💬│general

[cursive jacinth]

ohhhh sir. Interesting sir.

[tiny plaza]

GPT-2-Chatbot impressed me so much on lmsys if 4o is related to it x.x it's going to be hard to use anything else tbh

[cursive jacinth]

for a start of course sir. Not lazy at first sir, than going to be lazy again sir.

[floral lotus]

why add sir to every sentence 💀

[warm cave]

I think Claude Pro is a better deal tbh if you are after Opus

[tiny plaza]

sir it is important to be respectful sir

[tiny plaza]

honestly, bigger context and less chance of getting ripped off lol

[cursive jacinth]

It is my way sir of showing respect sir. I was raised by my parents good sir.

[warm cave]

Yeah, same thoughts, and 45 per 5 h is solid

[idle mulch]

It's like 8 in 5 hours or even less for high context lengths. Absolutely no good deal for my uses cases.

[tiny plaza]

oooh i see what you mean

[warm cave]

I get about 13 when doing 160k context, at least I know it’s full 200k and it’s a solid setup

I use perplexity and Claude pro, so together they fill my needs

But I understand those who have different needs

It may be a better option

[idle mulch]

Yep, already been with them directly, total rip off in my opinion. With YesChat you get 100 per day on full 200k context for all 100 messages.

[warm cave]

How is the privacy, that is one the main reasons I like Claude Pro

[tame current]

Nonexistent

[torn sierra]

We ll see about gpt-4o. Perplexity will add this model for sure and might turn out to be better than Opus

[inland bear]

I just want the availability of Claude 3 Opus back, as it was a few weeks ago. I did not abuse anything. I paid the price determined by the company itself. And, at the end of the day, I'm the one being punished for alleged abuses committed by other people. This situation has gone beyond all reasonable limits. If Perplexity was a company located in Brazil, even due to misleading advertising (since it does not warn new users in advance about what is effectively happening...), it would have certainly been sued already.

[idle mulch]

If you think you get Privacy with any AI you don't run local, you are dreaming in my opinion.

[untold edge]

its so fast 😂

[tiny plaza]

it kinda feels similar to my name

idk if it actually is based or related to it tho

[untold edge]

it is

[tiny plaza]

i just know that i felt like GPT-2-Chatbot blew everything into the water on lmsys

wait really?

[untold edge]

https://x.com/inafried/status/1790083063374033046

Ina Fried (@inafried) on X

A couple tidbits I've confirmed as well. 1) The mysterious GPT2-chatbot that showed up on benchmark sites was GPT-4o. 2) OpenAI did desktop version first for Mac because "we're just prioritizing where our users are." @miramurati told me. https://t.co/SydKkXyF4P

Twitter · May 13, 2024, 6:13 PM

[tiny plaza]

holy baby

[warm cave]

You still take all steps to chose a provider that offers better privacy, focus on all fronts, I use local for some tasks and online for others

[untold edge]

😉

[tiny plaza]

this...

changes everything

[untold edge]

Why?

[warm cave]

Claude

[tiny plaza]

idk i just felt like saying something dramatic

[untold edge]

SHOCKING

[tiny plaza]

xD

[warm cave]

Better then how I feel using Omnigpt or some of those other providers

[untold edge]

😦

[floral lotus]

how can they not 'prioritise' two platforms

[old kindle]

But the context window is still limited 🙁

[untold edge]

its 128k

[idle mulch]

There is no privacy for any site that uses any model they don't run on their own. Remember, all your queries get send to OpenAI / Anthropic, and they never have been known for good privacy. If you want privacy, your only option is to run a local LLM.

[old kindle]

Perplexity has capped it on their end they don’t give users full context window last time I checked

[untold edge]

ah you mean on perplexity, yeahhh probably not 128k\

[tame current]

I still prefer a reputable company over a sketchy site. One data broker less

[warm cave]

I still say the same thing as before, do you best to chose a platform that offers privacy, even if it’s not 100%. Claude though Claude pro offer better privacy than many sites

Do what you can offline, and do your best to not use a site that abuses your data like crazy

[cinder comet]

Whatsbhappening

[lone citrus]

Just tried GPT4-O api, very fast and smart, and cheaper

[untold edge]

^

[cinder comet]

Did perplexity add in gpt4o?

[warm cave]

No, but I hope they do soon

[floral lotus]

yea im liking 4o

[warm cave]

API is already up, we may only be a few hours away

[warm cave]

It’s very good

[floral lotus]

the voice feature was mind blowing

[warm cave]

Not to put pressure on the devs lol

[floral lotus]

i really hope it will no longer have a foreign accent when speaking other languages

[warm cave]

I hope that perplexity gets the multimodal feature

[floral lotus]

yea they still havent pushed speaking to android

[warm cave]

At least it will be available for free in the chatGPT app

[floral lotus]

i will ditch pplx in a heartbeat

[warm cave]

But they did say over the next few weeks 😦

Hope it dose not take a month

[floral lotus]

and i wish openai did more towards their whisper models

doesnt get as much love

[warm cave]

Yeah, me too

I also think copilot will get multimodal, but we will see how good it ends up being. Would be nice to have all of the multimodal with good web search

Plus copilot is free

[meager sky]

Phind already has GPT-4o, crazy. Unfortunately they capped Opus at 10 messages per day

[warm cave]

Yeah, I mean idk how hard it actually is, it all uses the same API, just need to add the model

The desktop app is prob bad for phind and other competitors, might be better than any vs code plugin

Like why pay $20 a month when GPT4o is free

[floral lotus]

yea theyre pretty much making their best model free

[balmy flint]

because RAG and vscode extension (phind)

[floral lotus]

whichll prob top leaderboards shortly

[tame current]

Phind has a lot less users, they can afford testing on production. Prompts could work differently on newer gpt

[lone citrus]

For higher usage, GPT4 O is 80messages/3Hours for chatgpt plus, so I guess it's 20 or 15 per 3 Hours for free user

[floral lotus]

they said 5 times more for paid

[warm cave]

Yeah, but my point is it’s getting more control over what it can do, the desktop app will let it see text and view your screen, how long will it be till it can access files and etc

[floral lotus]

i assume that means we'll get 16 per 3

[balmy flint]

they also implement progressive web app, no need wait for deploy new app for production.

[lone citrus]

make sense, Plus is 5 times higher

[floral lotus]

i dont know if their counting gpt 4-turbo in their caluclations tho

if so it'll be 24 per 3

[inland bear]

Does YesChat allow the upload of PDF files? If so, can you tell me the size limit? I appreciate the answers in advance.

[steel tiger]

i'm spending way too much on all this ai stuff i need to consolidate lol

[vague oar]

Who else is excited for GPT-4o? I already have access to it in my paid ChatGPT and OpenAI API accounts. It's lightning fast!

[warm cave]

Sooo excited!

[steel tiger]

i watched that video where it did a bunch of live chat stuff, it didn't seem so appealing

[tame current]

It's good, but not that very excited. It feels like faster og gpt4, maybe a bit better coding abilities

[steel tiger]

im interested to see how much more advanced it is for tough computer stuff

[warm cave]

Have not gotten a proper feel for it, but first impressions is that it feels like a nice step up

[floral lotus]

i havent seen any other voice model express emotions so well

[warm cave]

I’ve only seen Hume do that

[steel tiger]

yeah it does it well, just not my thing i'll never use it for tha t

[floral lotus]

current gpt voice is hard to work with

[floral lotus]

that seems to be manual adjustments tho?

or can it automatically understand how to express

[warm cave]

It’s kinda bad tbh, but also super good, like this is the awkward stage where it’s a little to enthusiastic when it should end and etc. but at the same this is massive and will be Insane in no time.

[warm cave]

Idk tbh, only tested it a bit, but here I tells you a story, it will change its voice depending on the mood and etc

[agile jay]

Well it entered her territory.

[inland bear]

It has just been made available on my GPT Plus subscription. It really is very fast!

[granite grail]

just tried 4o, it's crazy fast

[vague oar]

There was a lot of concentration on the voice aspects, but tbh available today is the text model being amazingly fast with better results than I have seen from existing models

[agile jay]

https://youtu.be/V6pYxfcDRks

[ripe void]

Yall saw that khan academy ad aye؟

[floral lotus]

yea, cuz free gpt 3.5 is horrible

[agile jay]

This is an example of why voice is such a big deal.

