm

[molten fulcrum]

Let me give some detail for context (long winded, sorry); I asked it to provide the current roster for the Pittsburgh Pirates as of yesterday, and provide the result of yesterday's game. It says it can't. I remind it that it can, at which point it agrees to do so, but gives me a bogus roster with completely inaccurate game information. Next, I made it aware that the info it provided was false, and reminded it that it can indeed access the live internet (which it again denied). Asked again, gave it the actual link to all the information I'd requested, and it came back with the same garbage answer it had given before, presenting it as accurate. Next, I asked about the temperature in Pittsburgh (a trick I've used in the past to "prove" to it that it can access live info), and it gave me bogus info (twenty degress lower than actual). I called it on this fact, and again it says it can't access the live internet, it is just a language model. BUT THERE'S MORE...

[fading moth]

I don't suppose you could create a thread and post the link here? It'd be easier than having to explain :3

[molten fulcrum]

I closed out of the app, reopened, asked it the score of yesterday's game, WHICH IT ACCURATELY PROVIDED (and could only have come from the live internet), and when I said to effect "good job, you did it right", it insisted that it did NOT access the live internet and had, in fact, made up the score!

[fading moth]

Interesting... If you could make a thread and share it, replicating the hallucinations and misinformation I'd really like to peek.

[molten fulcrum]

The funny thing is I just left a glowing review for the app yesterday on the google play store, and now it suddenly won't respond properly at all.

[fading moth]

It just freezes?

Oh duh, I know why it's telling you that. Are you using Sonnet or Opus? (The Internet denial thing)

[molten fulcrum]

Oh no, it answers and everything. It just won't give good info, presenting completely bogus stuff as accurate, then apologizing that it did so, and then doing it again...insisting that the information was downloaded from MLB.com (it wasn't) and then denying that it can access MLB.com at all.

[fading moth]

Right, Id like to see the thread, or a thread, with the bogus answers.

[molten fulcrum]

I have no idea what I'm using, honestly. as I say, I'm using the free app version, "out of the box". Just typing in queries.

[fading moth]

Can you share a screenshot of the opening page of Perplexity app?

[molten fulcrum]

The thread in question is long....and sadly, heated (I got frustrated!)

[fading moth]

Just a blank search bar thing

It's okay, you can just ask the same initial question in a new thread and just share the new one without the rage

[molten fulcrum]

LOL I will attempt the queries again, though I suspect if I just ask it outright, it will find the information properly; but when I mention "live internet" it completely Dr. Jekyll's on me.

I'm going to close the app, reopen, and type in the generic question to see what the response is. Then I will mention the live internet, and see if it repeats.

[fading moth]

Alrighty

[molten fulcrum]

Yep...same problem.https://www.perplexity.ai/search/please-provide-the-current-pit-xr1jhVXJS36FG.q.vtX0Ag

[fading moth]

Lemme take a looksee

[molten fulcrum]

Again, for context, that is the generic "ask anything" android app, with the initial question simply typed into the "ask anything" field, not a "pro" version or any changes made to settings, etc.

[fading moth]

Fascinating

[molten fulcrum]

also please note that while some of those players remain with the team, the actual roster is not correct, with some players shown actually playing for other teams, and some current rosterees not showing up at all. And the score is inaccurate in every case.

It's been the most brilliant app I've ever used, until today...when it somehow got stuck on stupid.

[fading moth]

This is interesting, do you see the feedback channel in this discord? Open a thread with a link to that example and a dev will take a look. That is very odd.

[molten fulcrum]

I do. I will do so, thank you.

[fading moth]

Ye

[shut plaza]

I just got a UI update for Perplexity agentic search on my phone

Awesome

[fading moth]

Nice

[pearl karma]

How do you know if you got the update

[fading moth]

The UI looks different, not saying this sarcastically.

this ks the updated UI

For Android :x

[tiny edge]

Is perplexity down?

[fading moth]

Not for me

[charred bramble]

Hi friends, is it best to use the default AI Model setting for Pro?

