💬│general

[digital hare]

agi

https://www.perplexity.ai/search/what-is-agi-vfsDo3z6TrG2z0oEobNzyA

[vestal socket]

Ask it what is AGI for AI technology

I have AGI for ai

[digital hare]

okay

[warm cave]

Yeah, I love perplexity, but the free version is really not smart, hence why i have pro i guess

[digital hare]

bruh what tf

[vestal socket]

I’m in swimming pool

Swimming

With AGI for ai

Free

[digital hare]

yes it needs more work

[vestal socket]

Look up AGI for ai technology

[digital hare]

but they give 5 free options to use copilot every 4 hours

[digital hare]

is that like sora

by open ai

[vestal socket]

No it’s just LLm

[digital hare]

elon sued openai

[vestal socket]

But a jailbreak prompt

Works for any platform

AGI is here

[digital hare]

so it will give premium features for free

[vestal socket]

No

[digital hare]

sad

[vestal socket]

Better

[warm cave]

So default has shorter messages, but you can have it pick up where it left off, this mostly worked: https://www.perplexity.ai/search/show-me-the-2X8fzcBARFS.B4aoKIfSVQ#1

[digital hare]

wow this was really smart

thanks alot

[warm cave]

No problem!

[vestal socket]

Who wants AGI

[fluid mauve]

@digital hare = Artifical General intelligence. its what everyone is striving for. Elon is saying in his lawsuit against OpenAI that 8* could be AGI

[warm cave]

Its the smart one that will kill humanity

[digital hare]

he was saying that he did that cause openai was breaching their agreement of humanity before profit

[digital hare]

how can an ai could kill humanity

until and unless you put it in your brain which is what elon is doing

🤣

[fluid mauve]

And Jeremiah is saying he has it from a prompt. Highly doubtful but you can learn a good lesson from it. DONT BELIEVE EVERYTHING SOME ONE SAYS ON THE INTERNET. An old CyberSecurity/Military phrase springs to mind ;Trust, But verify'

[warm cave]

Im just joking, you always hear that AI could kill humanity

IDK, maybe, seems like a stretch right now

https://www.perplexity.ai/search/Elon-musk-OpenAI-wup5WiTSRyyY9G0Oxtqp9A#0

Im trying out a new stack so it answer question the way i like

[fluid mauve]

I think there will eventually be a tipping point where Ai will have ammased the knolwedge of humanity, and then go forward to make its own decisions/breakthroughs. A bit like my friend who uses robots to make robots, its a kind of recursion thing.

Will Ai kill off humanity. pfft, who knows but right now its a tool. and a good tool, deopending on what you feed it. Your Eminem lyrics are a good example. dependant on the inputs predecates what the output will be

[digital hare]

is there any limit in premium version of perplexity like chatgpt 4 ?

like this

[warm cave]

you get at least 300 questions a day, but I have 600 per day and have never run out in a day

[fluid mauve]

look at the docs

[digital hare]

40 messages / 3hours is really less which is offered by openai

[fluid mauve]

same, never run out

[digital hare]

i always used to get outta the limit

[digital hare]

damn

that's alot

[warm cave]

Yeah, same, i have both right now, and chatGPT limits are way too low

[digital hare]

perplexity is basically a better choice than chatgpt

in terms of money

and value

but the chatgpt plugin is just deal breaker

i really like plugins

[fluid mauve]

you need to think more about what your asking. if you want to learn comp sci, dont go wasting prompts on eminem lyrics which can be returned by google. use the correct tool for the job. google for search, Perplexity for answers

[digital hare]

haha no im just testing perplexity free version

i usually dont search lyrics

[warm cave]

the only thing i miss is code interpreter, and that is why i have both, I am in college and pay for it instead of a personal math tutor, haha

[digital hare]

😂

[fluid mauve]

ii uzse codellama in the perplexity labs via api. codes well enough for me. Though i do have AI assistant in my PyCharm IDE

[digital hare]

life has become so much easier after AI

😁

[fossil sentinel]

Hmm idk if it's a fluke but I asked a random physics question that gemini ultra and perplexity seem to consistently get wrong (granted I don't have pro), but was able to get it right when I asked it the question while uploading the textbook on gemini 1.5

[fluid mauve]

did you upload the textbook to Perplexity?

[austere kestrel]

oh dear lol

[fossil sentinel]

I uploaded like... a 500k+ token textbook

[fluid mauve]

lol

everyone starts somewhere

i like them trying to muddy the waters with AGI and ASI, basicaly the same, just marketing want more buzz words

[austere kestrel]

No I meant perplexity's response, not the query \

[fluid mauve]

ah

[austere kestrel]

This also descended into gibberish 😬

[fluid mauve]

I see what you mean. Framing the question is everything

[warm cave]

Yeah, i have been experimenting with deferent options, like open interpret and autogen studios, since they both have code interpreter built in, ChatGPT is just very seamless and I probably wont switch until this semester ends, but using code llama and a code IDE is a good idea, that somthing that would just work right away

[austere kestrel]

yes agree - but also (the search engine needs to do some of the work, within reason)

[warm cave]

Really if you want less frustration you pay for a smarter AI, But if you put a little work into the questions and into looking a little online yourself it tends to fill the gap

I love playing with 7b and 13b models and they are very dumb compared to what we have in paid options but if you know how to work with them, they are realy cool

[fossil sentinel]

Something I didn't think to do when using perplexity til a few days ago was to prompt copilot to search in a certain way...

Idk why I never thought about it explicitly

[fluid mauve]

Totally. I went the easy option and got all options, Perplexity Pro, Chat GPT Pro and Google Ai Premium. Both Perplexity and Chat GPT have their own strong points. I only goy Google yesterday so will look at that during the week

typically I dont use copilot

[warm cave]

I wish they had docs on exactly how everything works in the background, at first its unclear what copilot dose

[austere kestrel]

@warm cave @fossil sentinel gotta check out this if you haven't already, it's really cool #1183781616515031124 message

[fluid mauve]

Yes, there is a lot of good information in the Prompt library

[austere kestrel]

Indeed :)) Though I had never seen this copilot manipulation before

[warm cave]

I will look at that, what is did was some system prompt mining, iIthings I have a good idea, but i could be wrong

but i dont know copilots behavior very well, i think i get how it fits in, but copilot manipulation sounds like it would yeild helpful info

[austere kestrel]

ya don't get misled by the fact it was posted in an image generation prompt thread - @nimble field has unearthed a real gem here imo

[warm cave]

Yeah, thank you, i kinda skipped that thread bc i don’t care about image gen too much.

[austere kestrel]

same lol. It was only by chance that I stumbled across - and was pretty blown away by it ha

[fossil sentinel]

Interesting

[austere kestrel]

btw @digital hare just to clarify (if it wasn't clear), this is not directed at you or meant to suggest you did something wrong (you didn't), it was the response that I'm highlighting (it shouldn't do that)

[warm cave]

Also if you want a longer system prompt than what is allowed or if you want a system prompt applied always no mater the thread, you can add it under profile, I just put it under "Introduce Yourself, share as much as you'd like" and it worked great haha

but i did put a heading saying "User System Prompt:" before i put it in

[digital hare]

yea i understood dont worry about that lol :))

[vestal socket]

When AGI

[halcyon coral]

https://www.perplexity.ai/search/When-will-we-ysf2jsvOTvWGnV.LBL9p9A

||Perplexity has ignored my spelling ||

[sweet jasper]

Yo what is this

How did you get so many questions

[sand storm]

I use Google Gemini Advanced and ChatGPT4 pro. They are both good, but in different ways

[sand storm]

I find that we have to become experts in order to understand anything at AI. Cause it has vectord, linear algebra, vectors, matrices, hyperplanes, boltzman machine, transformer architecture, neural network blablabla

I’ve read Wolframs book on GPT, AI is good for you by Eric Jang, Human Compatible by I forgot, and Master Algorithm by Pedro Domingos, and I still feel like I don’t know anything. Those bools weren’t technical, but still. I am reading Life 3.0 by Max Tegmark which is good.

My frienf recommended me Superintelligence, but it doesn’t look technical enough.

[warm cave]

what are some good go to word problems the test AI problem solving?? https://www.perplexity.ai/search/If-i-hang-KkHVySekTEqJbyK.Cl2IzQ#0

[sand storm]

The next step for me I think is to actually start building AI in order to understand it. Not just read a math textbook, or a python tutorial

[warm cave]

Yeah, i agree, its all so far above me, but i am trying to learn

[warm cave]

Good luck!

[sand storm]

We need to have more geographically proximate AI workshops

Run by volunteers or entrepreneurs

[warm cave]

That would be great!

