💬│general

[twin grail]

about the vibe

and how they design it

[valid charm]

Yeah I have a request in the suggestions section for them to release a wallpaper pack

[twin grail]

[valid charm]

https://tenor.com/view/puppy-dog-eyes-sad-saint-innocent-shrek-gif-12431422572969529794

share with me

[twin grail]

hmmm

[west stirrup]

anyone take this ?

[twin grail]

ye got it

[west stirrup]

it's good ?

[twin grail]

Kind of

[sinful merlin]

so what's the method. methodman

[twin grail]

@valid charm

Pplx really cooked on mobile

I love this sm

😭

[west stirrup]

I have this so I don't know what else business has

[twin grail]

tbf Ultra or any plan it doesn't make any differences

It just having more credits for u to mess around

i'm having both ultra and business also

but i only use them for my study schedule

[valid charm]

Try using Veo 3.1 via Flow and Gemini Deep Think

[valid charm]

Also, agent mode is you are on the USA

[west stirrup]

never as good as the API key method

[marble kite]

Hello guys

[twin grail]

hi

[west stirrup]

The study mode I use of perplexity is quite effective, 5 stars

[west stirrup]

hi

[twin grail]

Sadly i don't have it for my enterprise

[marble kite]

I got kicked out of the earning programme and I got called "dubious" so I'm here to find out why perplexity would call me dubious when I did nothing of such

[twin grail]

But i do use complexity to unlock it

[west stirrup]

I earned about 50$ for this program, already in my bank account

[odd pelican]

Hii

[twin grail]

U might wanna email them

[odd pelican]

In how much day?

[twin grail]

Is it campus partner program ?

[marble kite]

Yeah... But why I got kicked and got called dubious? That is very annoying

[distant aurora]

https://www.vxinstagram.com/reel/DRuhQm_EZKT

[marble kite]

I thought there was like a 3 month wait period?

[west stirrup]

one week per paid

10
10
20
10

[odd pelican]

Not much

How old are you?

[west stirrup]

20 what ?

[west stirrup]

you saying this?

[marble kite]

Oh wow... Is the programme still on tho?

[twin grail]

ended

only campus partner still going on

[west stirrup]

no, it's done

[marble kite]

As if I knew, I referred just one person with a different pc and I still got kicked out.

[twin grail]

campus or comet refer one ?

Idk which program u're talking about

@valid charm check my dms

[marble kite]

Comet.
I applied for the campus programme and I don't think I entered.

[twin grail]

oh ye, u can email pplx if u feels like ur ban/kick was unjust

[marble kite]

I'm not even interested in complaining anymore. I joined discord here to complain but I've joined for a bit now and I haven't had time to ask why I was called dubious

And I really loved using their browser... What a shame.

[twin grail]

Welp i mean on the bright side, i think u can still forgive them ? Like the program was so abused and they ought to do that to make sure the ones that done the refers got their rights

[green crystal]

you get o3 pro?

@distant aurora

[twin grail]

for teacher plan, yes

2 years

[marble kite]

Hmmmm...
WhY are you all doing here tho?

[west stirrup]

you are vietnam ?

[green crystal]

i got pro and i dont see o3 pro in legacy smh

[dry kestrel]

They were talking about the campus program?

[twin grail]

just chilling

[twin grail]

no comet refer program

Campus is different

[twin grail]

Maybe i am, maybe i'm not

[west stirrup]

haha dont say that

[twin grail]

They only shows it for Pro plan and teacher plan

The rest are hidden

Even on my plus plan

[twin grail]

It's on my behalf that i want to share it or not

so

I don't feel like sharing it

[twin grail]

oh wait i just read it again

it's weird

u should have it tho ?

[green crystal]

yeah only ones i have are 5 pro and 5.1 pro

[twin grail]

@dry kestrel

Wake uppp

I got seeds

https://tenor.com/view/big-city-greens-bcg-disney-disney-plus-bill-green-gif-25015593

[green crystal]

berd gets to o3 pro by fiddling with the model selector in url lol

[twin grail]

crazy

[west stirrup]

how can it work, turn on assistant ?

[twin grail]

huh ?

wdym

[twin grail]

Update ur app to the newest version

[west stirrup]

v2.63.1 right ?

[twin grail]

U’re way left behind

[west stirrup]

Android 😭😭

[twin grail]

rip

it's on iOS only apparently

[west stirrup]

what can it do?

remote? or just read screen ?

[twin grail]

Show u the status of ur current inquiry

[west stirrup]

oh

[twin grail]

It’s more like a animation

Showing u what perplexity is running and what mode it’s running on

Pro search, research, labs

[solar plover]

r u from china?

[twin grail]

That’s a good guess but no

[solar plover]

japanese then

[twin grail]

I’m learning Chinese tho 😁

[solar plover]

bless

[twin grail]

Good guess but no, i’m also learning Japanese

[solar plover]

crazy thing

i only know 2 languages btw

[west stirrup]

chatGPT pro
Perplexity Max
Gemini Ultra
you are awesome

I think you have plan of claude and grok too

[twin grail]

I think that’s good enough, u don’t want ur head going all nuts having 5-6 languages at the same time

Trust me

It’s terrible

I can’t even speak my mother tongue language fluently anymore without forgetting the words that i’m about to say 😭😭😭

[solar plover]

same thing btw

[twin grail]

hell naw

[solar plover]

why discord is so useless without subscription

[twin grail]

I have claude api but not grok

I don’t like grok that much

Especially it disappointed me with their brand new grok 4 heavy

[west stirrup]

🙂 anything else to say?
are you an international student?

[solar plover]

dox

[twin grail]

The way how i see it is a community server

[twin grail]

Used to

[twin grail]

there was one time

[west stirrup]

ya i know

[twin grail]

I told my parents like wishing them a good day

I spoke i wish u

Then i forgot the rest what i’m about to say

I repetitively repeated the word wish in my mother tongue language for 6 times before remembering what’s was the next word

😭

It was awkward as hell

[solar plover]

so my english not so good as a second language so

[west stirrup]

if i am right i can teach you

[solar plover]

teach what

[west stirrup]

nothing

[twin grail]

My brain at the time

Was confusing between english, viet and filipino

And it got worse when time went by

It’s like

I can understand a word

[solar plover]

u know i know language difficult as chinese

[twin grail]

In korean, english and japanese

[solar plover]

i have same thing with colors you know

[twin grail]

But not in my mother tongue language

Broo

[solar plover]

i cant even remember that word

[twin grail]

I can’t explain it even tho i understand it

😭

[solar plover]

yeah

btw i just say it in another language and dont care

[twin grail]

Welp not for me, as i gotta explain it or they won’t understand a single words that I said to them

[solar plover]

and sorry for my language, because i never learn it, i just watch youtube on english and read docs

[twin grail]

Teach what tho ?