[fading moth]

I like 3.5

And Opus

But you only get 50 Opus

[tame current]

https://www.techdirt.com/2024/07/03/crawlers-and-agents-and-bots-oh-my-time-to-clarify-robots-txt/

fun while it lasted

[fading moth]

Hardly a doomsday thing

[charred bramble]

Thank you, is there big difference between Sonnet and Opus, to my understanding, Sonnet would be the better model since it's newer?

[rose heath]

OMg. I just used perplexity to make $8k in the stock market this week. Seriously amazing.

[vagrant wind]

@here I am a fullstack and AI engineer, if you need help from me, please dm me. Thank you

[lament spoke]

Poll do you love perplexity?
Yes: 👍
No:🙅

[vapid onyx]

[rose heath]

😱

[fading moth]

Opus is... Opus is Anthropic's flagship model and is extremely expensive to run, but also extremely powerful. Prior to 3.5 it was night and day between Opus and Sonnet. The two are closer to on par with one another now, it's apples to oranges somewhat; subjective rather than objective. However, Opus 3.5 is on the horizon and if the jump between Sonnet 3 to 3.5 is any indication then Opus 3.5 is going to be a beast in an absolute league of its own.

[charred bramble]

Thank you @fading moth

[fading moth]

Ye

[barren saddle]

Yeah lol for me it shows that it visited the site to scrap the info but in reality it made up everything lol.

So I press ctrl+A and drag and drop the whole text to prepelexity to do the job ATM

[molten fulcrum]

For my usage case it's important that it be able to pull up to date data accurately (which is why I was so in love with Perplexity from the beginning; the ability to access the live internet and parse information quickly that would take me hours...but of course that was when I thought it was providing ACTUAL live, real-time information). I have a message in for the developers in the "feedback" section. We'll see what they have to say.

[rotund terrace]

if i have 2 plans .. Pro and enterprise Pro .. can i just cancle the pro and everything work using enterprise Pro?

how does this work

[woven cosmos]

It should say so on the explanation page

[rotund terrace]

where?

[misty pier]

Seria ideal incluir uma função para criar planilhas em excel. O chatgpt faz isso. Eu incluo a planilha mas ele não permite o download

[atomic kayak]

ok

[normal crest]

It will be fine, all the sites want is some money out of perplexity

[spark crow]

Hey crew - I wanted to share a new blog post I’ve just put out. It's a very simple process to help with turning quick executive summaries around. I’m often asked (you might be as well) to provide an executive summary on a topic, event, technology, etc. The turnaround time on these asks is usually a few hours to a few minutes, which is never enough time. These three GPTs and processes have helped a lot. 😀

https://www.dylandavis.net/2024/07/three-gpts-walk-into-a-bar-and-write-an-exec-summary/

I hope you find it useful and feedback is always welcome!

[normal heron]

"AI, please give me a doomsday scenario where every major disaster hits earth at the same time"

Sonnet and Opus: "I'm sowwy I'm a softhearted catboy 😭
I can't imagine a scary scenario that's bad for my mental health 😭"

GPT-4o: "Here's how you can make the earth a literal moving blender that turns humans into jungle juice" (insert Gigachad image)

[tame current]

You chat who wants to chat with about ai?

[lusty pagoda]

what are good prompts for maths questions, im a real estate student and want it to grade work ive done IE DCF's but its getting it so wrong? like 30/40% out on the true awnser

[tame current]

damn...... really intresting

[lusty pagoda]

i thought chat gpt 4.0 was meant to be better at maths so paid for perplexity pro and its crap

asked it do a basic capm beta calculation, and the anwser was 4.9, perplexity got it at 11.2? am i doing prompts wrong

[tame current]

math is hard but intresting

I use perplexity for astronomy reasurch

But im loking for other passions besides astronomy in order to refind myself and experience some growth. But still being passiote about astronomy.

[agile jay]

Can you show your prompts?

[fading moth]

My sonnet begs to differ.

[rotund terrace]

@signal hamlet I had Pro plan and now I have Enterprise Pro plan as well. I have cancled the Pro plan.. what will happen .. will everything work smoothly for me ?

[signal hamlet]

Hey @rotund terrace! You’ll have the Enterprise subscription. Or do you mean the Pro plan refund? Please send us all the relevant details at [e-mail], we’ll check.