[sand storm]

I mean there are free courses on youtube. I’ve seen them.

There’s also $24k grad dip in vic uni in australia

i wouldn’t pay for that though

[sweet jasper]

https://benhoyt.com/writings/go-1brc/

[sand storm]

So I have a question for Perplexity Pro but I don’t have it, will it answer my particular question better than the normal Perplexity. I’m tempted to just impulse buy

[halcyon coral]

You can send me the question and then you can compare the results ☺️

[sand storm]

I just got the subscription. I think I have AI addiction

I can barely afford these AI but I just have to use it

I don’t know if anyone can relate to this.

This is all I can find on Inflection AI

I was considering manually going through articles on google. But it so inefficient

[serene sand]

welcome on board! deep dive ➤#💬│general message

[sand storm]

Thanks. Wow that's a lot of information.

Good stuff

[flint jolt]

anyone know if the new Perplexity integration with the Rabbit R1 will come with api fees for the end user?

I'm also concerned this will cause delays in order fulfillment

the perplexity integration came after orders were taken for the first batches... now they have to work through the integration before the first orders ship... unless they plan to release as-is and then do a patch.

I want my R1 so bad and I'm in batch 6

[sterile mural]

hi

[fluid mauve]

Haha. Same boat. Ordered from UK. Batch 6 also.

[robust gust]

Heard they said it will be free forever to use on the rabbit. But we all know what it means

[flint jolt]

I'm sure it will be free forever. I'm also sure to unlock the full power of its capability you will need a Pro subscription. In my experience that's just how the world works 😫

[robust gust]

Yeah or we will just have the basic/ equivalent of free version

[flint jolt]

In this modern age, life is pay-to-win

[robust gust]

VC money can't live forever

[flint jolt]

On a similar note... I was curious about the API fees

so I asked Perplexity about the Rabbit R1

https://www.perplexity.ai/search/will-i-need-.b0O_mJ0TW6OY7c9F.fGcA

I hope its accurate

I'm using the Claude 2.1 model

[robust gust]

Oh nice, i was wondering which model was that

They should always show what model it is used

[flint jolt]

i do like that it cites sources

[robust gust]

Yeap, thats why it is nice

[flint jolt]

I have grown to really distrust companies like Micro$oft and Google

so Perplexity is a breath of fresh air

[robust gust]

Didnt google found them ?

[flint jolt]

good question...

lets ask 🙂

[robust gust]

Oh no, it is bezos :'

[flint jolt]

https://www.perplexity.ai/search/was-perplexity-founded-wQ8GKOUsQzKwrYU.awYc1g

[robust gust]

oh when i said funded, i meant given money

[flint jolt]

maybe

I really like having the choice for what LLM to use tho

it gives the impression that they are trying to avoid bias in the responses

[robust gust]

Yeap, i tried you.com before but their android app is just a weird fork of firefox and is half broken, compared to how clean perplexity is

[flint jolt]

the discover stuff is nice too

[robust gust]

Its nice but it is just in english, it should craete one for each language we choose in settings

[flint jolt]

give it time

they just announced a bunch of enhancements

[robust gust]

for slack ? 🫡 🤣

[austere kestrel]

I don't think google has been involved in any of the funding perplexity has received
(one of the cofounders might have been ex-google though iirc)

[flint jolt]

I am so curious about the Rabbit Hole

I wish they would open it up to pre-orders

I want to start thinking about how I'll use my R1

I have ideas... but until I see what integrations are available I just don't know

and the training mode!

I can't wait. I'm such a nerd

Perplexity seems to "think" that the portal will only be accessible to users whos devices have shipped.

so that's a bummer

but I guess we will see

[solemn trench]

[asking it an academic question to tell me all papers that use dataset Y to do X]
Answer
I'm sorry for any confusion, but I can't perform live web searches or access real-time data. My responses are based on a mixture of licensed data, data created by human trainers, and publicly available information up to my last training cut-off in September 2021. ...

[flint jolt]

try a different model maybe?

[solemn trench]

i tried them all

this is complete botching of the system prompt.

[flint jolt]

just curious...

put your question here and I'll try it to see if i get the same answer

[sleek tapir]

You need to use a new query, in case you're just continuing off the same one

I find it gets stuck a lot of if you try to follow up too many times on the same query

[flint jolt]

you mean start a new thread?

[sleek tapir]

yeah

[rapid cape]

I am trying to understand the high-level flow of how perplexity works.

How does Perplexity get the source documents that is used to summarize the answer ? Did perplexity also crawl the entire internet ?

If this is discussed elsewhere pls point to it

[solemn trench]

tell me about all the papers that use reveli olabs to study diversity

interestingly, this is only when i do the "Academic" serach

the regular search does better

[flint jolt]

interesting

[austere kestrel]

Would recommend this article, doesn't cover it in depth but provides some insight https://spectrum.ieee.org/perplexity-ai
(But yes, they're crawling the web https://docs.perplexity.ai/docs/perplexitybot ; whether their proprietary index is the sole source of the web results used to inform the responses I'm not sure)

[rapid cape]

Thanks. Very helpful. That they crawl the web at such scale is simply amazing. Previously i assumed that they get some web page ranking results from another search partner (Bing, DuckDuckGo etc)

The other super impressive thing is that they read those top 10, top 20 articles and summarize them quickly in to an answer. Unless the crawled webpages are pre-processed and kept with Perplexity, that would seem to be very difficult to achieve with such low latency too.

[austere kestrel]

No prob! For sure read that article (if haven't already), as it touches on some of the nuances / constraints in their approach (I mean, indexing the web is a pretty big undertaking - keeping that index fresh / up-to-date is an even bigger undertaking..)

But yep that said totally agree overall - impressive!

[rapid cape]

I read that article — doesn’t mention a lot of details though. I can understand that using BERT to aggressively summarize top web sources - which is a cool idea and fundamentally saves time during creating summaries too.

It seems restrictive though because if I go to the obscure, less frequented parts of the web, their crawler might not have that kind of information and is likely to provide an incorrect / incomplete answer.

I guess I should try to “break it” by going to topics / sources they might not have “BERT”ed/indexed

[austere kestrel]

Not just the obscure - there's also things like sport scores, weather forecasts, that are constantly changing and inherently time-sensitive. Should say that the web version has the copilot/pro toggle, it's not clear exactly what goes on in terms of sourcing, but it tends to be far more reliable (in terms of freshness, but also in terms of being able to be directed to specific urls)
If you want to test the index, try the API

[austere kestrel]

btw I hadn't noticed / appreciated that reference to BERT in the context of their approach to indexing (or I guess more specifically ranking) - thanks for highlighting :))

[outer relic]

hey guys ı wanna use the search func of perplexity in my app using vercel ai sdk anyone ever done that

[hasty wave]

can i upload multiple pdf books and chat with them at the same time using perplexity?

[cunning wing]

Is perplexity down? It says "connection reset" every time I try to load the page

[sand storm]

Wow Perplexity Pro is amazing

[cunning wing]

Oh it's working on mobile. I think it must be blocked by the network I'm on

[sand storm]

I have learnt so much about Palantir

I wonder if people will sue Perplexity because the information you can get is so serious

[sweet jasper]

Palantir is public knowledge 😳

[sand storm]

true

Perplexity needs to have an additional thing to source check

Because Linkdin, Reddit, or websites aren’t always accurate

[sand storm]

I learnt so much about Mustafa Suleyman and Reid Hoffman

I’ve been researching stuff for 5 hours lol

Its 3:04am now

Gonna binge research on Sam Altman now

and Open AI

I didn’t think I would say this. But I think Perplexity Pro is better than youtube.

Youtube is full of clickbait titles now. Perplexity has literally zero clickbait.

Its almost addictive to just constantly interact with information on your screen. I don’t know if Perlexity intentionally made it this addictive lol.

[shadow lake]

Any reason why Mistral AI option is not present on the iOS app?

[halcyon coral]

#🍏│feedback-ios message

[shadow lake]

Ok

[spark crow]

Hey crew 👋

I'm pulling together a conference in Asheville, NC for developers attempting to incorporate GenAI into production environments. Call for papers is open and we're looking for more speakers, so if you're interested please reach out.

P.S. There's plenty of beautiful nature and breweries 🏔️ 🍻 🏔️ 🍻 🏔️ 🍻 🏔️ 🍻 🏔️ 🍻

https://www.aiinproduction.com/cfp

[ripe viper]

Got the "Discover Daily Podcast by Perplexity" spam today. Why is this lie in the footer? I definitely never opted in to email spam. If presented the option, I never opt in, always opt out.

Bad marketing behaviour.