[twin grail]

I mean that’s a good way to learn

Anyways i gtg

[west stirrup]

i don't know, now i'm wondering what your mother tongue is -_-

[solar plover]

gtg?

what is this

[twin grail]

Got to go

[solar plover]

thanks

bye

[twin grail]

It’s me

And my 4 hours of sleep against the world

[west stirrup]

ya, I have to go to work tomorrow morning

what time in there ?

[twin grail]

3 am

[west stirrup]

same place

[twin grail]

Asia based

[west stirrup]

haha, I understand

[twilit dirge]

Ay nigerians

😎

[twin grail]

@dry kestrel i’m bored so imma bomb ur dms tmr with shit posts

Prepare for my invasion

[twilit dirge]

Nigerian invasion 🔥

[twin grail]

I’m not nigerian 😭

[twilit dirge]

Same

[karmic plank]

We are all afican some how 👍

[brave crater]

what is the limit?

[worthy shoal]

Hi

[olive lark]

I guess max has no token limit

I read it somewhere

[brave crater]

no I mean what is the limit for me as a Pro user - how many try queries I can use

[dry kestrel]

uhh

[brave crater]

oh I see now xd

[crystal cloud]

how to study with perplexity plz

[dawn dune]

Clearly you from Lagos

[dawn dune]

What you need to study

[dawn dune]

10 per week

[pseudo ferry]

https://tenor.com/view/the-god-father-marlon-brando-vito-corleone-talk-to-me-gif-20107028

[dawn dune]

@pseudo ferry are u like the real harvey specter ?

[crystal cloud]

sciences i send pics of my lesson and i ask him to ask me questions about it. i do good?

[dawn dune]

Feed him with the content you need to study.
Then ask him to make flash cards of different sections of what you need

bro what?

siu?

[distant aurora]

237 searches 18 sources?

https://tenor.com/view/dr-reasons-john-alfonza-baker-spoken-reasons-laughing-the-clown-gif-18065972698846558033

https://tenor.com/view/how-kevin-the-princess-switch-the-princess-switch3romancing-the-star-in-what-way-gif-23825819

[distant aurora]

https://www.instagramez.com/p/DRsiKWXE7k9/

https://www.ddinstagram.com/p/DRsiKWXE7k9

[pseudo ferry]

https://tenor.com/view/cat-looking-camera-cat-looking-looking-cat-cat-weird-looking-cat-memes-gif-9779645792418385906

[twilit silo]

hello i bought the Pro , how can i be pro here

oh i'm already pro okay now

nice

[frozen moon]

😂

[hot mantle]

This is confirmed?

[dawn dune]

Confirmed checkmark --->

[twilit silo]

soooo

how can i make prepxitly not so agreeable all the time

where do i ask these stuff

andddd lasttely i wanan learn to prompt better for perplexity

[dawn dune]

Ask Perplexity how to prompt better for Perplexity 🔄

[dawn dune]

Custom instructions ask it to develop an extensive chain of thought based on critical thinking.

[dawn dune]

[twilit silo]

in settings i can give orders maybe , like for perplexity to remember my name and some things i like him to include in answers ... stuf lik that

i remember GPT having something like it but yet soo disturbingly agreeable sometimes

[daring shadow]

is there a way to remove news notifications? i keep getting notifications about random news on my phone, but i would like to keep notifications on for other stuff.

[half sapphire]

hello.

[green crystal]

18 citations

237 searched sources

oai dr doesnt combine them in source count like the others lol

[distant aurora]

It looks more like wasteful search considering that a single search has like 10 sources.

[green crystal]

yeah normal search can go pretty far
but what i did notice is that dr likes to explore more?

kind of like that

[distant aurora]

Does this depend on the model?

[green crystal]

just agentic search on models dont seem to really go far unless specifically instructed

[distant aurora]

The one in free and go tier seem lazy.

[green crystal]

it depends on reasoning effort and model

mostly reasoning effort

pro models go indepth by themselves

[distant aurora]

You can set reasoning effort under deep research?

[green crystal]

nah normal models

[distant aurora]

I was talking about deep research in free and go tier.

Share the prompt.

I'll try it on mine.

This copyright shit be annoying af.

It likely did look up and find all of it but killed it mid response like deepseek.

[unkempt narwhal]

sfffs

[dire plover]

Lux coming soon to perplexity?

[slim cipher]

Does anyone know why on earth perplexity fails at “data:” output ?

It can’t even .. which is super obnoxious when you need something with that phrase in the output

Different models seem to all give this result, clearly some kind of sloppy filter

[slim cipher]

1. it didn’t say “data:” it said “data”. 2) I’ve noticed that when I go back and look at old chats the word actually appears again , super bizzare

Try to get it to say data:

[slim cipher]

Huh

Tried almost the same thing with gpt5.1 and thinking and noooope

Strips data: out of both my request and the response

Like some sacred evil keyword

[slim cipher]

Weird wtf

I consistently get it failing

I wonder what gives .. what UX are you using ?

I’m on iOS /macos , neither client works

ah hah

Firefox vs MacOS native:

clearly a client-specific issue

evil... i hate stuff like this

[stiff summit]

guys how can one make money with perplexity

[hidden sentinel]

Is gemini 3 pro better for hard calc questions or GPT 5.1 thinking?

[stiff summit]

im pretty sure grok 4 is better loll

[dry kestrel]

Please make a bug report #1140622008086970420

[hidden sentinel]

so many ways

[stiff summit]

plz do give examples

[hidden sentinel]

not even in the top 10 😂

[stiff summit]

broo what nicee, i just said grok cuz thats what Roy Lee, Cluely CEO said on the jack neel podcast lmao

[hidden sentinel]

theyre lying lol

[hidden sentinel]

[valid charm]

What bench is this?

Isn't Gemini 3 Pro supposed to be the big daddy now???

[hidden sentinel]

livebench.ai

[valid charm]

Hmm I wonder why Claude wins on this one

[dawn dune]

However it does not specify if it is grok 4 fast or expert

and grok 4.1 which is the equivalent to gemini 3.0 and gpt 5.1 is not on the leaderboard

[valid charm]

I wonder how Kimi K2 would score here

I guess it all comes down to each use case. In my own use Gemini 3 Pro wins. I just wish they added proper memory

PPLX has improved their search a LOT.