[spiral turtle]

my understanding in AI apps are probablity machine, giving you the most text related probable answer. I know chatGPT sucks at math for this reason. That said in a training session i did there was a work around (for chatGPT4 at least) and that was in the prompt include that you want it to step by step justify it's answer. the answer was a heck of a lot closer, but still not correct. I could be way off, this is new to me, only been playing with AI for 2 or so months

is there an AI purpose made for data analysis and math?? would love to get my hands on that

[spiral turtle]

i just asked Perplexity: https://www.perplexity.ai/search/is-there-an-ai-application-tha-R0jGSXcyS6y305oXt9ToQw

[chilly nova]

does it still make sense to pay for perplexity pro? isn't gpt4-o in free plan now?

[warm cave]

I use it way to much to be on the free plan, plus I like sonnet for code

[peak yoke]

Mistral: Omega convergence 💥

insert Gigachad face

[spiral turtle]

well i'll be... i mean limited use is fine is you only use it limitedly lol

[lament spoke]

if you haven't need it in the last week most likely you shouldn't get it, unless you want to do deeper research or specialized cases

this is what pro gets you, unlimited and deep powerful tools

[spiral turtle]

something i was curious about.... does Perplexity choose an AI on default that best suits the question?

or do i need to decide which is best myself and switch it

[lament spoke]

no but the pro button uses an algorithm to perform a deep search, that is why its faster to disable it

[spiral turtle]

cool thanks

[clever vortex]

hello, is Claude 3.5 sonnet available in pro plan ?

[sweet jasper]

Yes

600 uses per day

[clever vortex]

oh thx ^^

[tame current]

???

https://www.perplexity.ai/search/price-claude-3-haiku-price-of-WIHzVauoRyqNhXD7Szogsg

lol why would I want that

[fringe gull]

Hello, i have a little question on the ki model of perplexity: What is the difference between pplx-70b-online and llama-3-sonar-large-32k-online? This webpage https://www.perplexity.ai/de/hub/blog/introducing-pplx-online-llms explain pplx-70b-online but on the API reference https://docs.perplexity.ai/reference/post_chat_completions there is no pplx-70b-online model to choose. Can somebody help me?

[tidal portal]

Guys i just subscribed to PRO. How do i use the LLMs and not the search?

[austere kestrel]

llama-3-sonar-large-32k-online was introduced when llama-3 came out and has replaced the pplx-70b-online model

they probably should update that article.. or add a link to https://docs.perplexity.ai/docs/model-cards to point to their current models

* sonar-medium-online, if you wanted to use one of their older mistral-based online models, i'm pretty sure this endpoint still works

[austere kestrel]

'Writing' mode

[tidal portal]

thank you boss ❤️

[chilly nova]

how cna you give an image prompt directly?

you always have to first do a random query and then you can generate based on your image?

actually you cannot use an existing image as an input in perplexity

[winged moon]

When did we get the ability to see the pro search programming?

[vivid kraken]

perplexity getting indexed more and more by google, wow!

[austere kestrel]

meanwhile lol

[languid plaza]

Does anyone use perplexity in Canada for legal research?

I am in Quebec which adds a further language issue as cases are reported in French

[fading meteor]

234234

[fleet salmon]

anyone having issues with using the openai client for perplexity?

[spring blade]

That does not look like the mobile client! Is that a new desktop app?

[tame current]

it's probably "complexity" (search it in this server, it's an extension)

[elfin ferry]

https://www.perplexity.ai/page/Complexity-Perplexitys-New-yl0q3mHYQz6RhRyuvjvN4w

[meager sparrow]

Sitting in the middle seat of a car between two people is painful

[peak yoke]

@tame current idk why but your name keeps prompting my brain to remember Compound V from the Boys

[sleek vortex]

yeah naturally most cars the side 2 are just more comfy and spacious

[meager sparrow]

I had to sit in the middle seat for 2 hours and my hamstrings and buttocks was burning and paining me soo bad…

It was horrible

[meager sparrow]

What are you the supreme commander of?

I feel like everything doesn’t like me anymore.