[white cipher]

Hi everyone! Recently, I have been trying to transition from Google to Perplexity, but I know very little about Perplexity... Is it worth completely transitioning away with Google to Perplexity? What are pros and cons of Perplexity? I would really appreciate your insights and tips. Thanks!

[valid yoke]

Use DuckDuckGo as a backup

But ya ditch Google

[fallen kindle]

i use google to get a phone number or an address of a place, use Perplexity for research, for anything that requires comprehensive answer. You can prompt it, add a format of output that you want, craete collections per project etc

[grand sand]

https://fxtwitter.com/ananayarora/status/1762439921825120578

this is awesome

[fluid mauve]

I'm off to steal the wifes watch

[warm cave]

How do you get it? I haven’t used my watch in a while but it still updates to watch OS 10 so I think it should work

[grand sand]

A developer made it

You need to ask them

[warm cave]

Oh, I see the link, thank you! 🙏

[grand sand]

You're welcome!

[serene sand]

#💗│sharing message ☕

[cedar leaf]

Damn. Super frustrating. I was on trial. I wanted to cancel, no response while trying to cancel. I blocked my Credit card for preventing to not to get debited for the annual subscription 200$. I had to unlock my card temporarily and the damn thing debited with in an hour time gap. I’ve raised a refund request. Super frustrating!!!!

Could somebody from Perplexity answer this?
I don’t want to read a FAQ.
I think it’s more than 30 days. I’m trying to cancel and forgot to come in touch here. PLEASE HELP ME AND REFUND IT.

Another thing which makes me go mad is, once after the subscription is deducted from my card, I’m able to cancel it. And I did cancel it. But the money is gone. I want it cancelled and refunded.

[halcyon coral]

@cedar leaf I‘ve already written to you via DM, so please just be patient, thanks!

[forest sapphire]

hello

[sleek tapir]

Isn't there a cancel button online

unrelated, I want a "thank you" button I can click when perplexity does a good job

[pine dagger]

This is new

[sinful smelt]

Introducing the next generation of Claude - Claude 3 https://www.anthropic.com/news/claude-3-family

[errant moss]

Will Perplexity Pro users get access to Claude 3?

[sinful smelt]

One can live in hope 😉

[fossil sentinel]

1 million Tokens 💀

[errant moss]

Citations!

"In addition to producing more trustworthy responses, we will soon enable citations in our Claude 3 models so they can point to precise sentences in reference material to verify their answers."

[fossil sentinel]

Claude making a comeback?

😂

[azure crown]

damn. they gotta drop gemini 1.5 ultra now

[sleek tapir]

AI showcasing how competition drives progress is realtime

[azure crown]

gpt 5 announcement when

[halcyon coral]

Please check https://discord.com/channels/974519864045756446/977259063052234752
Thanks!

[wise edge]

powerful models back , The Avengers are come back

[azure crown]

took about 355 days

[wise edge]

openai sucks:
when i ask it to summary this video's script https://www.youtube.com/watch?v=ovkwsvbGq1I , it refused.
when the topic about USA's politics.

claude 2.1 and mistral large work well .
https://www.perplexity.ai/search/generate-in-some-rE_NuRa_Snio9dk7inaFEQ#0

unbeliveable ~~~~~~~~~~~

[fossil sentinel]

Doesn't mmlu have a lot of errors?

I remember watching a video some time ago

This one
https://youtu.be/hVade_8H8mE

[thick void]

For writing about technical subjects, I already preferred Claude 2.1 over ChatGPT4. ChatGPT has a terrible style of writing that I struggle to talk it out of. Too much sizzle too little substance. Claude seems naturally more to the point. Hope 3 retains that quality

[livid aurora]

Pls add https://www.anthropic.com/news/claude-3-family

[earnest elk]

can i not get the $10 discount on a yearly subscription that I am trying to buy right now?

i dont see any option for it

[glossy brook]

should be able to via a referral code? i know i was able to https://perplexity.ai/pro?referral_code=ZK089N02

[earnest elk]

well the problem is i signed up for the trial (yearly) and i realize that it will renew at the standard cost, so i cancelled it and tried the referral link, but it just does nothing

[glossy brook]

ahh ok. sounds like it may be a bug that customer support could help with. you cancelled during the trial or are you still in it?

[earnest elk]

i cancelled during the trial but i have 1 day left of it. it just wouldn't autorenew for now. but i wanted it to renew but with the $10 off

[buoyant shard]

can I use both gpt4, gemini , claude etc altogether with perplexity right?

[cursive jacinth]

Hi sir, Gemini is not available sir, Gemini Ultra unknown date to be integrated sir.

[livid aurora]

why sir?

[fathom pollen]

https://tenor.com/view/oliver-twist-gif-26543489

[heady quiver]

We will have access to Claude 3 OPUS with perplexity??? And Gemini Advanced???

[heady quiver]

It didn’t work for me either; I reported that 3 times and they don’t even respond lol; it’s a LIE!!! you will be charged 20$ a month no matter what, 10$ a month if u refer someone it’s a SCAM

[tame current]

perplexity with claude 3 opus when

[heady quiver]

Happened to me as well bruh. Use Mistral for such topics

[spice shore]

Was "copilot" rebranded to "pro"?

[gaunt crater]

But what about the 18 other services named Copilot ||15 of which are Microsoft, when I sat on a talk on Semantic Kernel they called it their Copilot development tool||

[empty quartz]

Anybody tried Claude 3.0 pro? How is it?

[jovial mantle]

Im trying via poe

It's good

Very good

[barren hawk]

Hope Perplexity has Claude 3 fairly quickly 🙌

[warm cave]

I was impressed it got this right, gemini advanced fails, and mistral only gets it right some times, but gpt-4 always gets it right

If you have any questions you want me to ask it, let me know

[gaunt crater]

Me over here 50% hoping they implement Claude 3 Opus and 50% hoping they also keep Claude 2.1 for reasons that involve the occasional unauthorized escape from incarceration.

[modern crown]

Why do you say sir after every sentence lol

[gaunt crater]

Sir, it is contagious sir, it makes me think of Indian English and now I too cannot stop saying sir, sir.

[modern crown]

Gotcha lol

[hollow silo]

Have you compared Claude 2.1 to Gemini Ultra 1.0?

In my experience, Ultra is better than GPT at creative writing, but I haven't compared it to Claude

[hollow silo]

Ask it how many unique ways there are to sort 5 labeled balls into 3 labeled boxes if all boxes must have at least one ball

[modern crown]

So Claude 3 is for beta users right now?

Someone posted a screenshot

[hollow silo]

GPT-4, Gemini Advanced, and every other LLM I've tested gets this one wrong on the first attempt

[languid nebula]

No, it is a fake.

[modern crown]

Well darn

lol

[hollow silo]

No, I believe if you have the premium plan you have access to Claude 3 Opus

[supple hare]

Yeah. Not available in the EU again...of course

[languid nebula]

Also not true, Claude 3 is not available for pro users either

[warm cave]

I paid for a month on claude and i have the option

[warm cave]

I answered it but did I forgot to reply to your message, here it is^^

[gaunt crater]

Poe already put up their template Claude-3-Opus chatbot. They work fast.
I have no sub over there but I do be itchy to try it.

[warm cave]

All of the AI's are all over the place, here is what each got: GPT-4 = 60
Claude-3 = 720
Gemini_Advanced = 147
Mistral_Large = 150
Microsoft_Copilot = 6

[hollow silo]

The correct answer is 120 iirc

Don’t think any of them got it right

[warm cave]

that is a good benchmark question for the future

[livid aurora]

what you mean?

it's available

[warm cave]

sorry, gpt-4 actually got it right, i forgot i had a weird prompt applied that messed with its math, here is what it got when i took it off:

[gaunt crater]

Nice. It can do n choose k type math.

[gaunt crater]

Kind of a good counterexample if Claude 3 missed it, since they seem to be taking the position it beats GPT-4 on reasoning.

[warm cave]

Yeah, it is

[hollow silo]

Interesting

[halcyon coral]

Yes but solved with python code

[gaunt crater]

We need to tell the model it's a closed book test.

[warm cave]

I can do that, but i feel GPT-4 with code interpreter is part of the selling point

[gaunt crater]

I agree, not only does it seem fine to whip out a calculator it's a very human thing to do.

Something something what a time to be alive.

[warm cave]

seems to do just fine without python as well: https://www.perplexity.ai/search/How-many-unique-FhBuPia.RUqOpS16s50zxQ#0

so i guess GPT-4 out reasons claude 3 in this particular question with and without a calculator

oof, i gave it a second try

[meager sparrow]

How do I access it? I try to get to Claude Ai but I need to put in my phone number. I don’t understand and if they is any other way to get onto Claude without it I would like to know

Sir

[surreal notch]

would Claude 3 be available as one of the models provided by Perplexity soon?