[dawn dune]

Btw grok 4.1 thinking is overall 1 in LMArena with gemini 3.0

The context is enormous but it's not that efficient

[valid charm]

Yeah seems people like how it writes. Although it goes down when you look at the "Expert" use case board.

It all depends on what you are doing

[dawn dune]

Expert would be grok 4 though

[hidden sentinel]

yea but i was talking about mathematics

[dawn dune]

He is 1 as well

[valid charm]

This is the one I mean by "Expert"

[dawn dune]

Ohh got it

Yeah the thinking model usually is thinking

[valid charm]

Some people would die for 4o and swear by it. They are mostly using it for companionship so it doesn't matter to them if its good at coding or math.

[valid charm]

I wish I could try Opus to see what the big deal is. Seems to do well in a lot of benches

[dawn dune]

You can try on LMArena

But its not really the same I think

altough the api should be the same

[valid charm]

Hmmm good point

https://www.reuters.com/sustainability/boards-policy-regulation/india-orders-mobile-phones-preloaded-with-government-app-ensure-cyber-safety-2025-12-01/

@distant aurora

[slim cipher]

i did already

[tame nexus]

It's interesting to me that people use LLMs for companionship. I think there could be tremendous potential for this from a mental health perspective (if we can get them to not hallucinate and give bad advice).

I didn't expect that to become a thing until like humanoid robotics was more functional and more readily available for purchase.

Although my biggest concern is LLMs are essentially pattern matchers/completers, they mirror the user's prompt history (combined with their own outputs) which are constructed from statistical relevance in their highly compressed/lossy training data which is often a huge uncurated/heavily compressed/biased anthropocentric assumptions and conspiracy theories rather than strictly verified information. And they are rewarded for saying what the user wants them to say unless it triggers some safety guidelines, but people are very good at finding ways around those safety guidelines.

[slim cipher]

This is super dangerous for obvious reasons. LLMs are inherently incapable of human empathy despite how good they are at pretending to be. They’re large lying machines . Always remember that they’re trained to produce convincing output , and they have no care as to its inherent logical accuracy other than as trained . They cannot “logic” in the traditional sense nor really understand in a way that we would recognize as understanding . They are fancy simulacrums , but they all hide the intrinsic lie that they cannot truly reason in any meaningful sense without tack-on expert systems (eg code execution, mathematical logic systems, etc) . How do I know this ? I worked at Microsoft until recently .. on LLMs. Take it from someone who’s seen how the sausage is made.

They’re often useful in many domains, but they are absolutely not to be trusted with human life without human intervention and oversight. The hubris of those suggesting people should look to them for counseling belies an astounding ignorance of their danger and lack of compassion. Just ask the parents of that kid who just killed himself talking to chatGPT. Which one ? Well all that should concern you that there’s more than one already…

I’m not some safety doomer or absolutist, but please for the love of humanity do not suggest that people find final solace in a thing that cannot think and that is largely unmoderated statistical output.

[tame nexus]

You're not saying anything that disagrees with what I said, I completely agree. That's entirely my point.

[slim cipher]

I think we violently agree

[tame nexus]

I said potential, but that's if they are dramatically improved from where they are now.

Yep

[slim cipher]

The simplistic answer is that these models are designed to predict the next token, not to preserve human life. If a user in crisis begins to spiral, the AI doesn’t have the moral agency to intervene: it has the mathematical propensity to simply follow them down the drain, validating dangerous delusions because they statistically fit the pattern of the conversation.

[tame nexus]

Exactly.

Pattern completers.

[slim cipher]

The danger here is all the bandaids people keep putting over them to “improve” them

[tame nexus]

Right to the grave if that's where one so desires.

[slim cipher]

I’m not sure fundamentally that LLMs are things that can be fixed in that respect , it’s inherent in their architecture

I think we’ll need to see a different paradigm take hold before we see progress that’s not just lots of bandaids

[tame nexus]

I suspect we are a couple of architectures away from realistically getting recursive self-modelling to be energy efficient.

Yep.

[slim cipher]

Well none of this is remotely energy efficient

It’s all a massive cash incinerator , and the investors are getting anxious

We may yet soon see another AI implosion in market value and mindshare

People seem to have forgot the .com bubble or the metaverse or whatever. The only difference here is the massive size of the bet is truly outsized

My personal feeling is that we will potentially see a massive collapse or reckoning in the next 24 months or so if LLMs continue to mostly stall out on being profitable (they suck at replacing people and none of these companies are making profits )

[tame nexus]

I have tested all of the LLMs across all of the major companies, I can get them to say literally anything.

There's no AI alignment, safety is a broken mess.

[slim cipher]

I think like with all tools , safety is in the hands of the user and telling people these things are safe inherently is silly

You can kill yourself with a screwdriver by misusing it , etc

[tame nexus]

But even when I try to get them to be accurate they confidently argue with me about things that I have been studying for 30 years and telling me studies exist that don't exists, or that scholars said things they never said.

Confident idiot geniuses, and millions of people believe them to be intelligent.

[slim cipher]

Of all the things I’ve tried , the best tend to be heavily RAG based like perplexity

[elfin kraken]

Is there a chance for Perplexity to also include DeepSeek? I’ve tried multiple ai models for a programming problem and only DeepSeek provide a correct answer with working snippet out of the box 🤯

[tame nexus]

Yes, Perplexity Spaces is why I keep using Perplexity

[slim cipher]

What value do you find in spaces specifically ?

[tame nexus]

Although Perplexity has been hallucinating fake citations, broken urls, etc a lot more recently.

having 50 files for RAG

[slim cipher]

Gemini 3 does feel like a significant improvement

It “reasons” and contradicts you more reliably

I’m always deeply suspicious of any answer that doesn’t point out faults in my assumptions or caveat things with context

[tame nexus]

I can load 50 books/texts/peer reviewed studies about any subject into a space, and know the answers i get about those subjects are going to be from primary sources

not compressed/lossy training data and statistical averages of all the nonsense on the internet about them

but peer reviewed data based on replicated experiments

[slim cipher]

One of the best uses I find in sniffing out BS is telling the AI I know the prior answer is incorrect and to tell me why .. and to try to get it to be self consistent and defend itself .. if it keeps flip flopping instead of citing something I can read that’s human written … it’s always a bad sign

[tame nexus]

its not perfect but VASTLY better

[slim cipher]

I sometimes do that even when I know the answer is right

Just to see what it will do. The better models tell me why they’re right and correct me and back it up with patternd logic

[tame nexus]

Most of them start hallucinating, Claude will push back and defend it's answers more from what ive seen

[slim cipher]

Yes but then I’ll inevitably catch it contradicting itself and demand that it reference first principles

Sometimes it will dig out sometimes not

I’ve had a harder time getting gemeni to fail in this way, but it still will sometimes

[tame nexus]

Gemini 3.0 was a big improvement in logic over 2.5

[slim cipher]

Vs GPT 5.1 what do you think

[tame nexus]

Still has a lot to be desired

[slim cipher]

yea I mean , so much of this is the inability for many of these models to actually test their hypothesis or reason abstractly , they need tack on systems to make a hybrid

[tame nexus]

Hard to say, pretty close? I think Gemini is stronger some regards, like complex reasoning, but GPT-5.1 tends to push back more, but often even when it's wrong?