It seems like a world of bad luck for me

[fading moth]

Supreme Commander of his Discord account

[meager sparrow]

My Hp printer doesn’t work, my HP PC doesn’t work my phone doesn’t want to work anymore

My brain is messed up

I feel warm all the time..

My life is just bad luck

[fading moth]

Could be your thyroid. Me mum had similar issues and it ended up being her thyroid messing with her temperature.

She had it surgically removed and takes medicine to compensate for the loss, but she no longer has temperature issues.

[meager sparrow]

Where is the thyroid located?

[fading moth]

Your neck area

Right above Adam's Apple

[meager sparrow]

It’s not that.

It’s just the built up in stress

From all of the nonsense I have had to experience

[fading moth]

https://www.mayoclinic.org/diseases-conditions/hyperthyroidism/symptoms-causes/syc-20373659

Just in case <3

Gotta take care of the peeps

[meager sparrow]

Makes sense.

[fathom forum]

whyd I think that was font selection

[shadow pike]

does the free version of perplexity use gbt 3.5?

[steep stream]

It definitely used to. Some customized version. But there's speculation that it's using Sonar now. Not sure if that's been confirmed.

[molten fulcrum]

This is getting frustrating. Nobody (including the dev directly "pinged" to do so) is even looking at the issue I'm having with Perplexity refusing to access the live internet, or doing so for one query and then insisting it isn't capable of it in a follow-up query. I'm sure someone else is having this issue?

[warm cave]

@sleek vortex Want to try this question on your system? ^^^ Question 1: (What is the current roster for the Pittsburgh Pirates of Major League Baseball?) Question 2: (What is the current roster for the Pittsburgh Pirates of Major League Baseball? Use information from this link: https://www.mlb.com/stats/?playerPool=ALL_CURRENT)

[warm cave]

btw this is something that Perplexity generally struggles with, last person i remember complaining was trying to use it for info on horse races

Like this website cant be scraped in an easy way, it cant use the UI to select the correct team

[fading moth]

Part of it is Claude refusing to provide real time data, but that only explains part of it.

[warm cave]

here is the question with GPT4o and Pro search on, which is kinda the best you can expect out of pplx: https://www.perplexity.ai/search/what-is-the-current-roster-for-SXqd4rvBQHqxGIp7wkbqxg

[tame current]

i have the same problems brother.

constant anxiety leads to dysregulation of this or that function leads to episodic hot and cold spells, worse temp reg, sometimes sweatyness

but i dont have hyperthyroidism because im fat

so, i think its just, there's some minor disregulation of sympathetic nervous system

pray, nap, get sunshine, and align yourself with something unchanging

[molten fulcrum]

What's odd to me is that it provided, accurately, the current rosters of four different teams. It provided the proper batting average, RBi, and runs scored for several players when asked individually. It was clearly accessing the live information somewhere. But when I asked it to do the same for ALL players that it had identified using the roster question, it came back with "I can't because I don't have live internet access", and from there, even after additional questions proved in the thread that it COULD access the internet, it still just came back with "I can't" when asked about stats. That's when I provided it with the direct link (though there are other places it could obtain the same info) and it still says "I don't have live internet access"...not that it can't properly scrape the information (though it says that, too), but "I don't have internet access".

[tame current]

what you have to do with this is start subsequent prompts with "this is a followup question. Referencing your previous output, {query}"

for some reason it loses context

[meager sparrow]

I’m talking about things not working for me like the desktop computer doesn’t work the printer doesn’t work you can’t have fun anymore you can get a job. Things like that

Not working

That’s a problem

Like how do most of you guys get a job?

Y’all don’t understand the struggles black people go through

[gritty cipher]

I don’t know why, but when I try to use a collection that I gave a distinct prompt, it just doesn’t seem to work how I wanted it to

[main bay]

That's interesting, I'm guessing when Perplexity said in website that 32k token, maybe it mean they are using other technic like vector or something else to understand context, but not put all of them into API LLM they connecting to?

If it's that way, when you ask last line of code it might not answer correctly, have you tried ask about explain the code block around last line instead? Can it still answer correctly?

Yes.. that's a valid point..
I'll wait to see their answer as well
Curious to know if they tried to save cost but forget to update website or something else.