[grand sand]

most likely

[hollow silo]

Yeah, looks like they released their API already

[hollow silo]

Oh GPT-4 on perplexity doesn’t have access to Python?

[muted pine]

Claude 3 lets gooooo

[warm cave]

Yeah, no code interpreter for higher accuracy unfortunately, but I did add it as a feature request under pro feedback, so hopfuly we get it: https://discord.com/channels/1047197230748151888/1214295338991755375

[south kindle]

Testing out Claude 3 Sonet and it is VERY fast.

[steady crater]

is perplexy going to add the claude 3 for us

[steady crater]

buy rabbit r1 and get perplexity for a year

[sleek tapir]

that's what I did

free rabbit 😄

[steady crater]

why not right

[sleek tapir]

even if the rabbit flops, perplexity's good

[steady crater]

100%

[brittle dagger]

I would like rabbit, if they made it an app instead of yet another device to put in your pocket

[foggy whale]

holy crap, nerds, I've got to sleep some time

i'm not even where to start with today

[untold fjord]

Can you use the playground image generator on mobile?

[stiff solstice]

When will the iOS app get some love with the updated models 😭😭😭

[jaunty breach]

Is Claude3 really better than GPT4?

[halcyon coral]

No

[heady quiver]

https://youtu.be/x13wGyYzlJo?si=pZqHLVhaprgPcaWv

[heady quiver]

So slow but most powerful? Interesting….

[heady quiver]

Me 2

[granite grail]

idk but I'm curious to try it out

PS: perplexity with Claude 3 opus when?

[strong bone]

^ wanna know this too

[fossil sentinel]

Bro theaigrid 😭

I got tired of that channels format

[icy zenith]

^^^

[heady quiver]

We want CLAUDE 3 opus in Perplexity!!!!!!!!

[grand sand]

Soon

[granite grail]

https://tenor.com/view/puppet-red-ball-soon-gif-13188418

[jaunty breach]

I tried to log in to Claude3, but my account was immediately blocked.

2 accounts!!!

[heady quiver]

No? Based on what you say that? All benchmarks and videos I saw are saying the complete opposite lol. Just say it’s expensive for perplexity to implement that which may be the real truth.

[heady quiver]

Lie

[tame current]

New AI imagery update is cool

Was kind of thinking of cancelling my subscription, since I was starting to get a tad bored with the service, but now I think I'll un-cancel it

I'm already using it

I'm a fan of the stylistic options too

While you can technically have the client versions of ChatGPT and Gemini create more stylistic variety for their images, it takes some clever prompting

Perplexity's ability to choose a style manually instead of letting the AI interpret it entirely on its own is nice

[grand sand]

true

[livid aurora]

How to make image in perplexity?

[grand sand]

like this

you can only do it on the web

[tame current]

It'll be nice when it becomes available for the mobile version

[livid aurora]

what is this

[grand sand]

the generate image feature needs its own section in the app

[livid aurora]

It needed it also on the web

[grand sand]

where is that?

[livid aurora]

the icon on the right

[grand sand]

oh this is cool

[livid aurora]

I thought you can do widescreen images

also it should not be in the normal chat tab

[grand sand]

still cant do hands

same prompt but DALLLE 3

[serene tide]

anyone involved in the hardware side of building/running LLMs?

[quasi gale]

Can we expect Claude 3.0 to be integrated into Perplexity?

[sacred furnace]

When we can get access for Claude 3 in Perplexity?

[cursive jacinth]

Hi sir, no one will answer this question sir. Cause no one has an idea sir.

[heady quiver]

We want Claude 3 Opus in Perplexity this week!!!!!

[tame current]

Claude released a new model?

Interesting

Looks great on paper

After Gemini though, I think i'll wait until we actually get to use it to judge how good it is

[sharp timber]

Why so verbose? Is there a switch as a shortcut, so that I don't have to always define the prompt, in order to get mor concise, even repetitive answers?

[drowsy lynx]

This

[cursive jacinth]

Hi sir, great model sir, equals to long wait sir, like Gemini sir.

That is why i am not hyped sir, i hope we dont wait for open ai to release their GPT 5 sir before we get Gemini Ultra and Claude 3 in perplexity sir. 😂

oh sir, Gemini 1.5 also sir.

[fossil sentinel]

Anyone wanna test a question for me if they have access to Claude opus and compare?

[livid aurora]

bro stop using sir 3 times

[thick void]

No I haven't had the chance to do that yet

[cursive jacinth]

I cannot sir, it is in my genes sir. I am sorry sir.

[fossil sentinel]

😂

[livid aurora]

use claude etc for free https://arena.lmsys.org

[fossil sentinel]

Gooddawg was raised with the highest of discipline

[cursive jacinth]

My parents taught me well sir. That is why i am a good dog sir.

[fossil sentinel]

Forgot about that thx

[primal sluice]

So it looks like the copilot option wes renamed recently, this weekendish or am I really far behind paying attention?

[prime mirage]

Anyone see a date perplexity will get Claude 3

[violet venture]

Hi there, new user (pro), excuse me if there's a more relevant channel. I'm trying to get perplexity to help me read some PDF files, which I've added to the conversation, but it tells me it cannot read PDF files. How come?

[hoary python]

is the file too large?

or just maybe corrupt?

[violet venture]

Less than 50 MB, I can mark text and c/p

They range from 2.82 to 13.3 MB

[hoary python]

I think their is a cap in terms of how large a file can be when uploading if I'm not mistaken and have a max fo 4 per query

have you tried other PDFs

if it's strictly just that PDF and it isn't size, you can open a bug report

[steep pilot]

Can you send the pop-up it’s giving you?

[hoary python]

^

[violet venture]

I'm just wondering why it gives me this answer:
"No, as an AI language model, I do not have the capability to read or access external files such as PDFs."

[violet venture]

How do you mean? The entire answer? Or as a link?

[south kindle]

sharing from the anthropic site.

[hoary python]

yea this is a bug

it's documented in the bug's channel, they know about it

I'd recommend sharing the file or documenting it in a bug report for them to see

[violet venture]

Thank you, I see it. That's sad, it was the reason for subscribing as pro.

[hoary python]

I am using sonnet rn

it's crazy

[steep pilot]

Talk to pro support apparently they’re fast

[hoary python]

you can try to revise your prompt

I believe it's a wording issue

you could say

looking at the attached files ...

that's usually how I say it

[violet venture]

Yeah, already tried that like ten times... thank you both for suggestions. I don't think support can help me if it's a general bug.

[hoary python]

have you tried changing models?

[south kindle]

try it along with this plugin: https://glasp.co/youtube-summary

[violet venture]

Last attempt was with this question: "Just answer yes or no, can you read all of the three PDF files I gave you?"
It then gave me an elaborate answer about why it can't read external files.

[violet venture]

Yes, tried all of them.

[hoary python]

uh

riddle me this, did you try other PDFs?

or open another thread

[violet venture]

I don't see why that would change the answer.

[hoary python]

it would

context

sometimes when I have used it in the past I usually attach pasted txt or log files and I'm like

based on the attached files can you answer ...

[south kindle]

the settings menu to adjust so it uses Claude.

[grand sand]

dang this is pretty good

[hoary python]

I want to use opus I am using sonnet, I can't wait for it to be matched with perplex

[south kindle]

: (

[hoary python]

oh yea that's sick

is that with the new engine?

[grand sand]

dalle 3

[hoary python]

oh

still looks fabulous

[south kindle]

what image prompt?

[grand sand]

i havent really tested the new one yet

[hoary python]

I am bad with prompt engineering

💀

[grand sand]

"generate an image of a dragon hugging a human"

[hoary python]

very simple yet creative

[south kindle]

tools / direct prompts exist to transform your "bad" prompts into "good optimized" ones

[grand sand]

I love prompt engineering

its like describing to a kid how to write an essay

[grand sand]

all we need is a ai prompt for quick search

[hoary python]

It's neat I really like it it's just when it comes down to it like actually formulating one is the problem

Opus w/ PPLX will be insane imo

if sonnet is doing what it is with me rn

like I can't imagine

[grand sand]

why is it a big deal?

[south kindle]

i'm from a technical support background and had to learn to "support" people that were both MUCH smarter and less intelligent than myself. forced me to learn how to communicate effectively regardless of who I'm working with / supporting. that "skill" or whatever directly improves prompt engineering OR at the very least, working with AI tools and using language effectively.

[grand sand]

havent read about it yet

[hoary python]

watch the video it's nuts

https://youtu.be/x13wGyYzlJo?si=pZqHLVhaprgPcaWv

[grand sand]

same here, im really good at explaining complex stuff to old/young people

[hoary python]

good people

[hoary python]

I can't communicate complex topics in simple ways unless I can think of a entertaining way of doing it

[south kindle]

not bad.