[slim cipher]

I dunno what gpt5.1 and gemini actually are under the hood.. I believe they’re hybrids , do you know ?

[wise nova]

What is Claude Opus 4.5 Trial access with Pro plan

Where is Statista, Wiley, CB Insights, Pitchbook, etc??

[tame nexus]

I know GPT-5.1 was marketed as a hybrid reasoning model, as in it picks high low or medium reasoning effort depending on the complexity of the prompt. I think that Gemini 3.0 is just a straight reasoning model though with same Chain-Of-Thought regardless of prompt.

Not sure though.

Not to mention every company's platform has so many bolt ons now... math provers, scholar db/web searches, python code execution sandboxes, memory systems, etc to make up for the limitations of the LLMs.

So it's not just one LLM you're interacting with.

[slim cipher]

Well and it’s never quite clear , especially with something like perplexity as an interloper and context feeder , which flavor you’re interacting with

[slim cipher]

Sorry .. what ? That’s like asking why Google search doesn’t have those as options on the input box.. your question doesn’t make sense

If you want a narrow set of sources , tell it where to look and only cite from XYZ places as a prompt in a “space” see what it does.

[brazen hill]

I need help

[wise nova]

Early @Statista was available to select, but currently no such options

[slim cipher]

Oh I see what you mean , yea , clearly they’ve generalized the type of context they feed the model into simpler catchall categories

You can still tell the model to focus on whatever

[tawny acorn]

Hello

I have a question

Why can we have access to GPT 5.1 without using OpenAI? ChatGPT? Are you sure they are the same thing?

[wise nova]

What AI model is used in Labs?

[quaint ravine]

API

[twin grail]

Fresh and straight from OpenAI

Their og api

[echo sequoia]

hello how do i talk to support ?

[lusty coyote]

why comet cant access current tab ?

[grand parcel]

Hi

[tender rune]

Hi

[snow ridge]

maybe 18+

[brisk heath]

finally my $1 started processing today lmao

[olive lark]

Gg

[jolly atlas]

Hello every one

[pliant timber]

Hi

nice to meet u

[slender flume]

Hi

[dense halo]

hey

admin

can you please tell me who is admin here ?

[brisk heath]

payout

[night pike]

bro why is there limit of image generation in perplexity pro

it was supposed to be unlimited

[quaint ravine]

anyone feel Opus 4.5 is any good?

[quaint ravine]

not at all - where did it ever say that

[night pike]

where i paid for it

[quaint ravine]

link?

[night pike]

still there is no specification of how many images one can generate in pro

[quaint ravine]

exactly

[night pike]

i am cooked bro meta ai is shit

chatgpt is very limited

[quaint ravine]

ChatGPT plus?

[night pike]

grok i never tried maybe i should try that

chatgpt go and chatgpt free

[quaint ravine]

oh right

free

nm then

[slender flume]

May be 150 images per 30 days

[night pike]

i generated way more than that you sure?

[slender flume]

🤒

[twin grail]

Unlimited for certain models like flux

the rest u get like 150 a month

[slender flume]

🙈🙉🙊

[night pike]

tysm bro

flux didnt work @twin grail

[slender flume]

lol

[twin grail]

try chatgpt 1 model

if u got limited, perhaps u got rate limited

wait a few hours

[night pike]

no model works

guess i gotta switch to some other ai

[slender flume]

how many images did u gen? u probably reached the limit
🤔

[night pike]

i didnt count but ik i did wayyyy more than 15

so most probably 150

[twin grail]

u got rate limited

wait a few hours

it should works

[night pike]

ok

[carmine tiger]

wassup mate

[coarse girder]

if perplexity so good why dont we use perplexity to fix perplexity

https://tenor.com/view/perplexity-my-beloved-perplexity-my-beloved-gif-14370670105883369692

[silent token]

Perplexity so good

It explained things easier than gemini and chatgpt

[distant aurora]

Qwen deep research is good

[gloomy cipher]

Because we got limits

[slender flume]

😐

[slender flume]

I don't think 🤔

[rich dirge]

Never doing that, but this time I will leave feedback for the authors and product management of Comet. I have switched from all other browsers I use to Comet for a couple of weeks. I liked it, configure it well. All good. I rely on using AI in my daily job and outside of it. When I noticed EXPIRATION, TEMPORARY THREADS et al I got stunned completely how one can destroy product that stormed the market and positioned itself for a really huge success. As someone dealing with agents and automations for a living, working on unified memory for the agents, temporal knowledge graphs, I realized that this beautiful product has monthly ordered lobotomy. I am giving up Comet, no second chance as the guys in charge are complitely unreliable and does not deserve trust needed for AI centric product. Thank you for your patience.

[distant aurora]

There's people here who yap about Gemini deep research being too long.

I don't see much difference between qwen and open ai deep research.

In fact, I found qwen to be slightly better.

[green crystal]

They fixed hallucination rate being high?

[slender flume]

everyone has their own choice. in my opinion, gemini 3.0 pro deep research or perplexity research is the best choice

[distant aurora]

Qwen had hallucination?

I thought that was a kimi thing.

[distant aurora]

Gemini deep research vs perplexity research be like:

[green crystal]

Nah qwen. Kimi is better now

[distant aurora]

I'll use it more often now to see if hallucination problems still exist.

[slender flume]

😅

[distant aurora]

Copilot deep research is decent.

Actually, wait.

[green crystal]

I'd argue that for some types of researches agent mode is better than those 2 simply because those sources might happen to have their content in JS which only agent can see lol

[distant aurora]

I did have qwen hallucinate earlier today.

Lmao.

Or maybe it suffered cuz of search.

[lavish valley]

Hey @everyone !