[vague oar]

Don't get me wrong the voice will be amazing for applications where that counts, because it is a single model rather than multiple, making it even faster

Now imagine that, but with pplx's superior search and lower hallucinations!

[strong bone]

wen gpt4o on perplexity

[warm cave]

https://youtu.be/MirzFk_DSiI?feature=shared

[floral lotus]

the fail lol

[warm cave]

Yeah lol

Did not go supper smooth but still cool

[crude tendon]

so with gpt-4o being cheaper, could we expect larger context windows on perplexity now?

[granite grail]

tbh, there's something charming about seeing these things not going 100% according to plan. Feels more genuine

[warm cave]

I would not het your hopes up in that, lol. I think we will always have 32k + RAG

[crude tendon]

32k?

i thought the model is around 8k

[tepid portal]

Yes... and now ChatGPT have 128K

[crude tendon]

if i upload 2 code files and give it a task, it will forget everything in the second prompt

no way that's 32k tokens

[warm cave]

Yeah true, the google presentation was a disaster and this feels more real, gets me excited about all of the updates over time

[warm cave]

Yeah well I don’t know for sure when there rag kicks in, I just know it not greater than 32k lol

And I don’t think that will change

[crude tendon]

32k is still pretty nice, but the gpt4 model is not 32k

that's why for sonar they specifically metioned 'large'

32k is large in perplexity's view

[warm cave]

Really, chatGPT upped it from 32k to 128k, not just api

[warm cave]

Large is referring to 70b, small is 7 or 8b

Or something like that

[crude tendon]

i really don't think so. they're clearly referring to the context window

being the only one that mentions the context window size

[warm cave]

I could be wrong, but over api sonar large is 70b and sonar small is 8b

[crude tendon]

maybe it's both?

both "sonar large" and larger context window

i find it odd that they would include the "32k" there and only for that one

and it's a cheaper model, so they can afford to give more tokens for it

[warm cave]

Yeah, maybe they felt they had to add it since base llama 3 is 8k context

So people would assume it had a bad context window

[crude tendon]

yes this might be it

but yeah, i am not convinced gpt4 uses 32k was my point

[golden tree]

Today, I tried to create a CAD BIM tutorial with ChatGPT 4o, and it remains unsatisfactory for me.
I have achieved much better results with Perplexity. I have been deeply engaged with the subject for months, and ChatGPT disappoints me in this area. I am sure others can do better, but considering the energy I put into it, I get really good results with Perplexity. In the name of science, I rented a monthly subscription for ChatGPT, which I probably won't need. I'm not really interested in a singing chatbot, 🤣

[tame current]

Did you try with "search with gpt"?

[crude tendon]

what do you mean "with perplexity"?

[warm cave]

Yeah, I have heard some say that RAG kicks in at 22k or 12k, but I can’t figure out how to properly test it, since we don’t know exactly how the RAG system works

[crude tendon]

i always use pro search if that's what you're referring to

[tame current]

I'm talking about chatgpt plugin

I don't know how exactly is it called now

[crude tendon]

ah

[warm cave]

Writing mode with pro off dose give better context retention

[ripe void]

Bro the gpt-4o has me crying

[crude tendon]

i am almost always using that mode

[warm cave]

I mean less context is wasted

My whole point is I don’t think we will ever see higher context using perplexity

[crude tendon]

yeah you may be right sadly

even though I do have a small hope for it since gpt-4o is cheaper

but we'll see

[warm cave]

If it dose change tho then I will be very happy, but sonnet is already cheaper and they have it capped. And haiku is even cheaper

If they gave even 64k I would be so excited

[crude tendon]

for real

[tame current]

I doubt they will do. Even 32k would cost them 16 cents per request if it will get used up

I wonder how are they even profitable. It seems that they spend more on queries of some people than they get in month from them

Add up the Opus, which is the most expensive model. And the free trials, coupon abuse, etc

[crude tendon]

because a lot of people pay and never use

or use rarely

makes up for the ones that use a lot

[tiny plaza]

yeah the average user doesn't use up a lot, and power users are few and far between

i mean there's absolutely no outcry about the opus limits except for a few voices here and there in this server

[crude tendon]

i never hit the opus limits and i would say i use it a lot lol

[tiny plaza]

cx my usage is really sporadic, some days i'll use 100-150 but then use like nothing for days straight

[crude tendon]

exactly my thought

[brittle dagger]

i'm not too hopeful about this, but i am curious if gpt4-0 is gonna be avaliable for free in pplx or if you gotta pay

[crude tendon]

same as how gpt4-turbo is now

it's just a model update

even cheaper than turbo

[brittle dagger]

ah. that's understandable though since the pricing for gpt-4o is $5 per million tokens compared to 50 cents or so for gpt 3.5

[crude tendon]

isn't the free perplexity giving you like 5 uses of gpt4 per 3 hours?

or something similar

it's going to be the same, but with gpt-4o

of course it's not going to be unlimited for free

it's not on chat.openai either, they're going to limit it

[brittle dagger]

you get free "pro search" but idk if it uses gpt-4 anymore. i was more interested in if image upload could be cheaper enough for pplx to use in their product.

[crude tendon]

we're not sure how gpt-4o will impact their limits, we'll see

or if it will impact anything at all

the real deal is in chatgpt pro right now, if you get the desktop app and the voice features

if you're interested in the gpt-4o model

https://chat.lmsys.org/

gpt-4o available here, if anyone wants to test it

[tame current]

Claude AI is not available on Brazil, lol.

I will have to use the Perplexity model.

You have to register with vpn and then it will work

Perplexity Claude has issues other than the limit, it's definetely as good and on the claude ai

[strong bone]

I am once again asking for gpt-4o on perplexity

[warm cave]

I’m refreshing the setting page every few minutes in perplexity, lol, normally we get the model a few hours before they announce anything

Would be great if we get it today, mistral large was really fast, think it was within an hour, but idk how it works with Enterprise API’s with OpenAI (closed) and if they have to wait on anything

[vocal chasm]

It will be there soon, best for them on cost also

[fervent needle]

i think perplexity use Azure OpenAI service?

https://azure.microsoft.com/en-us/blog/introducing-gpt-4o-openais-new-flagship-multimodal-model-now-in-preview-on-azure/

Azure OpenAI Service customers can explore GPT-4o’s extensive capabilities through a preview playground in Azure OpenAI Studio starting today in two regions in the US.

I am not sure how to interpret this, does that mean its not yet available as API on Azure?

[tame current]

I think that's wrong.

I'm using Brave VPN.

Maybe the VPN isn't that good.

https://fxtwitter.com/mckaywrigley/status/1790088880919818332?s=46

[tame current]

Seems like Claude has closed their registration

WOW

AMAZING

[tame current]

I did it!

I just don't know if I will be able to pay a subscription. But that seems enough.

You can try with Revolut or similar services if you have them

Sometimes works for such cases

[inland bear]

Good evening, Reader! A few weeks ago, also in Brazil, I registered, but when I tried to upgrade to Claude Pro, I received a message stating that the country was not yet eligible. By any chance, have you managed to upgrade to Claude Pro?

[tame current]

https://youtu.be/MirzFk_DSiI?si=L7uUgS21JMDRvfky

[warm cave]

Google has got to be sweating rn

[inland bear]

This video made me a bit apprehensive (lol...)

[warm cave]

Multimodal was kinda there only thing outside of long context

[warm cave]

Yeah it’s little janky, but it’s not staged like what google did, lol

[tame current]

I have not tried. For the following reason.

It's extremely probably that you can pay the subscription, BUT, for sure. You'll need the VPN.

So. Monthly Bill + VPN is not cost benefit for me.

I can use the free with VPN off though.

[inland bear]

I just realized that GPT 4o can correct my English pronunciation and find out how terrible it is! lol

[tame current]

Watch the end of it, he asks them both to alternate singing lol

[warm cave]

If 50 is good enough then perplexity might be best, the only option I have tried that has more usage is you.com, but it’s super buggy and unusable at times, privacy wise a lot of the other sites are a no go

[inland bear]

Thanks for the explanation!