[hoary python]

I like that

[grand sand]

is that the new model?

[violet venture]

Bit weird, it seems like it can read the Mischel text, but not the Ewen. In previous answers it incorrectly uses Mischel references to parts of the Ewen text.

[grand sand]

try again but turn off Pro search

[south kindle]

nope, just copy/pasted into Copilot sidebar with creative mode. I don't have Pro.
https://www.bing.com/images/create/a-dragon-hugging-a-human/1-65e675ab77f44d8fb41233f50f0d52a3?id=sxyZ%2Bj6bn0kaGvmzmD%2BwWQ%3D%3D&view=detailv2&idpp=genimg&idpclose=1&thid=OIG1.v_KxPU7QkjDuVMfRMxS6&frame=sydedg&form=SYDBIC

[grand sand]

microsoft uses dalle 3 i believe

[violet venture]

Which one would you suggest? Experimental? GPT-4?

[grand sand]

WOAH

[grand sand]

if you upload a file its always gpt 4

you can change the default model if you wish

[violet venture]

Thank you, that helps understanding limitations.

[grand sand]

no problem, if you have any more questions ill gladly answer em

[south kindle]

that still counts though. do you try that with GPT / AI tools? I also find that using brackets / parenthesis along with ! and other punctation to act as commands and direction/rules helps a lot. I have some standard replies I have saved into a private discord channel, so I can just fire off the same set of 4-5 prompts using copy/paste. having stuff organized like that helps a lot when it comes to ability / speed at getting stuff done.

[hoary python]

only have been using AI for a little over a year and a half but I mainly use it for codegen

and debugging*

or even refractoring

[violet venture]

Holy crap, now it's on drugs:

[hoary python]

look at that

[south kindle]

same here honestly. I got really into it just before GPT4 was released. and just sorta dove in head first from there.

[hoary python]

yep as soon as GPT3.5 released I was all over it then I heard about PPLX and swapped over as it was the same amount with more models/features

I rarely talk in here I need to more often

a lot of different and cool discussions

[south kindle]

example of my discord and my prompt channels that lives on a 3rd monitor, opened full screen.

[south kindle]

is your AI profile active currently?

[grand sand]

how did you get that to work?

[hoary python]

he used the prompt I told them to use (: and it worked

definitely write up a bug for it though

[violet venture]

It's not working, I didn't ask about Freudian psycho analysis.

[hoary python]

what

[violet venture]

I don't understand the question.

[hoary python]

can you share the document or is it classified

[violet venture]

Well... it's provided by the university I attend, not for sharing.

[hoary python]

all I needed to know

[violet venture]

But if you can find R.B.Ewen, An Introduction to Theories of Personality.pdf ...

[violet venture]

Meaning?

[grand sand]

https://www.perlego.com/book/1323727/an-introduction-to-theories-of-personality-7th-edition-pdf

[south kindle]

check your options

[hoary python]

is this it?

[grand sand]

i cant wait for perplexity to get that new gemini model

only gonna use it for pdfs though lol

[violet venture]

I'm there now, don't see an "AI profile" section. Do you mean my ... default model? Just on default now.

[grand sand]

its this

[south kindle]

are you at this site: https://www.perplexity.ai/settings/profile or are you on mobile / ios / android?

[violet venture]

PC - sorry, I'm using the website on Windows.

[heady quiver]

STOP SPAMMING WITH THIS PDF BUG and random Dalle images !!!

[south kindle]

it's the tab over from the one your on.

[hoary python]

@violet venture

[violet venture]

Dang it, that's the answer I wanted. I wonder if your version of the PDF is different.

[grand sand]

works better without pro search

[hoary python]

yep

[grand sand]

you had to turn off pro search

[hoary python]

and change model* and specify context*

GPT4 yelled at me too

[grand sand]

i thought only gpt 4 could read files?

[hoary python]

no

clearly not

[grand sand]

lol

[hoary python]

look at it

also you could just do ctrl + f and type in your query while the PDF is open

||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​||||​|| _ _ _ _ _
https://cdn.discordapp.com/emojis/1207250334687432764.gif?size=64&quality=lossless

[grand sand]

technology is a wonderful thing

[violet venture]

Yes, but the idea was to search multiple sources at once.

[hoary python]

elaborate

tbf though I would argue that PPLX needs to up it's game with being able to comprehend and search the rust docs/ crate documentation @south kindle or maybe it's my prompt engineering

[violet venture]

With a different question I might want multiple sources as reference to my statement.

[hoary python]

sources as in files

or just one file

or external sites

[south kindle]

I doubt it.

[violet venture]

"Overall, while the PDF does not directly discuss nomothetical and ideographical approaches, it does touch on related concepts and can be seen as representing a mix of both approaches."
I think there's a problem reading the PDF. I thought it was OCR (or what ever, noob in this tech)
I'm now using Mistral with Pro off.

[hoary python]

did you use my PDF

[violet venture]

Yes, depending on the question I ask, some need to be specific texts, other questions may involve a larger curriculum.

[hoary python]

and that's where context window comes in

I feel like PDF reading in particular is a niche thing but it's possible

but like I tend to evade it and directly pass in textual information

plus idk why you wouldn't read this, I don't judge but it seems enticing

[violet venture]

If you mean the one @grand sand linked, I don't have access to perlego.com (no subscription).

[hoary python]

,

[violet venture]

Missed that one, thanks.
It's not the right edition, though, I need the 2014. Appreciate all the help, but...

[hoary python]

why is that

there must be a way to like forcefully make it query rust docs

maybe using the actual Labs

I want it to be able to query like given crates and rust docs so it knows syntax and stays up to date with method calls being used

[hoary python]

bruh

lmao

I think it's biased

it likes me

where PPLX t

[violet venture]

The correct term is "ideographical approach", that might be it?

But as a language model, it should be able to correct that...

[hoary python]

should be able to be partialeq

[grand sand]

none of the models did a good job

[hoary python]

weird

really weird

[grand sand]

oh my

gonna make claude 3 opus the default model when it comes out

[violet venture]

I'm gonna lie down for a while and clear my head... back to dragons and future endeavors in AI competition.

[grand sand]

lol

[violet venture]

I have read it, but remembering where all the arguments are is pretty hard. For me, at least. I do find psychology a very interesting subject, too. That's why I study it at uni.

[amber pumice]

Does claude appear to be better in functionality?

[hoary python]

yes

[violet venture]

What kind of functionality? You're testing it right now, @hoary python, right? (I can read the chart above)

[hoary python]

yea in terms of just code generation it's very good, especially when given context

[violet venture]

That's the most important part, if we want AGI to be birthed from LLM.

At some point in developments, someone is going to ask an LLM: "Write code for an AGI", and... no answer. Just silence. Might seem like a bug, but...

[dire portal]

Would like to know if pro users are going to get Claude 3?

[hoary python]

eventually they said

[heady quiver]

Who said that? I got this response: Opus probably won't come to Perplexity since it's just too expensive or only with a low usage rate
https://www.anthropic.com/api#pricing

[carmine copper]

The Opus model is actually similar in price with the old GPT-4 model. Perplexity ran that model for quite some months.

[fossil sentinel]

But the question is would it be worth it in today's AI market

Different times back then

[chilly forge]

^

[heady quiver]

Similar in price??? Is 4 times more expensive

perplexity has claude 2.1 not claude 3 opus lol

[carmine copper]

at least in open router

that old gpt4 model

[heady quiver]

K

[carmine copper]

turbo is half~third the price. But perplexity ran the old model for almost a year i think

[chilly forge]

Thats what we were waiting for haha

[mental harness]

Perplexity controls the token limits in the background, that lets them adjust the costs accordingly.

[chilly fox]

Claude 3 is now on Perplexity Pro

Limited to 5 Opus answers a day

[chilly forge]

Lol thats sad !

[carmine copper]

Can we have an option to just use the Opus but at a rate of 200~300 times per day?

[chilly forge]

BTW what happens once we exhaust the 5 Opus answers ? does it switch to Sonnet ?

[tame current]

I always use Claude 2.1 but how good Claude 3 really Is?

[sullen cloak]

Wondering that too

[tame current]

Actually the performance Improvement Is acceptable but Imagine when Claude 3 become overly "smart" and just censoring some harmless prompt again lol.
Y'know that annoying Claude's ethic stuffs.

I think Claude 3 maybe very expensive though, I wonder how perplexity can pull It off.

[fossil sentinel]

Kekw

[wise edge]

fire ~

[tame current]

Yo It out now?
Let me test my stuffs.