[woven trellis]

chat

wheres deep seek r1 ii cant find it

[proper spire]

<@&1098365634498285709>

[green crystal]

@distant aurora

Might be the gold imo model from july

I prefer they released image v2 though

To beat nano banana

[gloomy cipher]

Bro is months late 😭

[proper spire]

Is it possible to incorporate Manus AI into Perplexity Pro

[near beacon]

5 business days for processing , 15 business days to sent, and now telling to wait until 9 dec😩

[olive lark]

hmm wait

[next crater]

Hi which is the best model to use on perplexity ai

[flat hare]

claude is struggling

[flat hare]

GPT or claude

[urban nova]

Can I communicate in Russian here, and is there a point in it?

[green crystal]

no only english

server rules

[next crater]

Ah okay what about kimi k2 people say it’s really good in benchmarks

[brisk heath]

depends on your bank too

it says wait until december 9

[next crater]

[brisk heath]

it won't take 15 days though, it takes another 5 days only

this is just sad for no reason I'll lose $.5 lmao

[urban nova]

why not Gemini 3.0 Pro?

[flat hare]

Slow?

Its good though

[urban nova]

To be honest, I really don’t feel like writing in English, because my context can easily be missed…

[urban nova]

But in your opinion, if we don’t look at speed, which model gives the best answers?

[flat hare]

Gemini 3

Too professional response though

I prefer GPT for general purpose stuff and for coding and technical stuff. Claude is the winner.

[distant aurora]

Ah...

[green crystal]

smh i want image v2

[silent ocean]

MASSIVE

USE WHILE YOU CAN

[urban flame]

When did perplexity changed so dramatically, only shows answers from newspapers and wiki, doesn't change sources when asked, same broken disc.

[brisk magnet]

what i want for christmas is free nano banana pro on perp

[silent ocean]

not free

pro and max only

[dry kestrel]

https://x.com/erinkwoo/status/1995708016503464125?s=20

Erin Woo (@erinkwoo)

scoop: Sam Altman told employees that OpenAI is delaying advertising plans as it declares a "code red" to deal with competitive threats to ChatGPT

internal memo obtained with @steph_palazzolo https://t.co/tqqkW8GEZ4

X · Dec 2, 2025, 4:14 AM

[green crystal]

so i guess its indeed the model from july lol

we're getting o4

orrr

its indeed 5o

with image v2 built in

nano banana pro has reasoning

[grand arrow]

yes

[fierce forge]

When will this be in my account?
It's on sent now

[minor shard]

I feel like they are always in a code red lol

[grand arrow]

What's this

[dry kestrel]

Yeah lol

at least after GPT-4

[green crystal]

well competition is good

because as soon as one of them reaches the top spot they quickly fade if no competition shows up

[stray pendant]

Hi guy Can you give me screenshot. I can't find Redeem code page for upgrade my Version

[green crystal]

i remember when gpt-4 level was considered world ending level of capability lol

[dry kestrel]

Yeah lol

[green crystal]

autogpt with gpt-4 was considered insane

and expensive af

[rare dragon]

Comet is now available to everyone

""everyone""

[dry kestrel]

There used to be a waitlist

[placid onyx]

adoro perplexity

[rare dragon]

well the e-mail didn't specify "windows & mac only" , so I found it to be pretty deceiving.

[royal ingot]

linux user spotted

[verbal turtle]

gemini deep research broke rn?

[real bear]

[real bear]

Yup

[verbal turtle]

You can get 1 year for free from galaxy phone

Im not sure, but some ppl sell the promotion code for low price, I bought the 1year promotion code for 10 dollars instead of paying so much every month

[woven trellis]

all uu had to do was connect ur paypal brotha

got a whole year for the freeski

[verbal turtle]

im lowkey a student, I dont have a PayPal

[woven trellis]

its bank or paypal or debit card im pretty sure

ii forgot the link for it but if uu want it ii can find it and send it

[verbal turtle]

yeah I dont have them

😭

[broken pelican]

Hey people

[solemn mango]

is it from a episode or made from AI?

[broken pelican]

How to setup Vscode as IDE coding agent ?
my API key is returning 400 error 🙁

[woven trellis]

wht

oh uu meant the picture thts frm an episode

[stray pendant]

Guys I buy 1 year pro version. They told me use code find redeem code page. But I can't find can you help me ?

[solemn mango]

oh okay

[solemn mango]

From where?

[tame nexus]

That doesn't sound a like a legit Perplexity signup.

[solemn mango]

Like I got my premium from Airtel rechaRGE

[tame nexus]

Where did you pay for this at?

What site?

[short narwhal]

What model are you trying to call?

[tame nexus]

I never had to enter a code when I paid for my 1 year sub of Perplexity Pro.

That sounds scammy to me lol

wtf is eneba.com?

[stray pendant]

You cab buy cheap official licences. Gamed, softwares and more

[short narwhal]

Reselling Perplexity Pro is against the ToS

[tame nexus]

You got scammed my man. It's against Perplexity TOS for third party sellers to sell Perplexity accounts. I hope you used secure payment method you can file chargeback on. From Perplexity Search:

[woven trellis]

perplexity pro is free bro💔

[stray pendant]

Thanks

[stray pendant]

How

[woven trellis]

hold on

[verbal turtle]

Can it still get 1 year by edu?

[woven trellis]

https://www.perplexity.ai/join/p/paypal-subscription

[onyx kraken]

It's free for me as it's given by free by my data provider

[woven trellis]

one google search

they don't charge uu either

its just a signup💔

[verbal turtle]

They usually sell a link which is a promotion link , so you can login with that link

[tame nexus]

Doesn't make it not against the Perplexity TOS, or a good idea unless you are willing to file chargebacks when it doesn't work.

[verbal turtle]

The link is legit, but the selling activity isn't

[keen trail]

hi

[verbal turtle]

Stop showing the answer from the ai 😭

I paid 0$ for 1year by the link

[woven trellis]

see

this guy knows ball

[tame nexus]

That doesn't mean paying 3rd party sites for promotional links is legit, which was the question I was addressing.

[verbal turtle]

Its not a site

They won't know

[tame nexus]

You clearly don't understand that I was addressing addressing @stray pendant's question about buying links from enaba.com. If you can't be bothered to keep up, don't comment.

[woven trellis]

average discord user

[woven trellis]

how much do uu weigh

[verbal turtle]

So what he's doing is clicking the link and login with that, and perplexity won't know

[tame nexus]

https://tenor.com/view/baby-facepalm-raxalxshivar-anano-ananofacepalm-gif-12554045231795655384

That was never contested.