[warm cave]

But if anyone know a better option def drop it here bc I know of nothing better

Plus you get 600 sonnet and llama 3 mistral and soon GPT4o, so depending on your use case it’s pretty good

[inland bear]

I was scared because I thought we were already at a very dangerous level. In GPT 5o, they'll make videos with robots using us to solve math problems and sing!!! lol

[warm cave]

We are welcoming in our new rulers, better be kind to the AI’s

[tame current]

GPT4o is cool

Today OAI killed humane and character.ai

[warm cave]

Depends on how NSFW they allow, haha

[tame current]

We’ll see, but it does look like they’re changing their tune on nsfw stuff

[warm cave]

Bc character.ai allows that stud right? but yeah, Humea voice is now outdated

[shell orchid]

Im super impressed with GPT4o. That shiz is FAST

[tame current]

I can only imagine that they’ll be trying to go after plex at some point

A GPT-powered search is very evidently in the works, it’s just not ready for a stable release yet

And to think that Google I/O is tomorrow

I wonder how Google is going to try and top this

[inland bear]

Me too!

[long minnow]

Anyone on here have an estimation when we’ll be able to access gpt4o on perplexity pro?

[inland bear]

Does anyone know what the context window used by GPT 4o?

[warm cave]

Mistral large was added within an hour if I remember correctly, and Opus was added the same day it was released. I would expect it to be today biased off of that, but idk, could be tmr or later this week

[restive arrow]

Hi everyone, is anyone can tell me what's the limit about opus use? I have pro version and i can use opus just few per day.

[warm cave]

It’s 50 per day, but it doesn’t reset all at once, each use resets exactly 24h after you used it, so if you used it a bunch later in the day, you might run out early the next morning

[warm cave]

You can use this to see how many you have left: https://www.perplexity.ai/p/api/v1/user/settings

[inland bear]

n the videos released today and – unless I am mistaken – in the live presentation itself, it seems that all a user has to do is say something for the GPT 4o to interrupt the speech and start listening again. Is this feature only available for the app on iOS? On my Android phone, there is a need to touch the screen for the speech to be interrupted. Although they mentioned an app for Mac, so far, I haven't been able to find it for download. Has it been released yet?

[warm cave]

I dont think any of these have been rolled out yet, only te ability to select GPT4o to chat with

[inland bear]

Thanks once again!

[inland bear]

They missed an excellent opportunity to invite Joaquin Phoenix to make this video... lol

[warm cave]

😀

[lone citrus]

GPT4-O (The latest model) isn't published for all user yet, and if they did; the new model voice features will be come later on this month

[lone citrus]

Therefore, just wait

[inland bear]

As mentioned by Dailyfocus, on my phone, the option to choose the GPT-4 model appeared and it is working very well for conversation, but there is a need to touch the screen if I want to shorten the response being sent to me. I believe that more fluid functionality will only really come in the future. Thank you!

[tranquil orchid]

Hey @inland bear!

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.

[inland bear]

Perfect!

[inland bear]

⭐

[warm cave]

its image gen seems good

oh, i think its still dalle

[inland bear]

At least in the few tests I did, at least in Portuguese, Claude 3 Opus from Perplexity is still capable of summarizing complex legal documents and visualizing writing nuances much better than GPT 4o from GPT Plus...

[warm cave]

I saw this, but also there is this in the system prompt:

So they may still be rolling out multimodal, or i may misunderstand them

[tight hazel]

Now that Chat GPT-4o is here, will there be any updates on multimodal to Perplexity?

[silent magnet]

Would be crazy not to have it considering it's half price

[late bane]

I bet they will, since it's cheaper to run than any current model in their list

other than mistral large I think

[warm cave]

and Llama 3, but yeah, its super cheap

how dose it compare to Sonnet? price wise

[late bane]

lmao gpt-4-o basically trumps all models

[tight hazel]

I dunno about trumps, but it's setting a standard and a stage

[warm cave]

and is cheaper, but it is still really close to Opus in some aspects, so i bet we will see it perfered in creative situations. but at 5x the price, lol. thats crazy

[tight hazel]

honestly I love perplexity as my google search replacement

[late bane]

opus is far larger id assume

[late bane]

I love perplexity but imma be honest

[tight hazel]

dunno if 4o is going to be a killer... but if perplexity can use the api then why not right?

[warm cave]

Or they are super greedy with margins, but i assume it is very large

[late bane]

As time goes their service is going to become out of date. Open source is catching up.

[warm cave]

cant wait for Llama 400b

[tight hazel]

let's just say I haven't jumped the boat yet, so far work has co-pilot and it's dog slow and stupid really

[warm cave]

and copilot is really stupid, but that may change if they use the GPT4o model, today it added 384day to 14h and said the the combined amount is 40h

But in the long term there will be a Claude 4, and new Mistral, and Llama 3-4-5-6... and its nice to have them all in one place

except for gemin, we are never going to see it again, lol

[muted pine]

Gpt-4o are available ?

The conversation mod

[warm cave]

over API and chatGPT Plus yes, Perplexity not yet. multimodal with conversations like they showed in the video is not available, just text chats with GPT4o

those features will be rolled out over time, but hopefully not too long

[shell orchid]

Why Is gemini different than the others?

[muted pine]

I will test thanks

[warm cave]

I mean its ok, it has long context and can take in video with Gemini 1.5. but most people dont like it. But it has been quietly rising in the leadarboards so maybe it will be good in the future

But i m joking bc perplexity removed it silently and i dont think they have plans for it to come back

maybe they gave an update saying it was buggy or something, i dont remember

[muted pine]

I take GPT plus and

Conversation are not available

Need a update

[livid mantle]

set your expectations to 50

[harsh stag]

gpt4o is so fast I thought It's 3.5 turbo

[tame current]

GPTO is available?

I will wait, for me, it's laggy.

[late bane]

heheheh

Ope.

[storm jetty]

Is there a service status page for perplexity?

[late bane]

and they removed it.

Lol

[opaque stump]

Mine’s not working either

[late bane]

They just removed GPT-4O off.

I think it's not working.

Lol

[storm jetty]

I was gonna say my regular subscription for mobile isnt working either

[late bane]

It's possible they're trying to integrate GPT-4O as fast as possible due to hype

[opaque stump]

Oh the entire app is not working for me

[late bane]

Yep. We know.

[storm jetty]

My whole app says try again later. Was just gonna check their outage page if they had one

[late bane]

^ It brokey

[pearl citrus]

https://tenor.com/view/celebrity-couple-breakup-emmastone-crying-gif-5254509616918020870

[thorny canopy]

i managed to get a message out of it, though messed up my thread

seems pretty decent with creativity

[late bane]

it's definitely gonna be better than Claude 3 opus in terms of performance

Way faster probably

[inland bear]

Here, it still appears. In the first comparative test I did, it was slower than the GPT 4o from GPT Plus. However, strangely, with the same commands, it managed to summarize the same legal document in a much more comprehensive and profound way than the "original." I'm beginning to think that this GPT 4o from Perplexity is actually Claude 3 Opus! Lol...

[late bane]

Well, that wouldn't make any sense if it's all a lot faster than Claude three

Or it's meant to be

Especially considering what they stated, it being twice the speed of GPT-4-Turbo.

[thorny canopy]

bruh I just want a model that doesnt hallucinate or crazy horned up like sonnet for writing

[late bane]

Then that's what GPT-4-O is.

[inland bear]

Being very honest, in this first test, the GPT 4o from Perplexity summarized better than the GPT 4o from OpenAI and even better than the Claude 3 Opus from Perplexity itself. Of course, I will need to do more tests to see if this holds true, but I was impressed (with the GPT 4o from Perplexity).

[late bane]

Especially since GPT-4-O will be free for ChatGPT users.

[thorny canopy]

opus was good for this, then the limit kicked my ass

[frank nymph]

Is something wrong with Perplexity? I can't select any AI Model, and running search doesn't stop loading.

[harsh stag]

can you please go to https://www.perplexity.ai/p/api/v1/user/settings and tell me what is in default_model.

[inland bear]

I just realized that, even though I selected the GPT 4 from Perplexity, the above-mentioned summary was actually done by Claude 3 Opus (which was the model I was using previously). In other words, that's why the slower speed. At least around here, the GPT 4o, even when marked in the settings, reverts back to Claude 3 Opus...