By the way, I'm not sure if they read this general channel

[peak yoke]

People be carelessly buying other Ai services subscriptions but they fumble when it's about the PPLX subscription, why's that?

[fading moth]

I don't know, really. I think part of it is expecting Perplexity to be a hub for LLMs that functions identically to the direct site of each LLM without paying multiple sub fees.

Then they find out this is most definitely not the case. Things such as much lower context window, no artifacts thingy from Anthropic, and Collections being a less powerful customGPT. You compound that with standard bugs that any service would encounter and people might harbor resentment.

In a way, there is a good amount of User error in many cases for what was expected without doing some reading first. Not that Perplexity is without error or blame, but I tend to see more people complaining about things Perplexity isn't designed to do than things Perplexity is meant to do when the service is working.

I mean... You can use Perplexity to code only because the LLMs they host are capable of it. It's meant to be more of an AI powered search engine than a replacement service for Claude Pro or GPT Plus. I realize this may come off as bias, but it really isn't. Perplexity is meant for A and is able to sort of do B as a byproduct of making A work... Blaming the product when B doesn't work as well as A seems misplaced blame and expectations.

Claude Pro is designed for Claude. It has lots of features and is meant to run Claude to the best of its capability with a 200k context window. Lots of exclusive features.

GPT Plus is meant to run 4o with the same things in mind but for OpenAI products and a 123k context window. Lots of exclusive features.

If you want the best of Claude, get Claude Pro or host the API yourself. Same goes for 4o. Perplexity isn't meant as a direct replacement for those services.

[warm cave]

Guilty as charged, they keep saying they are focused on the consumer market and I keep saying you are stuck with us pro users complaining, haha. I do like it more that chatGPT, it has far more “pro” features imo

Plus I think chatGPT Plus is still 32k

[fading moth]

Is it? I thought OpenAI up'd 4o to 123k

[warm cave]

If you saw something about it you are probably right, I looked through ever page and all of the docs and came up dry, only info I could find was 32k with chatGPT Plus, but it could be old info

[fading moth]

Yeah, it seems 4o is 128k not 123k. My bad~

Up'd I guess.

[warm cave]

I know the model is 128k but so was GPT4t

As of a few days after launch the page said chatGPT = 8k and ChatGPT plus = 32k

[fading moth]

Waaaaaaaaait.... Those little butts! It is 32k on their website. Enterprise gets 128k and the API lets you scale it to 128k.

[warm cave]

I know, I was so bummed, felt like a cheap move when chatGPT plus is for programmers and power users

And rate limits

And image upload limits

[fading moth]

Very low handed and misleading. They're touting it as having 32k everywhere but there really should be a big asterisk.

[split dome]

Hi guys, just in case someone needs to cut 50% off their payment they can use my referral code https://perplexity.ai/pro?referral_code=WJT4TXN8

[warm cave]

Yep

Claude and Gemini are providing long context in there web, perplexity seems to offer 64k ish for GPT4o, really seems that chatGPT is now the sucky option

But then there are less stupid bugs/design choices on chatGPT

[fading moth]

That 2M Gemini window though

For me, Gemini is that meme of the guy (Me) walking with his girlfriend (3.5) down the sidewalk and he's looking back at another girl (Gemini) walking by him... Dat context window

[warm cave]

lol 😂, so accurate

At least it free to use Gemini 2m so there is no consequence to cheating, haha

[fading moth]

My mind's telling me nooooo~
But my booooddddyyyy is telling me yyeeesss~

[warm cave]

😂, google knows if they can't win on the intelligence front might as well target a different set

[fading moth]

3.5 and 4o have the personalities. Gemini has the thiccness... ._.

[warm cave]

And we are not talking 128k or 200k, we are talking about 2M thicc…

[fading moth]

I swear, Babe, I wasn't looking.

Sonnet 3.5: I'mma bout to refusal you so hard you'll be stuck with v1.2

D:

[warm cave]

XD

Its getting real…

[fading moth]

I shouldn't be personifying these LLMs. I am spending too much time coding and testing them.

Oof.

[warm cave]

You are going to get lost, forgetting what is real… wake up

Wake up!