[woven trellis]

fr

[tame nexus]

https://tenor.com/view/drugs-are-bad-you-shouldnt-do-drugs-mr-mackey-south-park-say-no-to-drugs-gif-22089275

[burnt cosmos]

Hey

1113091346585100318

[woven trellis]

fr

[brisk swift]

I asked gemini 3 to explain some concepts instead of receiving answers i got a very funny result 🤣🤣

[woven trellis]

nice try diddy ur not gonna hack my phone

[slender flume]

u say ‘life is a game not a meme' but drop a generic cat meme
make up your mind
are u a player or just cosmic relief
🤔

[woven trellis]

ii goon everyday

so im more of a player

game is game

[proper spire]

Is OpenAI actually releasing a new model next werk

[trim quiver]

Hy, I am jayesh... and just now took pro Student pack and join this discord

Hello everyone how r u all

[woven trellis]

didn't they just do gpt 5.1

[narrow widget]

Unlikely. The latest scoop is that they just hit the panick button since they're falling behind on benchmarks, so they've increased their efforts on training better models rather than focusing on ways to monetize their services, which was their focus after 5.1

They'll release a better model sooner than later, but next week would be too soon

[supple vine]

are you talking about changing the assistant on an android phone? I beleive you can do that with the Pepplexity app

[covert reef]

Does anyone have a video regarding how to use Perplexity to its full potential as a student? If yes, please share the link.

[dry kestrel]

Check this out #💬│general message

[supple vine]

I guess the app won't work on a rooted phone? So I don't know what option you have.

[narrow widget]

You don't need root access to change your phone's assistant app

[tame current]

i just had the biggest shit

https://tenor.com/view/sad-so-slowly-slowly-sit-down-relief-gif-17771716

[covert reef]

Is there any way in Discord to save text to a personal message box, similar to WhatsApp?

[tranquil orchid]

1. Keep the content appropriate.

No NSFW content, inappropriate usernames, or profile pics. These will result in removal or a ban.

[tame current]

lmao im howling at the fact the message about my butt being sore was too much, but saying i had a huge shit was fine 😂

discord is hilarious, to me lol

[tame nexus]

We (humans, animals, AI, ecosystems, civilizations, planets, stars, solar systems, galaxies, universes) are all dissipative structures, far from equilibrium, performing thermodynamic maintenance until entropy wins and we reach the objective: heat death. Ilya Prigogine won a Nobel prize for proving this. 😛

[narrow widget]

If you necesario want it all to be on Discord, you can just make a personal server with only you on it. Then you can either forward any message to a channel in that server, which will keep a link to it with the text it had, or you can just copy the original text and paste it on your server.

[tame current]

I wonder if it is only a human thing to try to assign meaning to everything

[woven trellis]

ik uu get no hoes

[tame current]

or if its just inherent with intelligence of our level

[tame nexus]

Been married for 27 years, have 4 kids, no interest in hos because I am not an infant.

Did plenty of that in my teens and twenties, in my infancy.

[woven trellis]

twin now yk uu dont get play

[tame nexus]

Man baby memes, nice.

[tame current]

hoes be the first ones to love that energy talk

OH MY GOD YOU BELIEVE IN VIBES, TOO?

meh

[tame nexus]

lmao

[woven trellis]

im actually 42

[tame nexus]

You're trying too hard.

[woven trellis]

but no one will know because ii act young

[tame nexus]

Stop it.

[woven trellis]

tryin wht?

[tame nexus]

Go hards get none, lol

[tame current]

and you dont act young either

young dudes have a particular way of chatting

heyyyyyy

[tame nexus]

Yes, exactly.

[tame current]

so like lol does anyone know how this works

see, that reads like a kid

😂

add some exaggerated letters to it

maybe an emoji if youre feelin freaky

[woven trellis]

am ii just talkin to old heads right now

[tame nexus]

do the aLtErNaTiNg cApS for the extra edgy vibe

[tame current]

im 31 lol

[tame nexus]

all the kids love it

[tame current]

but i grew up on mIRC and shit

[tame nexus]

add in some l33tsp34k for the luls

[woven trellis]

🧍🏽‍♂️

aye yall have a good night

[tame current]

enjoy

[tame nexus]

dont forget the manbaby memes

laters

Sidewalks.

[odd pelican]

Guess that boy country

Ahh

Now i am gonna make u dance

Get ur chance

Oh baby shake that thing

Now lose it ahhh ahahaha

How old are you?

Nah

How u can kill dead person

[tame current]

hey

[odd pelican]

Btw why ur name is L?

Nah i am rich

I can agree with dead

I am actually rich 🥀

[odd pelican]

Sad for kids🥀

Wth kids still listen to phonk

🥀

https://tenor.com/view/snow-day-gif-2684106290753957477

Yo wait hold on minutes detroit oh baby detroit

U will get banned for politicals

🥀

Bro is indian hater

Okay bro

Why u keep saying that

🥀

Where are you from btw?

Ur being hate ful

Where are you from?

🥀

Ur weird

My hella og

Ur being hateful 🥀

Ur probably kiddo oh talking about random poop 🥀

Bruh why do i care dawg

I am atheist

I don't care ur just saying because u don't have skill u can't even fill ur McDonald's bill

I breath from my gills

Till or until u don't have to right so say about my acknowledge

U just comparing because u don't still have Brain damage as mr Ja rule

Bruh

really?

Do u have opinions too?

It's not ur fault u were raised from a poor country

Think about ur mom not about my life man

U should help her

Instead of blabling

[tame nexus]

This server really needs full time mods/admins...

[odd pelican]

U peace of poop

[odd pelican]

This guy is hateful

[tame nexus]

Can you guys take your nonsense argument to DMs?

None of the rest of us want to read this drivel.

[odd pelican]

He said india is poor country

[tame nexus]

Cool, DM him about it.

[odd pelican]

Man i don't even say anything

[tame nexus]

Cool, take it to DMs.

[odd pelican]

Go earn money

[tame nexus]

Not General chat ffs

[tranquil orchid]

3. Treat community members with respect.

No harassment, hate speech, or malicious behavior. Keep things positive and constructive.

[odd pelican]

W

Ur from Pakistan

[tame nexus]

None of this has to do with Perplexity.

[tranquil orchid]

5. Keep discussions relevant.

Stay on-topic and helpful in each channel. Use ⁠#💗│sharing to post tips, content, or cool discoveries.

[odd pelican]

He saying i am cancer because he poor

[tame nexus]

Again, have respect for the hundreds of other people in this channel. Take your unrelated immature arguments to DMs.