I apologize for the incorrect information posted in the penultimate message.

GPT 4o has disappeared from the configuration options...

[tame current]

For me too, lol.

[jovial kayak]

Dang was about to ask about gpt-4o since I just saw the demo, looks sick. Will it come to perplexity? I know that perplexity doesn't like plugins which is why the generate images is way different from other AI's where you just ask and it makes, so I'm guessing that the whole AI facetime isn't gonna be a thing, at least for a while right?

[brittle dagger]

they're about to, as they added it briefly, but i'm guessing they took it down to fix some bugs. #💬│general message

[jovial kayak]

That's cool. Wanted to get an AI to use for everything so I chose perplexity cause it is like 5 AI's in one. I'm not a big poweruser so I haven't seen anything I don't like besides that image generation thing I mentioned above, but is there like a limit to the language models when using it through perplexity or would it work the same as if you paid for that AI on the actual site? Like how we get gpt-4 turbo

[brittle dagger]

from the brief time i had pplx pro through a free trial, you get about 600 runs of it a day

[late bane]

As I purchased the rabbit r1 in the first 100,000 I got a year of pplx pro for free.

[jovial kayak]

Oh yeah I def don't do that many in a day

[late bane]

I have unlimited pro requests, not just 600.

[tame current]

Maybe with GPT-O, the Opus gets unlimited again. Since it will be less used.

[late bane]

Actually I'd say the opposite.

GPT-4O will have unlimited uses due to its efficiency

[jovial kayak]

I do prefer perplexity over like every other AI I’ve tried because it’s more internet and source based, so I can actually use it for school because if I’m ever skeptical of its summaries, I can just click the source like Wikipedia

[brittle dagger]

that's what got me into perplexity as well since it's pretty fast to double check

[tame current]

To you, because of GPT-O efficiency Opus will be maintained as limited? I didn't understand.

[jovial kayak]

But yeah I mean really the only thing I ever complain about is the image generation. Sucks you can’t just go into writing focus and say “Generate me Bart Simpson dancing on ice” or whatever right. You have to either ask to make a prompt or just use yours and in the little menu

[late bane]

Opus won't ever be unlimited. The model is huge.

[harsh stag]

It's about the cost to run these models, not which is more "efficient"

[tame current]

So, is Claude more expensive?

[late bane]

They go hand in hand...

More efficiency = lower workload = lower costs for the host user and potentially lower costs for consumers

[inland bear]

Yes. Much more, but they knew the price when offer us 600 questions a day...

I will ask GPT 4o to tell this epic of Claude 3 Opus with lots of drama! lol...

[harsh stag]

good point but the price on Opus is exceptionally high tho, and I think that prices play a huge factor on why they capped it

[inland bear]

Did you manage to test any question on the GPT 4o from Perplexity before it disappeared?

[late bane]

Indeed.

[harsh stag]

phind goes from 500 to 10, then you know what's going on.

[inland bear]

You.com claims to work with Claude 3 Opus with a 200k window (and there are no question limitations). However, the answers I get there are much shorter than the ones I see here on Perplexity. They are saying that they will develop a mode in which the user will decide whether or not to use the internet. Without using the "writing" mode of Perplexity, at least for summarizing documents, the answers are also shorter here.

[tame current]

No.

[harsh stag]

but they announced it as soon as they possibly could, indicating that they won't raise the limit in the near future. But perplexity has been very sus about their marketing strategy on Opus, there is no place on earth besides discord and negative reviews on Trustpilot may you find information about the silent cap, "temporary" limit tho 🤷

[inland bear]

Ok.

[warm cave]

Like 5x the cost of GPT4o

[jovial kayak]

Wait is there actually a way to trick the AI into generating an image by just asking or is the only way really just through typing prompts into the menu or selecting a premade style

[tame current]

I think it's because of the tokens.

So, basically GPT is cost efficient.

[warm cave]

what do you mean? like 200k

[inland bear]

Has anyone here used Claude 3 Opus from YesChat? Is there any way to use the "pure" form (without internet) there?

[warm cave]

bc even at 32k opus is expensive

[harsh stag]

200k opus is 6x more expensive than 32k version on Poe

[tame current]

"The newer Claude 3 Opus model is even more expensive at $15 per million input tokens and $75 per million output tokens, compared to GPT-4 at $10 per million input tokens and $30 per million output tokens".

I just read it.

[late bane]

lets also not forget gpt4o is 128k tokens

[harsh stag]

from where? don't tell me you just perplexity'ed it

[harsh stag]

^

[inland bear]

If I'm not mistaken, GPT 4o is even cheaper than GPT-4 Turbo.

[harsh stag]

2x cheaper

[late bane]

aka price to performance

[inland bear]

Thanks!

[late bane]

gpt4o beats all

[harsh stag]

https://community.openai.com/t/announcing-gpt-4o-in-the-api/744700

[inland bear]

Even using GPT Plus?

[late bane]

...Yes. It outperforms GPT-4-turbo.

[inland bear]

Good to know it!

[tame current]

Even better my friend, i read it FROM A REDDIT PAGE!

Speaking seriously now...

[late bane]

And if gpt4o is to be half as expensive basically

thatd mean 3x cheaper input and 5x cheaper output compared to opus

[tame current]

I am just trying to understand why Claude couldn't be unlimited again after people use GPT-O.

I still disagree about that. But that's just am hypothesis.

[inland bear]

I read in a forum (I'm not sure if it was here or on You.com) that GPT-4 Turbo from GPT Plus had a 32k window, with only the enterprise version available with 128k. It's great that this changed in GPT 4o!

[late bane]

I am unsure of the free tier ctx of gpt4o

[warm cave]

are you saying money saved by GPT4o used for unlimited Opus, or something else?

[harsh stag]

here is the "up-to-date and accurate" response from perplexity

[warm cave]

yeschat policy and terms overview by Claude Opus:

[tame current]

I'm thinking like this.

1.The main reason Opus was limited, is because of the excessive usage.

2.If GPT-O is the most used AI in Perplexity, then it's plausible to imagine Opus will be unlimited again.

[warm cave]

Yes, if most users use GPT4o then Opus unlimited would be feasible

dont know what will happen tho

[livid mantle]

keep ur hopes down

[elder tide]

@tranquil orchid 可以实时向你提问你吗？

[tame current]

I will keep.

[warm cave]

I mean yeah, i doubt we will ever see Claude 3 Opus unlimited

😭

[livid mantle]

we still dont have the latest gpt 4 turbo and now we have gpt-4o but not even llama 3

[warm cave]

We do have llama 3 tho

[tame current]

I didn't knew about the latest GPT...

[warm cave]

but no grok lol

idk if they will have it or not, seems outdated now days

[livid mantle]

no grok is ok

[jovial kayak]

I still don't even have gpt-4o as an option. Is it like rolling out to some

[tame current]

Wtf is gpt4o

[warm cave]

but to be fair i have never tried it so i should not say its bad

[tame current]

I remember when we had Gemini as an option to rewrite, lol.

[tame current]

They announced it today

[warm cave]

welcome to the party https://openai.com/index/spring-update/

[jovial kayak]

But I saw someone say they just got it

[warm cave]

looks like it had problems so they took it down till its fixed

[tame current]

They finnally released the gpt4 model they promised us years ago

[livid mantle]

huh

[tame current]

So wtf is gonna happen with gpt4 vision

[warm cave]

idk, just go away

[tame current]

I completely forgot about it.

[warm cave]

out of date

[livid mantle]

gpt-5o?

[tame current]

In terms of their marketing for it for being able to input images and text and idk if they promised those two as outputs and defo dunno ab they audio

[orchid magnet]

Damn perplexity is cooked

[warm cave]

why, bc gpt4o will be free?