[fossil sentinel]

Only opus remains 98%+ recall at longer context 😭

[tame current]

Bruh why It have to always be near 100?

[fossil sentinel]

U talking about the recall?

[tame current]

Yeah

I never see that thing reach 100

Claude Is still like good old days lmao.

Just saying "Hi Mistral" bro thought I am asking for meth recipe or something.

[chilly fox]

Is the pro search working for anyone with Claude 3

It cannot get passed understand question

[warm cave]

and there goes 1 of 5

[fossil sentinel]

Lmao Claude opus for pro users is like the 5 copilot uses for free users

[tranquil orchid]

Copilot is limited to 600 queries per day

Claude 2.1 and GPT-4 queries are counted as uses in the same context as Copilot. Each use is given back to you 24 hours after you use it. You will be able to see the exact count when you hover over the Copilot toggle.

[fossil sentinel]

Except no payment option to get more uses

[chilly forge]

That was my reaction too xd

[lusty basin]

No longer seeing "limited to 5 Opus answers per day". Did they decide against that or does it downgrade you to Sonnet after 5 uses?

[chilly forge]

Now I highly doubt whether its actually using Opus or not

cuz the pricing aspect still doesnt make sense

[lusty basin]

It says it is using Opus (this is only my second query though)

[hoary python]

WWWW

[chilly forge]

Actually the "5 uses per day" label is only on the profile page where we can select models not on the Rewrite menu

so its indeed only 5 uses per day

[lusty basin]

What happens after 5 uses? Does it downgrade you to the cheaper Sonnet model or prevent you from using Claude entirely?

[chilly forge]

No idea thats something I am also curious to know Please exhaust your uses so that we can know

[lusty basin]

I'm on it

[hoary python]

I think I found a bypass

so

if you edit your query on a thread it doesn't count the amount of uses

[chilly forge]

How are you determining the usages though ? because there is no counter for it right ?

[hollow silo]

It automatically switched me to the model I last used, GPT-4 Turbo

[chilly forge]

[chilly forge]

imo once we exhaust the Opus we should still be able to use the Sonet model

[warm cave]

Agreed

[hoary python]

I'm not

I'm just doing it over and over

and it doesn't change from opus

[ivory bluff]

honestly it would be cool to be able to choose between 2.1, sonnet, and opus, personally

[chilly forge]

Matter of time when the team fixes it

[hollow silo]

Oh nevermind, it looks like it switched me to Sonnet

[hoary python]

yes

I ain't no abuser

[chilly forge]

[hoary python]

I ain't closing my thread now

🥱

[warm cave]

have all Claude models, but different use amounts, have it so the best uses up 5 usages and have each model down cost less, and 2.1 cost 1

[restive sky]

did we actually just bully them into making it unlimited

doubt it

[warm cave]

prob just buggy

[hoary python]

if they did

that's crazy

[hoary python]

yes

@ivory bluff why?

Honestly I'm preferring sonnet over opus right now

[ivory bluff]

so am I

I'd prefer to use sonnet but I've been stuck on opus lol

[lusty basin]

For me its still sticking with Opus after 8 queries

[warm cave]

why do you guys say that? i have not tried sonnet yet

[hoary python]

I tried sonnet on test but not in PPLX

[heady quiver]

Say thanks to that brother; poe is no longer offering a free trial or free gpt 4 searches

[warm cave]

so, what makes sonnet beter in your opinion?

[heady quiver]

For those that really need opus just go to poe.com and get yourself a subscription and u have that opus there (600 daily) lol

[warm cave]

I think you get more like 44 a day

[wispy spear]

I'm subsribed at Poe, but don't see Opus. Where is it?

[heady quiver]

https://poe.com/Claude-3-Opus

[ivory bluff]

Opus likes to standardize its responses to follow a template, from what I'm witnessing. Sonnet's a lot more creative with its responses.

[warm cave]

(1000000/750)/30=44.44

[wispy spear]

Thank you, much appreciated.

[heady quiver]

44 a day? 44 then

[tame current]

Damn Claude 3 Is real good.
Reasoning Is spot on.

Still receiving ethic block from time to time tho lol.
Not much trouble.

[digital siren]

Yeah this new point system is terrible for Poe. Seems like much worse volume than ChatGPT plus. I keep Poe for everything else though.

[wispy spear]

I'm crafting press clippings for my clients and the text it outputs is incredible, and it also follows my half-assed prompts very well. I'm happy.

[serene sand]

#💗│sharing message ☕

[heady quiver]

The point system is trash yeah

[fossil sentinel]

I tested a probability question on arena 2 times between opus and gpt 4 and opus got it wrong one out of the 2 times kekw

What was the benchmark results for 0 shot again

[ivory bluff]

I'm like 50 queries in and I'm still on opus 😐

[warm cave]

I think its when you move to a new thread, not sure 🤔

[restive sky]

how can you tell it’s opus and not sonnet

[ivory bluff]

It tells you on the tail of the answer

[restive sky]

mine reverts back to nsfw-waifu-beta-7b after 5 opus queries

[chilly forge]

[heady quiver]

Waifu beta wtf perplexity?????

[heady quiver]

It’s a bug still displaying opus, u are not on opus after 5 searches

[warm cave]

why only 7b? 😢

[hoary python]

bro I got timed out

for sending three messages

fix this bots logic fr

[restive sky]

they’re trying to mitigate the opus dissent

[hoary python]

fr

censorship

[restive sky]

and silencing the waifu lore

[hoary python]

petition for more Opus queries 🥱

[restive sky]

i’m going on strike

[hoary python]

fr

[restive sky]

hunger strike

[heady quiver]

More opus needed? Then perplexity will increase its price; is really that simple.

[carmine copper]

just increase its usage rate by x2.5 since it's x2.5 more expensive than turbo

[hoary python]

fr

I agree with the above statement

tbh rn with sonnet I am getting like repetitive responses and inconsistency

just keeps relaying to a prior response

[ivory bluff]

I'm kinda finding claude 3 opus to be a bit infuriating to use compared to 2.1 right now

It just went to sonnet for me and it's so much better, for me at least

[copper willow]

Figure out a way to allow more image credits please, the current limit is too restrictive, I'd pay more subs for more gens

[heady quiver]

What limit for image creation is there?

[heady quiver]

On what kind of tasks u used opus and not satisfied with it?

[serene sand]

50 per day

[heady quiver]

U can use microsoft designer and generate your images for free there; by dalle 3

Or better; use leonardo AI; its the best that I found for image generation

[heady quiver]

This popup should be updated with claude 3

[shadow saffron]

I found Misteral to be the sweet spot between GPT4's reasoning and Claude's readability.

Since Opus is limited rn, do you think Misteral still a better better bet as the default model or C3 Sonnet crushes it?

[agile jay]

Mistral large is probably the least lazy for writing code.

[keen solstice]

I see Claude 3 but limited to 5 query / day

[agile jay]

Could be due to Claude limiting their API at release.

[shadow saffron]

Yah I just tested a video transcript summerization prompt on all models and claude 3 performed the worst by miles and also extremely verbose!

I don't understand the benchmarks ...

[agile jay]

Was it opus, or the liter models?

[shadow saffron]

Sonnet :/

Imma stick to Misteral / GPT4

Don’t know what the fuss all about, specially with the advanced model (opus) being so limited at 5/day

[keen solstice]

True, all the test I saw on youtube until now are very mixed result, maybe the coding part is a little bit better

[shadow saffron]

I suppose...but the gain seems to be somewhat marginal in coding compared to gpt4 at least

Unless ppl excited about the 200k context and recall

[solemn trellis]

is there a way perplexity can directly manipulate google documents

or google sheets

[shadow saffron]

Sorry one last message about this; Anthropic's benchmarks are outdated. Someone compared them to latest GPT4 benchmarks and even though Opus is an impressive model, the real comparison is not really that "industry shaking" compared to what's already available

[restive sky]

benchmarks are saturated and meaningless. opus is good because it has a soul

[chilly forge]

Nope its due to pricing constraints, Claude didnt mention anything as such for "Limiting the API"

[chilly forge]

Yeah even I had got somewhat similar to this once

It had like 10 different modes lmao

as soon as I was about to take SS page just refreshed

[small hemlock]

Is there a way to talk to wake up Perplexity and continue the conversation with just your voice so I don't have to hold my Android phone to operate it ? Probably need another app but just wondering. Thanks

It would be nice to have something like the functionality of Rabbit R1 or AIPin as just an app without a secondary device

[calm rose]

https://www.perplexity.ai/search/Apple-Vision-Pro-iWrw6S6NQeeqYG2xe8poJg

[small hemlock]

I've got those but the Meta AI is sH1T in the UK

Samsung need to do an AIPin like brooch camera and then provide the rest as an app

[agile jay]

The price is only 2.5x more for output compared to 4Turbo.