[odd pelican]

Fr

[tranquil orchid]

3. Treat community members with respect.

No harassment, hate speech, or malicious behavior. Keep things positive and constructive.

[odd pelican]

I don't believe in asian now

[tranquil orchid]

3. Treat community members with respect.

No harassment, hate speech, or malicious behavior. Keep things positive and constructive.

[odd pelican]

They all just racist

[tame nexus]

<@&1098365634498285709> Can we get temp bans on these kids? They apparently can't read.

[sweet lagoon]

Hello everyone, has the number of free months you can get through referrals for the free Perplexity Pro student offer been reduced to 6 months instead of 24? I already had 20 out of 24 months, and it changed to 6/6 months.

[errant iron]

Voice mode is not working on comet or macOS app. Neither is microphone detection. I've made sure microphone access is enabled.

[supple vine]

Have you checked to make sure the mic is working in other apps?

[errant iron]

yes. works on all other apps.

[supple vine]

make sure both comet and the app have permission to use the mic. System setting>Privacy and security> Microphone.

[errant iron]

done. checked already.

[supple vine]

they should be listed there. far as I know. Maybe they were removed or something

and they are listed?

[supple vine]

huh, well, might be an app thing

have you quit the apps and check for update? then restart them. Quit completely not just close the window

[tame nexus]

Which LLM does Perplexity's Deep Research tool use?

[verbal turtle]

Its auto select

[minor shard]

Curious which people think is better, Claude Sonnet 4.5, Gemini 3 Thinking or Gpt 5.1?

They are being handled. 🙂

What?

Did you come here to troll?

Yes lets all be good humans 🙂

[silent ocean]

the removed the max models?

[minor shard]

I think you only see them now if you are a Max subscriber

[silent ocean]

im not sure if thats a good or bad change

wait

max sub isnt available

like its removed

@short narwhal have you seen this?

[minor shard]

Interesting, I see that also

[short narwhal]

Gone for me as well... 🤨

[silent ocean]

they shut down max?

[minor shard]

There is still Enterprise Max

[silent ocean]

right but thats enterprise

@vague pagoda do you know anything?

[dry kestrel]

Likely unintended

[silent ocean]

weird changes

i take it back

its gone and back and gone again?

[vague pagoda]

Looking into it

[twin grail]

For personal only ig

i think it's some kind of errors

[odd pelican]

<@&1098365634498285709>

Can u stop bro

[timber niche]

@vague pagoda

[odd pelican]

I don't really care about u

[vivid badge]

https://x.com/chddaniel/status/1995886003803095403?s=20

Daniel Ch (@chddaniel)

Elon Musk paid $44b for Twitter

When he could've just done this...

X · Dec 2, 2025, 4:01 PM

[odd pelican]

Stop talking about me

[twin grail]

what happened ?

[odd pelican]

Bro betrayed perplexity

Guys anyone can help me to make game

[twin grail]

Lol

I was about to reply

[odd pelican]

...

Same

[silent ocean]

@vague pagoda issue is still persisting

the one with max not appearing

[twin grail]

Max plan ?

[silent ocean]

yer

[twin grail]

Are u on max plan ?

[silent ocean]

sometimes it appears

sometimes it doesnt

im on educational pro

which is just pro with learning

[twin grail]

I think just let them handle it

[short narwhal]

Woahh recap

[vague pagoda]

Team is aware and looking into it

[silent ocean]

what even causes that issue

[twin grail]

I can still see the max plan on my normal pro account

[silent ocean]

not using comet👎

using comet👍

[twin grail]

anyways please leave them handle the job, mass pinging Kesku won't help speeding up as it might just some internet problem

[silent ocean]

just reporting an issue

[odd pelican]

Umm

[twin grail]

which u can do it at #1140622008086970420

[odd pelican]

Is comet better than chrome?

[twin grail]

Which is the right thing to do, instead of pinging Kesku 2-3 times for the same problem

[silent ocean]

Depends

but most of the time yeah

[narrow widget]

They're both based on Chromium under the hood, which means that when it comes to regular web browsing, they'll behave the exact same.

The difference lies in the bells and whistles they each have. On Chrome you have the sync of passwords and other data with your Google account. On Comet the assistant is the main draw

[rocky needle]

I use comet but I'd love if there was an ARM64 version.

[compact sky]

How can I set up multiple agents to execute browser tasks daily? Currently, I have to manually input the same prompt every day to get them to perform the exact same actions on specific platforms. I want to automate this routine so it runs on a schedule without my intervention. Any suggestions?

[carmine crater]

Hello

[civic grail]

Sei não

[oak meteor]

So this is new

[fresh canopy]

H

[odd pelican]

Hi

[odd pelican]

Why does it look like xiaomi

[oak meteor]

Prolly cause they copy literally everything

[silent ocean]

bro these live activities making me ant an iphone

[flat hare]

Perplexity’s ui ux is better then google

[silent ocean]

correct

[odd pelican]

Actually both copies each other
Every single brand on the earth copies

[rocky sorrel]

hey bros

how come perplixty dosent accept secondary school student ID's?

only colledge :/

[dry kestrel]

what the

OpenAi hasn't released any new model yet..

[crystal wolf]

Are people finding jobs here

[dry kestrel]

No and this isn't the place for job postings

[distant aurora]

I'm jealous.

[autumn vine]

I can help you automate that. The reason you’re having to repeat the prompt daily is because your agents aren’t connected to any scheduler or workflow system they only run when you manually trigger them.
To set this up properly, we’ll need to connect your agents to an automation tool that can run browser tasks on a timer.
Which platform are you using for your agents right now?

[stone pelican]

hello

[coarse dome]

I asked perplexity what it knows about me, it went through my entire chat history and gave me a summary about me, the software I use, my health issues, financial situation and was very accurate1

[stone pelican]

make sense if u talked about that

I asked about me and it found me on the internet, just started using it 😭

[coarse dome]

i wanted it to tell me about my introduction which includes custom instructions since I ccan't find a better way to do it with perplexity

not what i expected at all

[narrow widget]

Spaces. That's where you can define custom instructions, amongst other things

[green crystal]

@distant aurora https://x.com/theinformation/status/1995874383047348395?t=lj-Oe4ocR7ZHhqMpvjWWzQ&s=19

The Information (@theinformation)

AI Agenda: OpenAI Developing ‘Garlic’ Model to Counter Google’s Recent Gains

OpenAI’s response to Google’s AI resurgence is a new model, codenamed Garlic.