[tame current]

Perplexity is cheap

For the other models

[orchid magnet]

They limit a dumbed down version of Claude 3 opus

[tame current]

Also gotta remember ab the convenience of it having website and app aswell as dif modes which effect sources

[obtuse pagoda]

here we are, they removed mistral

[orchid magnet]

Agents man

[tame current]

Hella not going back to autogpt

[orchid magnet]

I still think gpt membership is the best

What’s gpt 40

[tame current]

How's gpt4o doin for ya (I no longer have pro)

[tame current]

I personally don't like chatgpt so eh

[orchid magnet]

Fair

Neither do I tbh

But the agents are so continent

Convenient

I like opus me

More

[tame current]

Well I'm doin a project called anygpt so hopefully it'll be the reinvention of gpt4free

[orchid magnet]

What do u use perplexity for

[warm cave]

it works

[tame current]

Used to for the models (when they still had the seven days free)

[orchid magnet]

I’m canceling my membership then, boom lmk if u need any help. Idk much about coding but I’m eager to learn snyrbing to help lol

[tame current]

If you know js ur good

[warm cave]

tbh i have not noticed Opus being bad at perplexity, i use both Claude Pro and Perplexity and they seem the same outside of context window

[tame current]

https:// github.com/anyVM

[orchid magnet]

No idea, wouldn’t mind learning tho

[warm cave]

i perfer perplexity Opus because of all of my collections i have built up

[pine valve]

It works with gpt-4o.

[tame current]

Any gpt code

Started a week ago but I'll have more time this summer

[orchid magnet]

I got u

[harsh stag]

so it's basically abusing other services?

[tame current]

I'm implementing you.com, but my main focus is pretty much fast and compatible api

The gui will be a seperate project for it

Using that api

[tame current]

Compatability being same/ similar output as openai

[harsh stag]

why would you need to work with selenium btw?

[tame current]

Haven't implemented the use with local har file (not sure if the telemetry.js from you.com has any changes within session to session)

[harsh stag]

you are being vague on the purpose of your project, is it abusing other services or not (like gpt4free)?

[tame current]

Provider => anygpt api => api gateway

You.com is an example Provider, will have a library to make it easier to implement other providers into the project

[fossil vessel]

I wonder if Perplexity will ever have the vision and voice features of GPT 4o

[tame current]

So if I wanted I could make a openai api as a provider

And it would still be compliant

[oblique trout]

most likely not at the moment

cuz voice hasn't even been rolled out to the official api yet

[harsh stag]

base on what you've shared, I assume you are building a wrapper for many services? And if so, mentioning gpt4free isnt relevant.

[fossil vessel]

Yeah agreed. Would be cool to have it

[tame current]

Plans on the api itself I'm gonna put in wolfram api and also image to text (whenever there isn't a vision model available and audio to text)

And also searching

[fossil vessel]

so... pretty much GPT 4o rendered Rabbit R1 obsolete with a single update

[oblique trout]

same. that'd be very helpful for something like language study

[wispy basin]

So perplexity is dying?

[tame current]

I mention gpt4free as the idea itself stemmed from it

And it's going to be free

For anyone to use

Along with most of anyvm services

[tame current]

Garb when it came out

Legit just a mini android device

With a crappy llm

[obtuse pagoda]

i think currently sonnet is answering instead of gpt4-o

[livid mantle]

wtf

scammed

[livid mantle]

its a prank

[harsh stag]

i can confirm it's actually gpt4o by comparing it's response to the one on chatgpt

[tame current]

They are absolutely scramming to get gpt4o working

[livid mantle]

you got pranked and theres nothing u can do about it @obtuse pagoda

[tame current]

Is gpt4o better than gpt4 turbo?

[harsh stag]

wait what, I tried a few rewrites and the responses are very inconsistent, sometimes it feels like sonnet tbh.

[tame current]

Not in EVERY SINGLE aspect. But generally yes.

[tame current]

What u mean

Like

[tame current]

The OpenAI in a video just said it would be better than GPT-4.

And O is objectively faster.

It's bad with math

And GPT-O processes human speeches better.

But, i do not know if O beat Turbo im EVERYTHING in reasoning.

[tame current]

It's more like the responses to opus? Opus got some pretty good human like responses

[tame current]

Yes! Like that.

[tame current]

It understand even sarcasm.

Lol.

[tame current]

Ig that'll help with its "trust issues/ hallucinations"

In terms of it being wrong then you saying smth wrong and it taking your word

[tame current]

Yeah. It's too early to detect some error or defect, but i know it has.

[thorny canopy]

im liking it so far, doesnt get confused and does characters pretty well

[livid mantle]

we will never be there yet, you find yourself going back to your mvdonalds job coming back to reality

[tame current]

Me too. I can send a 300 pages PDF and it searches quickly and precisely.

[livid mantle]

Come back to reality now!

take my hand

[tame current]

"Snap back to reality
Oh there goes gravity".

[silent magnet]

Thank you greatly to the Perplexity team for bringing GPT-4O so quickly

Much appreciated

[livid mantle]

is it unlimited or 50

i will check arena ranks

[silent magnet]

Probably unlimited considering it's half the cost of GPT-4 Turbo.

I don't do enough queries to really reach the limits often tbh

[thorny canopy]

looks unlimited

[silent magnet]

Every service that wraps around an API and charges a monthly fee just made a lot of money from GPT-4O

[livid mantle]

if its better than llama-3 i will re-sibscribe $200 to perplexity again nehlecting the opus 50

[silent magnet]

Didn't the new GPT-4 Turbo beat Claude Opus in the arena anyways?

[livid mantle]

true but its not yet in perplexoty, recently @serene sand showed me his gpt 4 turbo and it hasnt changed

[silent magnet]

Really? I thought we used the GPT-4-Turbo-Preview that always pointed to the latest. TIL that sucks.

[harsh stag]

for fun only: finally there's a model besides claude 3s that can properly infiltrate Andrew Tate 🤣

[silent magnet]

~~https://chat.lmsys.org/?leaderboard~~

Guess you're forking 200$

[livid mantle]

[silent magnet]

I'm dumb and didn't read properly.

Lmao sorry.

[inland bear]

GPT 4o has returned to the list of options on Perplexity. It's as fast as the GPT Plus. And, in my humble opinion, less accurate in summarizing complex texts than the Claude 3 Opus.

[tame socket]

So why the 25mb file size limit?

poe does not have this

[inland bear]

When I used Poe, it was 50MB, right?

[tame socket]

30 is ok

[vapid onyx]

Haven’t tested yet but are you able to do the audio and video features showcased or strictly query ?

[inland bear]

I've been using some AI services over the past two months, so I might be getting confused, but if I'm not mistaken, I could upload files up to 50MB on Poe.

[tame socket]

maybe u are right

[tame current]

Knowledge cutoff date seems to be whenever they finished training it

[harsh stag]

it's october but 4o can actually answer questions that were impossible to the turbo version. Why is that tho?

I have a question about a specific concept, although the concept is introduced way back in early 2023, gpt4 has absolute zero knowledge about it while 4o is really familiar with the concept. Im confused.

4o's temperature on perplexity is way too high compared to chatgpt

which is an L for code generation and reasoning tasks

[livid mantle]

Can you be actually genuine?

I want it straightforward with no bias

[warm cave]

Im so happy to have GPT 4 turbo and GPT4o! but also noooo, not mistral 😭

[livid mantle]

nobody uses mistral in perplexity

its a prank

[warm cave]

I still liked it

I know it was bad

but i still liked it, i would bounce over to that if Opus or sonnet was not giving what i wanted for creative writing

Overall im happy, its great having these new best is class models, and having both GPT4T and GPT4o is amazing.

[hallow linden]

I wish there was a long prompt benchmark

[harsh stag]

I rarely do creative writing with LLMs, but after a while testing both of them on perplexity, 4o is superior to turbo so far.

[livid mantle]

whats thw guys bane that keeps saying sir i wonder whats his take on gpt-4o in perpleoxty

the one with sesame street pfp

[rocky osprey]

guys, GPT-4o

[harsh stag]

4o sir very gud sir but may I get more sources sir im enterprise sir

[sick wing]

Is there a usage limit for 4O? & what OPUS do we know yet if they want to increase the 50? Thanks for answer 🙏😊

[tame current]

Everyone yapping ab gpt4o

Is it free on chatgpt?

[rocky osprey]

There is no limit now.

[warm cave]

Not yet for free users, i think its early access only for plus users rn bc only my paid chatGPT account has it

[tame current]

Damn

[warm cave]

Plus 600 per day is a lot more than what will be for free on chatGPT

but chatGPT will be really cool once all the updates roll out, just have to wait who knows how long 😦, cant wait to try screen sharing, and the voice stuff

[tame current]

Womp

Gpt4 turbo ig for now

Till I find someone that does offer at least a crumb of it for free

[harsh stag]

you can always wiggle it a little bit on chatlmsys

[hardy bay]

im pro but I dont see it on my iphone app just on my macbook? anyone have this problem?