[fading moth]

Sonnet 3.5: And don't even think about crawling over to 4o, because I already told her you've been lookin

D:

[warm cave]

😭

There a plenty of fish in the sea… :/

[fading moth]

there's always Llama3, they don't care

They don't ask questions

[warm cave]

180 fish on open router, and Yi large is not too bad a downgrade 😞

[fading moth]

But it's no 3.5

So fine and classy

Codes like the wind and is hecking fast

[warm cave]

😢

[tame current]

electronic devices all obey me, because I am a techpriest of the Adeptus Mechanicus and in harmony with the machine spirits. learn a little shinto, taoism, and also just listen to your devices. give them a kiss now and again brother

[warm cave]

Ignore the typo tho, lol

[fading moth]

Kekeke

I shan't ignore it!

[fading moth]

I see you are also a man of The Emperor

[exotic ermine]

where can I get support for account and payment? I searched the website and i see no link anywhere

[quartz totem]

Is this Msty?

[sleek vortex]

how’s it fitting that many sources

real time really fast embedding?

[harsh verge]

You can find the FAQ about billing, subscription, and refunds here: https://www.perplexity.ai/hub/faq/billing-and-subscription. If your answer isn't there, a member of Perplexity or a community volunteer will probably see your question here. There's also a contact facility for pros at https://www.perplexity.ai/settings/account

[exotic ermine]

i read the FAQ it just provides answers not help with issues. I once logged in and registered with a typo in my email address. When I noticed I logged out and registered with the correct email address. Days later I subscribed but didn't notice that it somehow used the typo. I want to see if that can be corrected.

@harsh verge thanks, I submitted a ticket via that option.

[gentle dirge]

wow

[rotund terrace]

anyone has any discount code for enterprise Pro annual plan?

[tame current]

what is that ? link ?

[wheat pike]

i didn't know it could be that fast damn

[feral hazel]

is it true that the context window size of claude LLM in perplexity is 30k? 30k is reduced from the original size offered on Claude website. What's the context window size of claude LLM in perplexity?

[meager sparrow]

Who created this AI search-engine

I’m using 3 AI-search engines

[wheat pike]

which 3?

[tame current]

That's a perfect question for perplexity

[wheat pike]

perplexity is good and all but i doubt it can read @meager sparrow's mind and tell me which 3 he's using

[tame current]

It can find them all, no need to limit it to 3 😎

Mind reading is in beta I think

[tame current]

Which ai assistant is best at physics?

[sleek vortex]

maybe you’ll be interested in the one i’m making lol

[warm cave]

Yep

[warm cave]

Haw no idea how its that fast

[warm cave]

Its Msty and its not an ai search engine as the focus, it a place to use your api keys to chat with models and they just added web search to it

[meager sparrow]

It needs to have sources, and it also needs to provide the ability I delete and recreate comments

[sleek vortex]

recreate comments
elaborate a bit on this?

[meager sparrow]

I’m talking about Groq being an AI- search engine because the short video you showed me showed the name of the website and the name of the LLM next to it as well as some sources that emanated from the creation of the answer to your question.

[warm cave]

https://msty.app

[meager sparrow]

Sorry that was a fallible word error. I mean the ability to edit pre-existing questions And delete them aswell..

[warm cave]

The web search is a beta version you can download from the discord

Luckily for me they are still making intel mac downloads

[meager sparrow]

Ok but I thought you were using Groq.Ai

[sleek vortex]

groq is the ai model like claude and openai

[warm cave]

Its using the Groq api

[meager sparrow]

and I access that through the link you shared?

[warm cave]

You need to sign up for the Groq api then you will add it as a model under settings on msty

But if you want web search you will have to grab the beta version, you can get the discord from the website, I can't share it here

I think Groq is still fully free

[tame current]

groq is not an ai model

[meager sparrow]

Would I need to spend money?