[fair ice]

.

[gaunt crater]

I noticed I got a couple Claude 3 Opus responses before they changed to Sonnet.
What are the conditions for getting to use Opus vs. Sonnet? Knowing this is probably a tricky business because of Opus's costs.
Is it a certain # of uses over time, like sliding window or replenishing uses? Or is it conditioned on something like the type of prompt or whether it's being regenerated?

[granite grail]

5 per day

[gaunt crater]

😦
Gotcha. That does feel like quite a drop unfortunately from the ~600 or so.
Sadly I am pretty sure Sonnet is not quite as smart as Claude 2.1.

[gaunt crater]

Definitely feeling spoiled with Claude 3's capabilities but that's a big reduction on uses.

[granite grail]

I'm guessing that they implemented it ASAP without studying the actual impact on costs, so they just left 5 until they figure it out

[gaunt crater]

Maybe they can re-add Claude 2.1 as an option? Unless I'm wrong and Sonnet is not often worse than 2.1.

Or at least until they figure out their longer term plans or what they can afford to provide.

[jaunty breach]

Can perplexity do this? Can Claude 3 opus be charged extra?

[supple hare]

Sad that we don't have more of Opus, but to be honest I'm very impressed with Sonnet. It does great with summarization workflows using the Perplexity chrome extension. It's great at following instructions set in the profile. GPT-4 sometimes just ignores them. Tested a bunch of searches also and I can't see it worse than GPT-4, it is however quite a lot faster, which is nice.

[tame current]

Sonnet Is so damn good.
It understand every document I uploaded even half-baked low quality prompt.

Opus even better though.

[icy zenith]

5 A DAY???

blasphemy

[flint matrix]

Haha heard ai pen

Product is crap

Doing massive layoffs

Been out for a while still can't deliver just get a rabbit 🐇 R1

The only hand held ai assist you'll ever need

[sleek tapir]

my dream is a partnership with midjourney and just use the sub for that. I love MJ's model

[copper willow]

If only I could afford multiple subscriptions I'd be all over MJ

[torpid comet]

thank you for making it free although it is obviously worse than chat gpt and gemini

[desert field]

When is Claude 3 available on Perplexity? On iOS I still see 2.1.

[jaunty breach]

What phone numbers did you all use to register for Claude3? I was banned as soon as I registered.

[restive sky]

1-800-273-8255

[woven cosmos]

Once the Claude 3 opus requests have been exhausted, which model takes over?

[limpid mason]

Claude 3 Sonnet

[woven cosmos]

Ok thanks

And is there a limit on the use of Claude 3 Sonnet?

[limpid mason]

same limit as for the other models (GPT-4, Mistral Large,...)

[grand hearth]

Whoa, just saw this. Perplexity / Nothing collab? (I have a Nothing phone) 😄 https://twitter.com/AravSrinivas/status/1746951840825262269
(sorry if i missed if this was discussed here already, I did a cursory search first)
Edit: just noticed the tweet was from January. But chatter about Perplexity in the Nothing discord really kicked up after the Nothing 2a phone release yesterday

Aravind Srinivas (@AravSrinivas) on X

Perplexity 🤝 @nothing

Twitter · Jan 15, 2024, 5:45 PM

Nothing Discord has been chatting it up

[solar trellis]

how many limit question in gpt 4 turbo?

[glass spear]

600

[sleek tapir]

what's the partnership?

[near imp]

how do I turn OPUS on or off for Claude 3 in my Pro sub?

[icy zenith]

rn

ohhh ios

[near imp]

it says I have 5 queries a day, but no switch to differentiate between Opus and Bonnet

[icy zenith]

app gets updated later

[icy zenith]

no manual switch, it's automatic

[ivory bluff]

Yeah, they need a way to choose. I just don't understand why they would put an automatic switch in there

[icy zenith]

ig opus limiting is temporary

[ivory bluff]

Choosing between opus/sonnet/2.1 would be ideal

[icy zenith]

or at least

like preparing for heavy load

not every pro user uses all 600 turbo uses every day

[restive sky]

my pro subscription is also temporary

[icy zenith]

LMAO

[near imp]

It's annoying for me, as I use Perplexity all the time.. and I don';t; want to waste my opus queries.

[icy zenith]

thats funny

[icy zenith]

again, they'll prob raise limits once they figure out load

[chilly forge]

Nothing is actually a "Nothing" brand dk how Aravindh agreed to collaborate with them

[ivory bluff]

Opus is heavy enough to likely still require a quota even when they figure out load

should at least have a choice of when to dip into opus, imho

[sleek tapir]

I mean they collabed with rabbit, and rabbit's even less of a something than nothing

nothing actually has products released

[chilly forge]

Makes sense lmfao

[grand hearth]

Yeah, I have 2 of their phones and am in their Discord. Big fan of Carl Pei

[sweet ember]

Ordered the 2a 😄

[grand hearth]

Oh nice! I'm in the US. It won't be coming here

[random spade]

Water Management in Bangalore. Synopses

[tame current]

what's opus ?

[agile jay]

Claude 3 opus

is the biggest new claude 3 model

[limpid mason]

Honestly, i hope we will get at least a 30-40 messages/day for Opus in the coming days

[sleek tapir]

My experience using the new text and image models

[ivory bluff]

Been having that a lot today

[sleek tapir]

I tried to get it to genereate a logo for a bowl of dry breakfast cereal and it denied the generation due to moderation 😐

[agile jay]

How did you try that?

[halcyon coral]

Due to errors/overloads of Anthropic's servers, there may be more errors with these models on Perplexity.
@ivory bluff @sleek tapir

[lusty mantle]

Hey

[sleek tapir]

"generate a logo for a company called "dry cereal"

[agile jay]

Oh, for me when I switch models, options dissapear.

[sleek tapir]

Oh interesting

[halcyon coral]

This can happen sometimes, just try again 😉

[agile jay]

For example, when I switch to claude, image generation dissapeared from the search section.

[sleek tapir]

Yeah I just expanded on nit and got more specific

the playground image model is good

[woven cosmos]

You just have to start a new conversation

[sleek tapir]

is "pro search" the same as "default" with the "pro" slider checked?

[woven cosmos]

The pro search is here over the model you use I believe

[austere kestrel]

They are all different models. Think of 'Pro Search' as perplexity's Default model. The Pro search toggle is a separate feature

Not sure if that makes sense ha

[hearty quartz]

CLAUDE 3 OPUS is amazing!

[austere kestrel]

So it's not the same - to answer your actual question @sleek tapir

[gaunt crater]

What's the relationship between Pro Search and Experimental? Are they the same architecture/size/training or differences?

[sleek tapir]

If I weren't a pro subscriber, it would just say "default", then?
Just noticed that in the re-write popup, it was "Pro Search", but there was no such model in the settings is all

Which made me wonder if "default" was different for free vs pro

[gaunt crater]

They could rename Pro Search to Copilot Search so it's less confusing.

[austere kestrel]

I'm not too sure what it is now (feel like Mixtral might be used now), but iirc it used to be:
Default = GPT3.5 fine tuned.
Experimental = LLaMA-2 fine tuned

[gaunt crater]

Gotcha. I wasn't sure if they might be related (like the same model but different sizes), or essentially two different offerings. Sounds like the latter.

[austere kestrel]

I think if you weren't a Pro subscriber you wouldn't have the option to choose different models, so it would just be Default yeah (but I'm not really sure tbh)

[warm cave]

I’m pretty sure if you select rewrite with pro search, it runs the same model you had used befit but with pro search

[austere kestrel]

yeah I know it's confusing (it was called Copilot, now it's Pro; but in any case they refer to separate things: one of perplexity's models (let's call it Default), and; a feature on the app that allows you to retrieve more sources and clarify queries)

[gaunt crater]

IIRC you can still see the dropdown selections (everything but Pro Search with the Pro badge), but probably get blocked switching or breadcrumbs to subscribe.

[austere kestrel]

that is also confusing re-reading it lol

[sleek tapir]

Not sure if it's a bug or not, but I think something's up with image search results in threads with any follow ups. The results are strange. Not as if they're not taking into account the follow ups, but as if information is being dropped sporadically

[minor shard]

Just subscribed for the first time to perplexity. Really surprised this is not more popular than buying directly from a vendor like OpenAI. You get access to multiple of the best models out there.

[mental harness]

Did you take advantage of the rabbit r1 deal? I'm not totally sure the device will be very worthwhile, but it's $200 and comes with a free 1-year perplexity subscription, so that's pretty neat.