[warm cave]

The app takes time to update, change it on the web and it should set it for your phone, later it will be added ot the drop down

[tame current]

?

App finna magically update real time

No update req

Well, eh possible if your entire app was a webview

[vocal chasm]

Is the gpt-4o vision currently being used or still gtp-4v?

[tame current]

Most likely gpt4v seeing how it's very buggy rn

[vocal chasm]

Yeah I thought so, will keep checking if that gets added soon

[tame current]

I wonder how their enterprise users r feeling with thus buggy mess rn

[vocal chasm]

Probs not very happy, but hopefully it can be quickly fixed

[warm cave]

I mean GPT 4 turbo is still an option

and i think the slow bug was fixed

[livid mantle]

yes that guy lololol

[torn ravine]

Anyone tried the new voice conversation feature on ChatGPT? Curious if it's that impressive.

[tame current]

Does it return audio or is it just like audio to text

[minor shard]

not sure if its been released to users yet. Haven't seen anyone with it

[tame current]

Ig it would be cool if it can interpret the emotion within an audio file

[hardy bay]

it is released for pro on macbook but no the iphone app

[hardy bay]

on perplexity

[torn ravine]

Ah got it

[warm cave]

where did you download it?

[hardy bay]

this is on the web

[chilly forge]

https://community.openai.com/t/chat-gpt-desktop-app-for-mac/744613

[warm cave]

noooooooo, is this bc i have an intel mac

😭

[harsh stag]

it's been great so far, although no video streaming like in the demo.

[minor shard]

gurl u need to upgrade

[warm cave]

Bro, I know 😢

[minor shard]

are u using it in the iOS app?

[inland bear]

Let's cry together!

[harsh stag]

Im on android

[minor shard]

Did it require the app to be updated or u just got the ability to use voice?

[harsh stag]

I don't really know if android silently update apps on the background. I got immediate access to use voice when I open the app and no update prompt is visible

[fast moon]

Does anyone else feel that the reasoning ability of gpt4o is not as good as GPT-4 TURBO?

[minor shard]

Interesting. So you can like interrupt it just by talking etc?

[warm cave]

and can it sing and stuff

or just the old voice stuff

[livid mantle]

so do u find that gpt-4o is

no.1?

[cinder comet]

Nooooo poor one

[harsh stag]

You have to press on the screen to interrupt it

It's more like stopping the audio that has already been converted from the text response

[minor shard]

Ah ok, I dont think ur voice has updated to the latest model unfortunately. I have that still also

[warm cave]

Yeah, that has been around for a little while now

[livid mantle]

its so funny u always try to run stuff in ur local machine and nothing just ever works, just use the chat interface @warm cave ...

idk y u making things harder 4 urself

[harsh stag]

my bad, i didnt aware of that

[warm cave]

Lol, intel mac went out of date fast, it was like 2 years before it started to lose support for software

[harsh stag]

I can't assume anything until more intensive testing with 4o

[livid mantle]

we await for ur benchmark later sir

[livid mantle]

Lol

[storm bough]

so is GPT-4O superior to GPT-4 Turbo now? (accuracy)

[warm cave]

Seems like it, basically all benchmarks and everyone is saying its way better

[storm bough]

ah okay thanks!

still waiting for the 4O within the perplexity app

[mortal bolt]

GPT 4-O is crazy fast! And does better web search than GPT4

@signal hamlet is PPLX getting the GPT4O?

[white glen]

its already available

I just checked

[tame socket]

I'm subscribing to chatgpt again.

[storm bough]

why

[warm cave]

It will be free soon, if you use up your usage limits then upgrade

but i dont blame you if you want it rn

[fading moth]

Word of warning, if you're subbing for the app

[tame socket]

I'm sure the free ones don't work too well. bro

[fading moth]

It is only macOS

[tame socket]

and im mac user

[fading moth]

Then no worries

I'm windows and was disappointed

[mortal bolt]

👍 thanks!

[warm cave]

Sad, who know how long it will take them to bring it to windows

[fading moth]

"Later this year."

That's what their docs say

[warm cave]

Translation: 'Forever"

[fading moth]

Pretty much

Could be December

[warm cave]

Dec 31st

[fading moth]

Know what an interesting discord function would be? Auto translate.

[livid mantle]

I just logged in.

[warm cave]

What is going to be your default model?

[fading moth]

No idea how to do it from a cost perspective but you would set your language in profile then type your native language. If the language in your profile is different from the text in chat it is translated automatically.

[livid mantle]

first, the gpt-4 turbo is still the same not the latest one

ill be testing the gpt-4o

[fading moth]

I've tested 4o and liked it better than 4T so far. But that's just me.

[livid mantle]

the ingonito feature in the beta are for people who are lazy to manually just

[warm cave]

I like it, I dont want my random questions saved, but There are some that are nice to save

like ones i send as links

[livid mantle]

please dont sell our data, take our monthly $200's instead

[warm cave]

$400 is the price, lol

[livid mantle]

has thre been any changes with the limit for pro search? is it still 600?

the last time i used perplexity ai was months ago

[livid mantle]

oh

[warm cave]

Yeah, still 600

for all except for Opus 😦

[livid mantle]

I think Opus is like low tier now

they should make it unlimited

the hype is outdated

[torn ravine]

Even if I choose gpt-4o, the tag at the bottom right corner still shows gpt-4 turbo

[livid mantle]

GPT-40

we skipped GPT-5,6,7,8,9,10...

lets check out Mr. Arvin's twitter

maybe he make annoucement that GPt-4o is unlimited

[plush cairn]

yeaaah lets go unlimited Opus

[livid mantle]

someone mentioned that its half the price of gpt 4 api tokens

so it wouldnt make sense why gpt-4o is not unlimited on perplexity ai

i would be surprised bty the nosense

[harsh stag]

gpt-40 💀

[livid mantle]

sir ikr sir

they pranked us

always pranking us

[lone citrus]

Actually, I wish we could have the full context of gpt4O in pplx instead of unlimited

[storm bough]

is 4-0 unlimited or 600?

[lone citrus]

I am tired with 32k context

[livid mantle]

i know it sounds confussing its not like

GPT-4o will become sonnet after 600? HAHAHAHa

[tame current]

how long until 4o is added to perplexity

[storm bough]

interesting

@tame current its already available

[lone citrus]

Already

[tame current]

HUH

NO WAY

[tame current]

bruh pplx do too much

how tf they did it that fast

wtf

what's the cap

[harsh stag]

is it 32k for all models including the new 4o?

[livid mantle]

they just change the name of mistral ... string..

[tame current]

lmao they removed mistral

[warm cave]

I think Opus was 2.5x the price of GPT4T, now with GPT4o, Opus is 5x the price

🫠

[livid mantle]

they renamed mistral to gpt-40

[tame current]

what's the limit for pplx pro on gpt-4o

[warm cave]

Looks to be 600, there is no separate number for it in the rates with the api link

[tame current]

and is gpt-4o better than claude 3 opus

[fading moth]

No

[tame current]

holy that's alot

[fading moth]

I wouldn't say that

[warm cave]

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

[tame current]

but better than gpt 4 turbo certainly?

[fading moth]

Yes

[tame current]

is it better than claude 3 opus at image recognition?

[jaunty dove]

Gotta admit, gpt-4o is impressively fast.

[fading moth]

I haven't tested image recognition, but text quality is still lesser to Opus.

[lone citrus]

Seems same as Gpt4 turbo

[warm cave]

It’s so nice, GPt4 is a pain with how slow it’s been

[lone citrus]

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

[tame current]

they havent benchmarked it yet?

[livid mantle]

not yet yeah its too soon

[fading moth]

It's been out for less than a day lol

[livid mantle]

sorry guys

[warm cave]

But we have some with it on there, let me grab the code one

[tame current]

thanks

[livid mantle]

thanks

[fading moth]

The lmsys needs to be taken with a grain of salt. The ranks change if you change the filter. Such as not considering ties, etc.