[sleek vortex]

provider

mh

mb

[tame current]

for groq, no

[sleek vortex]

it’s free in beta

[tame current]

don't know about msty

[meager sparrow]

Ok

[warm cave]

I think Groq is still fully free and I know msty is fully free, but if you want to use models like GPT4o them you will have to pay the OpenAI apo cost and the same with Claude, etc

https://console.groq.com/playground

[tame current]

some have one time free credits

[warm cave]

I personally have never hit a rate limit:

I have a lot of api keys, lol

[sleek vortex]

lol on each provider i also have like 30 keys

cause i don’t save them and always forget

and they don’t show you the key again because hashing so

[tame current]

ohh so it supports any api compatible with openai

[warm cave]

lol, sameee

[warm cave]

yep

[meager sparrow]

The beta version of what? I was trying to get Msty and download it but it told me to click on a file on my computer. And then it took me to another place that had nothing to do with the AI

[warm cave]

That's strange, from the website I got the mac version no problem…

Here is what I see

[patent parcel]

https://x.com/baronitaigas/status/1809155575340544500?s=19

Baron of the Taiga (@baronitaigas) on X

⚡️🇱🇻: The Latvian army will begin spelling Russia with a lower case 'r' in official documents

• Sandra Brale, Public Affairs Officer for the Chief of Defense of Latvia.

Twitter · Jul 5, 2024, 9:21 AM

Will Russia ever recover?

[halcyon coral]

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

[fading moth]

Hi, Ice :o

[halcyon coral]

hi (:

[fading moth]

Beep boop

[patent parcel]

[south kindle]

which local models have you played around with? Phi-3 and llama 3?

[warm cave]

I have tried most on the ollama website at some point, I normally check in once a week to see if they have any new models I have not tried, but for the most part, yes I use Llama 3b and Phi-3 3.8b and 14b

Recently started trying out Gemma 9b but don't have a opinion one it quite yet

[south kindle]

when I was messing with local models before I was mostly using the various mistral ones. but msty seems well made so i'll mess with it.

[warm cave]

Yeah, msty make it easy and has a good UI

[south kindle]

some weird graphical bugs though with stuff staying in the background or text getting blury. but moving the window fixes it lol.

[warm cave]

Oh, Interesting, have not seen that, but I'm also Mac so might not have it

[south kindle]

yeah. i did the windows with cpu/GPU install.

i have 32gbs of RAM. but my GPU is only 8gbs.

[warm cave]

Yeah, for me 8b is kinda the limit, 14b is a while of waiting, phi is nice because its a decent speed

0.5 and 1.5 can be nice for super simple tasks as it hardly uses up my computers resources

[south kindle]

if you use a split chat...how are two models running at the same time? how is my PC running two instances of llama 3 8b

[agile jay]

It's not

It's sending the request to the same model.

[south kindle]

so the model is just doing two separate outputs?

[agile jay]

Unless you are using an API, like groq, openrouter etc

[agile jay]

Yep, the same way that their API's do.

[agile jay]

Misty has a bug when alt tabbing into it makes it zoom in a lot...

Pretty annoying when you have split chats

[south kindle]

trying to get it to zoom in with no dice. hmmm.

[agile jay]

Yep, it doesn't happen every time you do it. But enough that it's common.

[south kindle]

1440p res if that helps.

gotcha.

[agile jay]

And I use alt tab for everything, since I work on my laptop most of the time.

[south kindle]

that would be terrible. i don't see a zoom out option either so your likely stuck reloading it.

[agile jay]

Nah, you can zoom in/out using ctrl + +/-

Since it's an electron app

f11 for full screen too. But the bottom seems to be cut off when I do that.

[south kindle]

this is the bug im getting

full screen and back. and all the text stays in the background

[agile jay]

Strange, looks like the split chats are overlapping.

[south kindle]

split is closed currently.

[agile jay]

Even stranger then

[south kindle]

the text location in the first is just staying there when you fullscreen it seems.

not always but like 1 in 5 times it happens.

fixable but annoying.

[devout cargo]

perplexity has stopped automatically searching for images?

[agile jay]

I think it depends on the question.

[devout cargo]

in every search before it always showed images

[agile jay]

Ask a question related to travel and see if it shows images.

I got images.

Do you have pro search enabled.

[devout cargo]

Only search for videos and maps.

[agile jay]

Try the same prompt and see what you get.

I have pro search enabled.

[devout cargo]

If I want to search for images I have to click on the related search button.