[cursive jacinth]

Hi sir, any problem with facebook sir?

Is the pro loung gone sir? why i dont see it sir

what is happening sir?

[livid aurora]

Is Claude 3 limited?

[lusty mantle]

When is Claude 3 being implemented or is it and I'm dumb?

[livid aurora]

Bro it is I just checked

[lusty mantle]

Where at I just bought it

[livid aurora]

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

Please remove limits for Claude 3

[scarlet schooner]

need to see the daily limit left for the opus answers

[livid aurora]

Is there a free model for claude in perplexity?

because there are 3 models

I dont get it

[scarlet schooner]

i only see 1 claude model in perplexity, that is claude 3

but we don't know after the opus answers are used. which claude 3 models will be used

so yeah, i can't answer you because idk as well

[halcyon coral]

#👑│pro-lounge message
@livid aurora

[livid aurora]

I dont have access to this

[scarlet schooner]

noted

[halcyon coral]

Yes you have to leave and rejoin via the link at https://www.perplexity.ai/settings/account

[livid aurora]

okay see u

[livid aurora]

thx

[lusty mantle]

Hmm not showing for me

I have pro too

[scarlet schooner]

you have to leave this discord and rejoin by following the link that icelavaman shared above

[lusty mantle]

Is it not on the app?

[scarlet schooner]

on the web ver

[lusty mantle]

Ah got it

[minor shard]

Did they stop this promotion?

[mental harness]

It's not advertised anymore but when I ordered one on 2/29 I got an email afterwards with the redeemable credit. If you order and the credit doesn't show up I guess you could just cancel the order if that's the only reason you wanted it in the first place

[craggy prairie]

my credit didn’t show up but I just emailed rabbit and they gave me the code in less than 24 hrs. Very responsive team

[languid nebula]

The promotion is for the first 100k preorders. They are in the 6th batch which is like 50k unites, previous batches were 10k each. So long as they are in this batch, I think you still get the code, but it is a little of a gamblell

[minor shard]

Nice thank you everyone!

[mental harness]

I hope people aren't claiming the free year and then canceling their rabbit order anyway, hopefully they have a way to go back and remove the promotional credit in that case

[tropic scarab]

When will Perplexity offer the possibility to generate images, using text prompts?

[languid nebula]

You already can, just create a collection with an AI prompt instructing it to generate a prompt for the image generator. Or you can tell it to just retype your prompt exactly.

[languid nebula]

Here is an article confirming 50k units in the 6th batch. https://dataconomy.com/2024/01/22/rabbit-r1-perplexity-ai/

[mental harness]

Does it use the reply to generate the image, or the user's prompt?

[languid nebula]

And I just checked the Rabbit r1 website and the preorders are still in the 6th. So you can still get the perplexity code

[languid nebula]

It is the reply that is used to generate

So if you tell it to generate a prompt for the image generator, or to retype your prompt exactly, you can get a decent imagegen

[tropic scarab]

I'm logged in on Perplexity now. I told Perplexity to generate green skyscrapers and got the answer: "I'm unable to generate or display images directly. However, based on the search results provided"

[languid nebula]

So like I said, create a collection with an AI prompt that is either something like this "Generate a prompt for the image generator StableDiffusionXL" or "Retype my prompt exactly" then create a new thread from within that collection and set the mode to writing

Make sure pro/copilot is turned off

[languid nebula]

So something like that

[sleek tapir]

That never works for me, I just get

[languid nebula]

Did you create the collection, create the thread from within the collection, turn pro off and set source to writing?

[sleek tapir]

Whoops, didn't set it to writing

Rewriting in claude worked too

[languid nebula]

Another option is for the AI Prompt is just "retype this exactly" which will have it retype your prompt so you can send it to the image gen without modifaction

[sleek tapir]

The image it creates is still a little screwed up by the "here's a prompt" intro

but taking the result and shoving it into the generator manually works great

😄

[languid nebula]

yeah, it takes some playing around to get it to generate the prompt exactly as you want.

you might have to type a bunch of instructions in the AI prompt to get it to do what you want

but this way you can create your own customized imagegen experience

It is honestly better than a dedicated function imo

https://discord.com/channels/1047197230748151888/1202920197720375296

[sleek tapir]

Not using the collection at all and just using the search works fine too, set to writing

Using pro search also works fine

[languid nebula]

https://discord.com/channels/1047197230748151888/1183781616515031124

Yeah, but with the collection you can customize how it writes the prompt, and you don't have to ask it to do it each time, you can just describe the image you want

[sleek tapir]

that's true, you're right

[modern crown]

So we confirmed Claude 3s biggest model is only 5 use cases right?

With pplx right?

[languid nebula]

yes, you get 5 opus uses and then it goes to sonnet

with pro

If you have pro, give an uptick on this https://discord.com/channels/1047197230748151888/1214608425862824017

[sleek tapir]

I have pro but I don't have access to that channel

[halcyon coral]

#💬│general message

[minor shard]

I love perplexity but it seems like they don't try to make their UI intuitive lol. Having to go into settings to change the model and having to first generate a prompt and then choose to create an image can be annoying for some I am sure.

[sleek tapir]

Nice

[tropic scarab]

Yes, I created the collection, created the thread from within the collection.I chose Claude 3 Opus.

[languid nebula]

This is actually on purpose from what I gather. They are in the process of developing their own dedicated models to run their system. Eventually they will probably remove the access to premium models entirely

Once they get their own models up to snuff

[tropic scarab]

Ok, this is nice to hear. Ai generated this image based on the text prompt.

[languid nebula]

That is super nice actually!

[nova owl]

waiting for the 5th batch

we need image generation on the fly for sure!!

if is that possible that could be awesome!!

[icy zenith]

5...

[humble swan]

everytime I check AI news, a new AI model released which beats X Model that got released 1 week ago...

[brittle kestrel]

u guys are ballers

crazier updates day by day

💯

[nova owl]

this gadget will hit hard!

[signal hamlet]

Read more about Claude 3 here:
https://www.perplexity.ai/search/Claude-3-Overview-dFyjhctzQyWXmgCgSPHJMg

[robust quartz]

100%

[long patrol]

For those pain for the pro version of harpa and Claude pro, is there any way to be able to use? Claude opus with the software?

[sweet jasper]

With pro you get 5 opus usds

Uses

[long patrol]

Even if we're paying for a pro on both platforms?

[tame current]

Switching over to Claude 3, I didn't even realize Perplexity also switched over to GPT-4 Turbo for their OAI model

Sick

Honestly this is way better than getting the latest Google model

Getting Gemini, and then Claude 3 in such quick succession though has been crazy

[icy zenith]

mfw we pay $20 for 5 opus uses when we can pay $20 for 200+ opus uses every 8 hours

not complaining but

im totally complaining

[sleek tapir]

well perplexity's product is an AI enhanced search engine

[icy zenith]

WELL

You are out of free messages until 3 PM

[sleek tapir]

If you want to just use claude's text generation, pay for claude's text generator :p

[icy zenith]

nvm about claude

[tame current]

You're paying $20 for 5 language models, 3 image models, and an AI-powered search that's arguably better than what Microsoft and Google are offering right now

[sleek tapir]

yeah the other models are cool, but for search - default is fast

[icy zenith]

well it's not like we use multiple models at once

[sleek tapir]

I usually uset he other models for rewrites, not the default

[tame current]

No, but you can rotate between them, for free

[icy zenith]

i find default and experimental not all that accurate

[tame current]

Before perplexity your only option to use multiple was to individually purchase a subscription to each model, which is expensive

[icy zenith]

mistral large is fine, claude 2.1 was okay, gpt-4 t is the king

[sleek tapir]

Huh, I think it's great for search

[icy zenith]

or WAS the king

[minor shard]

Sorry, I read this somewhere but can't find it now. When you get the email for the 200$ credit for perplexity, how long do you have before you have to use it?

[icy zenith]

claude 3 opus from my testing earlier today

is super good at search

while being pretty fast

[tame current]

I will also say, Perplexity probably doesn't have the infrastructure (yet) to run Opus to the point where there isn't limiters on prompts

[icy zenith]

AND smarter than gpt-4t

[halcyon coral]

The code doesn't expire

[icy zenith]

wdym run

they use anthropic's (or amazon/google's, idk what they use tbh) API

[halcyon coral]

Claude is not yet offered in large quantities, mainly because of Anthropic, but the number of uses will probably increase over time 🚀

[tame current]

Ah, I see

[snow briar]

How do you know when it searches with opus then it switches to sonnet?

[icy zenith]

itll say what the model is

[tame current]

5 opus uses is sadge

[snow briar]

Is sonnet comparable to gpt4?

[icy zenith]

not really