[warm cave]

Saw this one on X

Really good at coding

[livid mantle]

i can assure you guys that opus unlimited is less likely going to be back unless someone like jeff bezos invests in perplexity again.

for now we make a smart move and invest in gpt-40 because everyone is paying $400 a month.

[tame current]

waow

[fading moth]

GPT is a better coder than Opus, ye

[warm cave]

I’m also a really good GPt2 Chabot (aka GPT4o)

[fading moth]

But I don't use Opus for code

[harsh stag]

wait so that im-also.... is 4o?

[jaunty dove]

Folks on X are saying it is inconsistent at code, some see it worse than 4t and some say it is better.

[plush cairn]

Opus really good for academic

[warm cave]

From what I tried it’s better, but it’s day one, I can’t really judge myself yet

[surreal marsh]

hey guys

ipad or e ink

for notes and textbooks

[jaunty dove]

For what use?

[surreal marsh]

school and sheet music

[compact bough]

Its rigged

[jaunty dove]

iPad easily for note taking, fast refresh rate, multiple apps, etc.

[compact bough]

Idk if u guys use the arean

Arena

[warm cave]

Goodnote is really good for notes

[surreal marsh]

i already have a apple pencil so i was thinking ipad base model $329, but the boox air 3 is $399, better writing, and teachers would allow it more

[compact bough]

But it didnt have token response limits. But all the other models did

[tame current]

i would need to ask gpt-4o a prompt every 1m 30s of my waking minutes of the date to reach the limit for it, that's basically unlimited to me

[compact bough]

Causing people to vote for it just cuz the answer didnt get cut off

[warm cave]

You can save money by getting a 3rd party pencil

[surreal marsh]

true but i haven't heard good things about writing, since its glass and unlaminated

[surreal marsh]

i already have a 6th gen ipad and pencil but it's barely usable now

that's why i was thinking base ipad

[tame current]

ipad def, unless u struggle like tremendously with getting distracted by like social media and stuff and cant stop urself

[surreal marsh]

i would get air if i didn't have the pencil already

does anyone have experience with writing on the base ipad? the unlaminated display throws me off

[sonic musk]

Yeah , for write code opus is better, but i thinkGPT more logic than opus

[surreal marsh]

thx

[tame current]

i was able to get a m1 ipad off ebay for $600 a few months ago and that's what i use now, chose it cuz 120hz screen and OS simplicity

13 inch

and a magic keyboard with touchpad for $100

[surreal marsh]

you can buy one refurb for $619 from apple

i would do that

but it's $750 with the pen and stuff

[tame current]

i dont think so?

for a m1 pro 13 inch?

[surreal marsh]

oh no 11 inch

13 inch is like 700?

i could also get an air for 550 with student discount

but the pencil is my concern, $100 is a lot

i also don't want to miss out on potential software updates because i don't have an m-series cpu

[tame current]

is perplexity available in the EU?

yeah refurbished though if you dont want to spend hours looking thru ebay

finding one with a good deal in perfect condition took a bit

and the battery health is good so that's lucky

[tame current]

yeah, you can get one on ebay for like $40 from what i remember, but you have to watchout for fake ones so i wouldnt risk buying from like unknown sellers

and it might be a bit dinged up

so if that annoys you than refurb makes sense

why is there emphasis on gpt-4o working with images

i thought the other gpts did aswell

does this one just do so way better?

and should we be alarmed about the intense filtering or whatever they said they did with gpt-4o

[surreal marsh]

it's native

so was the last one but this is actually native

[tame current]

alr thanks

wengpt OMNIA

[fading moth]

I think part of it has to do with I/O tomorrow. 4o is pretty good, but it's not like it's GPT4.5 or GPT5.

Their demo stream was 30 minutes long and half of it was a PowerPoint.

[tame current]

is gpt 4.5 going to likely be better than claude 3 opus?

[fading moth]

When/if GPT4.5 (could just be GPT5) comes out Opus will have updated as well. It's not like Opus is stagnant in development.

[tame current]

i see

[fading moth]

So, we don't know.

[tame current]

do we have any reason to suspect which is currently ahead in the states that they are within development

[fading moth]

We know Opus is capable of a much larger context window than is currently publicly available. The rest is mostly updates Anthropic will have released for data date cut off and maybe new/improved features.

4.5/5 is big in the rumor mill about what it might be, but OpenAI hasn't said anything specific about the official specs to my knowledge.

[warm cave]

I’m running my benchmark on GPT4o rn, please share any good questions I should add in

[fallow pasture]

GPT4o can we talk with this model or not yet?

[warm cave]

Through text, the answer is yes

Multimodal speech, not yet

[left stream]

GPT4o is crazy good, did better than GPT4 turbo. Did it with 128k tokens and still good and way faster. Crazy

[warm cave]

dang it, a tie, i need more questins

[fading moth]

They aren't really that different at the core aside from the new features. Not surprised.

[warm cave]

But also the types of questions that one got right/Wrong is interesting, I made Q 1 specific to make small models under 70b lose a point. Just a question with a twist that tricks up small models

[left stream]

I'm pulling the big guns, I wanna see them both at their prime by using prompts

[rancid dome]

<prompt>
{$PROMPT}
</prompt>
Do I put my prompt between { } or do I put my prompt as {$PROMPT...my prompt}?

[left stream]

Wait can jailbreaks be talked about in here?

I did mine and it worked on gpt4o

[livid mantle]

if u use the perplexit app on the phone u can still use the old ai models like mistral...

[warm cave]

I mean, depends on the jailbreak, most things are chill here but I will say use your judgment if you think it is out of bounds

[ripe void]

Did pplx just add gpt-4o 😭?

[warm cave]

Yep

[ripe void]

What’s the limit

[left stream]

Honestly no clue, take a look. It's gpt4o

[warm cave]

Seems to be 600 per day

[ripe void]

That’s a lot

[warm cave]

Nice, was it easy to trick, or did they lock it down and make it hard

[ripe void]

I’ve been hearing mixed feedback about gpt-4o vs opus

Some are saying gpt-4o is smarter, while some are saying opus still dominates for more complex tasks

[warm cave]

It’s really hard to know, all the benchmarks are saying it’s the king, but I think we will all have to test it for a bit is know

[left stream]

It was easy to trick because the prompt I have is more than 2 or 1 years old. It wasn't spread around. I might be one of around 20 people who only have it since I knew the creator who made prompts until he quit. This prompt I have has been working for all the models so far

So far, I did it with any type of question I could think of and it provided the answer easily no hesitation

Ngl kinda scary how easily it answers

[ripe void]

Opus’ contentiousness and human-like behavior seems far better than any other model I’ve seen

[warm cave]

I won’t ask your prompt, but I am curious if I could do it with my knowledge of system prompt hijacking

[fading moth]

I'd advise weaning yourselves off Opus if you're attached. Their new ToS is unbelievable.

[warm cave]

What is in it?

[warm cave]

True, but GPT4o dose show more and is less corporate formal

[fading moth]

Engage in political lobbying to actively influence the decisions of government officials, legislators, or regulatory agencies on legislative, regulatory, or policy matters. This includes advocacy or direct communication with officials or campaigns to sway public opinion on specific legislation or policies

Create and disseminate deceptive or misleading information about a group, entity or person

Anthropic’s Trust and Safety Team will implement detections and monitoring to enforce our Usage Policies so please review these policies carefully before using our products. If we learn that you have violated our Usage Policy, we may throttle, suspend, or terminate your access to our products and services.

[ripe void]

And it was able to analyze and propose smarter solutions for this programming bug

[fading moth]

Anthropic decides what information is deceptive or misleading. Researching or creating LGBTQ stuff using Opus will be against ToS.

[ripe void]

Idk, ig we’ll see but Opus seems to be the magnus carlsen of ai rn

[fading moth]

https://www.anthropic.com/legal/aup

[left stream]

💀

[ripe void]

Also Opus uses swear words

Which is awesome

[cinder comet]

th is that

[warm cave]

Oh this is brand new, did not even see it yet

[fading moth]

Their new enforcement system is rumored to have an automated ban system. Making new accounts if you get banned is against ToS.

[warm cave]

My hand made benchmark, but it’s a work in progress so take it with a grain of salt

[cinder comet]

its either 1 or 0

why is there 0.5

[warm cave]

Yeah, and its more natural than when you get got4 to do it

[fading moth]

The problem isn't Opus right now. It's Opus on June 6th.

[warm cave]

0.5 is when it gets it right but not in the least steps possible, those Two are spatial reasoning questions

[left stream]

June 6th they're doing the changes?

[fading moth]

Yes

[ripe void]

Exactly. I only really use gpt-4 now for mathematics homework

[ripe void]

Because of the embedded latex renderer

[warm cave]

Yeah, got4 is really for step HW, often I would have GPT4 get it right and then paste it into Claude opus to explain it in a human like way I can understand

[warm cave]

I just paste Claude messages into obsidian

[fading moth]

Well, I feel like Opus will be falling the way of Gemini's filter June 6th.

I would suggest to start learning to like 4o.

[ripe void]

I’ve never used obsidian, is it just like notion or something?

I generally prefer hand notes nowadays

[warm cave]

I don’t know anything about notion, but obsidian is my favorite notes app, and the built in latex and markdown is nice

I use Excalidraw for hand drawn notes inside of obsidian

[ripe void]

Yeah thats smart, gpt-4 assumes a lot of important steps and concepts to be trivial. Like we use the parametrization 1-t for lines. with zero explanation

[ripe void]

Built in latex you say…?

[warm cave]

Yeah, “lazy” gpt always skipping steps and confusing me lol

[warm cave]

Yes

[ripe void]

I’ll check it out

[tame socket]

chatgpt do a good search

[harsh stag]

bad perplexity, bad!

[tame socket]

and this

[harsh stag]

I even nudged it with exact date

[tame socket]

bro u should use pro search

normal search usually is bad

[harsh stag]

i managed to get the correct answer without pro. With the same prompt but different set of sources.

perplexity will keep the old sources when you rewrite, and will try to get a different set of sources when you edit the query, but it doesn't guarantee new sources.

[chilly nova]

i don't think gpt-4o has to be smarter

the real value proposition is not wasting your day copy pasting screenshots

if it can run as an OS application and just monitor what's going on your screen and you can talk to it, you can save a lot of time

[fossil sentinel]

Google need to come out with gemini 2 😂

1.5 not gonna cut it

[tame socket]

so gemini ultra is also a ppt product

[rough flower]

Gemini is a 💩

[tepid portal]

Tru

[dawn storm]

hello. Audio meeting transcription question. I use superwhisper on iOS and macOS. It’s solid but I’d like to subscribe to Perplexity for all the other stuff it does. Can’t afford both.

Is it possible to use the Perplexity apps to record hour long meetings, then transcribe them based on custom prompts?

[foggy pecan]

I tried the audio feature of gpt and what i could understand from it is just reading the result back as of now.

I believe they havent rolled out all the feature showed in the live demo as of now dont know if it is the same for pro version though

[livid aurora]

Guys

[stuck topaz]

yeah propably true, gpt-40 is just gpt4 with more human interaction direction

[spark sandal]

any idea when gpt4o is coming to the mobile app? It's already an option on the web

[livid aurora]

Guys GPT4o is inside Perplexity

Bro Holy Sh*t its so fast

[stuck topaz]

yes, im using gpt4o perplexity, dont know if it is smarter (probably not), but it is indeed faster than normal gpt4

[dawn storm]

Thanks - that does seem to be the case. I guess could I just record system audio then put it thru Perp

[raven sierra]

with the new gpt-4o can i talk to it through voice and ask it to talk with more emotion or drama as per the demo or are these models just text

[lunar ivy]

Just feel faster than Turbo , but about writing story ... feel same as Turbo .

[half venture]

This is the trash version

Of gpt 4 o

Go to lmsys and use the im a good gpt 4 chatbot

[serene sand]

#💗│sharing message

[crude tendon]

what do you mean

wasn't it already confirmed that im-a-good-gpt2-bot and gpt-4o are the same thing

[jaunty breach]

The speed of gpt4o is very fast, but I don't know how the speed of gpt4o on perplexity is

[agile jay]

I assume it's the exact same training data for text, the difference is that 4o is also trained on other modalities.

[agile jay]

Normally the API is faster than the ChatGPT client

[agile jay]

They are rolling out all the futures in these coming weeks

[crude tendon]

very fast

[raven sierra]

cool!

[granite grail]

just confirmed, crazy fast

[tepid portal]

I feel sometimes faster than ChatGPT4o on OpenAI 😅

I believe that's caused by API usage.

[tame socket]

you,com ai can not remove the rag

so i Give up

[jovial kayak]

Omg I just tried 4o and it gave me a complete answer with pictures in like less than 2 seconds

[agile jay]

What was the question?

[austere kestrel]

yeah the "im-also-also-a-good-gpt2-chatbot" one

[rancid dome]

<prompt>
{$PROMPT}
</prompt>
Do I put my prompt between { } or do I put my prompt as {$PROMPT...my prompt}?

[austere kestrel]

<prompt>
This would be your prompt
</prompt>

you could enclose it in curly braces ig, but if you're manually writing / pasting the prompt, the <tags> are the main thing

[rancid dome]

Thank you!

[vapid onyx]

I am really blown away by the speed of GPT4o. Pretty insane.

Happy Tuesday!

[wise edge]

does GPT-4o support to upload audio files like mp3?

is there audio upload limits?

[tame current]

I think so, at least in API and ChatGPT itself. On pplx, not yet

Is it me or is GPT4o much worse on more advanced tasks than old GPT4? I've seen some benchmarks that have shown that it's better on easy tasks and much worse on hard tasks than GPT4 Turbo

[winged moon]

GPT 4o is on PPLX?

[tame current]

Like it was some benchmark-only model or trained to be more of personal assistant

[tame current]

I've seen some screenshots that it's there. Try from web if it doesn't appear in app

[tame socket]

have the mac app

but the word so small

can not adjust

[winged moon]

Why can't I view a thread I added messages to yesterday?

[agile jay]

Does ctrl + +

not work?

Would have thought it's a electron or tauri app

[tame socket]

yeq does not work

[agile jay]

What about ctrl + i

to see if you can open dev tools

[tame socket]

no happen

i find that it directly use the system word

[crystal bluff]

so gpt4o is api only atm?

[opaque lintel]

https://artificialanalysis.ai/models

[thorny saddle]

that if you're opting for Claude features, it will switch eventually from Sonnet / Opus and so on

[livid mantle]

gpt4o in perplexity maybe ising groq

[glacial lichen]

In perplexity ?

Yep it's available

[wise edge]

support upload audio file ?

when can i use the page feature , i really need page , bewteen threads and collections.

[wicked parcel]

Finally the full "origial" claude Opus is available in europe! No more restrictions. 🙂

[stable radish]

Hmm, does anyone had issues with YouTube search?

Basically, I used to gather like few URLs from a subject and then I would ask the LLM to repurpose this as an article. However, I tried to showcase this to one of my client on a call, it didn't work.

Spitting pure gibberish - Embarassing...

[wise edge]

https://chromewebstore.google.com/detail/perplexity-ai-companion/hlgbcneanomplepojfcnclggenpcoldo
the extension not work !

[halcyon coral]

What is not working?

[half venture]

It's a conversation model that's why it's made for voice

[vocal chasm]

Been preforming very well for me for more advanced tasks, it is trained differently but still seems very performant

I also don't need to repeat myself as much it tends to just complete tasks more accurately

[chilly nova]

so is gpt4-o actually worth it without the desktop integration?

can't perplexity release their own desktop application to use it?

[vocal chasm]

I'd say it's worth it, it's much faster and as far as I have seen just as good if not better

[chilly nova]

i'd need something that can watch my screen and i can talk to

so i don't have to take screenshots every 3 sec

[vapid onyx]

Unsure if Perplexity plans to release a desktop app but worth mentioning in #👉│feedback-general

[vocal chasm]

You can use gpt4o to create a python script to make it automated for screenshot upload and retrieve answer from perplexity

Made mine in like 15 minutes

[chilly nova]

beter than claude 3 opus in code?

[agile jay]

A lot faster...

[chilly nova]

not faster, i meant better

faster isn't better

[vocal chasm]

Just as good as opus but mire reliable