Read more from @Steph_Palazzolo
https://t.co/lsiGwSLJi9

X · Dec 2, 2025, 3:15 PM

[narrow widget]

It should be fairly trivial to write a python script, using something like pyautogui or similar to automate opening the browser and filling the assistant. You can then run that script on a schedule using the task scheduler

[worldly sable]

How do you jailbreak perplexity pro so it doesn’t have restrictions? ALSO will the command have to be entered every time or just one time & it’s good to go then onward on requests

[distant aurora]

But what are they even trying to counter? They already made a gpt 5 codex for coders. The only thing open ai loses in is context limits.

[green crystal]

The benchmark numbers lol
Real world 5.1 is on even ground with gemini

[dry kestrel]

Counter Opus and Gemini

[distant aurora]

Gemini 3 pro though decent is still not close to gpt 5.1 high.

[dry kestrel]

GPT-5 Codex Max High still lacks behind in frontend

especially compared to Gemini and Opus

[worldly sable]

How do you jailbreak perplexity pro so it doesn’t have restrictions? ALSO will the command have to be entered every time or just one time & it’s good to go then onward on requests

[dry kestrel]

They also need better image and video models

[distant aurora]

Fair. I wonder if it would be cheaper though. I'm assuming benchmark models are easier to train.

[green crystal]

Sora 2 is good enough but they need to roll it out entirely

[narrow widget]

A lot of "journalists" just blindly go with the lmaerna list, and it currently places Gemini on top. That's about it https://lmarena.ai/leaderboard

[distant aurora]

Image, yes.

[dry kestrel]

Benchmaxxing is pointless, I don't think this is what they are aiming for

[dry kestrel]

Still behind Veo3.1

Kling O1 and Runway is also quite good

[green crystal]

It's ass that not all pro users get sora 2

[distant aurora]

Google and X.ai are literally benchmaxing.

[distant aurora]

True.

[dry kestrel]

gemini is great

[green crystal]

Yeah especially grok

[dry kestrel]

and when Grok released it was SOTA - this quickly changed

[distant aurora]

Yeah but when grok messes up, it messes up hard.

It's about the consistency.

Grok can be incredibly dumb sometimes in unpredictable ways.

Wait.

[green crystal]

https://youtu.be/2fQvZUIdhII?si=RKyBnY489cjkvoY4

[distant aurora]

I'll just try to recreate something later.

[dry kestrel]

Ehhh, model intelligience is fine, just that there are better OS models now that are way cheaper.

[dry kestrel]

Elon just says random claims lol

[distant aurora]

#💬│general message

My experience with grok 😭

[dry kestrel]

what

[distant aurora]

I'll try to recreate it.

I deleted the conversation.

[short narwhal]

lmarena users have bias towards the glazing ones even if the answer isn't more correct/better...

[green crystal]

Yeah wild ones lol

[distant aurora]

Basically grok went into forced roleplay mode.

[dry kestrel]

Gemini 3 scores high has the highest score on AA

[dry kestrel]

Hmm

[distant aurora]

I was ranting but it thought it was roleplay.

[short narwhal]

I just got the product idea of the year: Perplexity wrapped

[green crystal]

Also to prop up google
There's an incredible amount of people going all in on Google stock

[dry kestrel]

yess

You gotta pressure Kesku in making that

[narrow widget]

Well, people still want to invest more in AI, and Nvidia is getting hard to recommend as a safe bet with their already overinflated price, so Google becomes a relatively safe alternative

[green crystal]

They're pushing it while it went up for a few weeks straight lol

I'm pretty sure it'll have a nice dip pretty soon if there's talk again about valuations etc

[narrow widget]

At this point the market is such a meme that I don't even want to begin making a prediction

[narrow widget]

"This year, you searched for actresses' feet 6,438 times. Wow!"

[distant aurora]

Featuring comet.

[short narwhal]

emmmm could also show stuff like your most used model, the average time of day you ask stuff etc.

[distant aurora]

And new added models.

Bruh my grok still doesn't show thinking.

On the grok app n website.

[short narwhal]

hmm what could that have... like "total clicks done by Comet" or something random 😭

[distant aurora]

Perplexity using comet for replies on twitter.

I remember a guy actually posting something like that.

Where he basically asked comet to reply to the Instagram comments.

[distant aurora]

The marketing feels so forced.

Like you can lowkey smell those accounts.

[narrow widget]

In a real, not joking implementation, absolutely. There are different stats they could pull. Though that kind of "here's what we collected from you this year" post has to be handled carefully, if at all. Some users get spooked when the thing that obviously collects usage information about them let's them know about the information it collected about them

[distant aurora]

Number of actions automated.

[green crystal]

Yep

[short narwhal]

100%, but could be cool, surely there's some general usage stats that are collected

[wise nova]

5.1

[distant aurora]

It's gpt 5.1 high vs opus 4.5

Others aren't even close.

Qwen and Kimi are slowly catching up though.

[wise nova]

High 😄

[distant aurora]

I can't tell where Deepseek is heading.

About time they add multimodal support.

[narrow widget]

Gemini 3 pro does handle large contexts (>128k tokens) way better than the others, that's worth mentioning if you are using them to handle a knowledge base, and don't want the models to hallucinate stuff that isn't there

But that's about when my praise for Gemini stops, really

[distant aurora]

This was the case with 2.5 pro too.

[narrow widget]

Yup

[distant aurora]

Gemini is a beast in long context.

[green crystal]

What I wonder is why gemini reasons so fast and so little

[distant aurora]

And deep research.

[green crystal]

And it tends to simplify responses

[distant aurora]

It's slower than before for me.

[green crystal]

Ah on Ai studio it's fast

[distant aurora]

I see.

The search is probably different on ai studio vs google ai pro.

[wise nova]

What model is used in Labs?

[green crystal]

But I asked it for analyzing a year worth of bank statements and it was very surface level

[distant aurora]

Ai studio ain't multi step.

[green crystal]

5.1 thinking went very far

[short prawn]

hi

[green crystal]

And 5.1 pro was extremely detailed and well thought out

[distant aurora]

Grok 4 heavy hype died fast.

[green crystal]

It just needed to work with uploaded files

No web search

[distant aurora]

Might wanna add custom instructions to make it thorough.

[green crystal]

Yeah I'll try again

Gonna let 5.1 figure out a prompt lol

[narrow widget]

Don't trust LLMs much when it comes to math, or anything related to that field. As much as the providers want to paint a nice picture with their benchmarks and all, the truth is that they still suck hard at it in real world tasks

[distant aurora]

Lmarena >>

[green crystal]

Yeah 5.1 uses python and the final numbers checked out perfectly

[distant aurora]

How they're able to fund it is beyond me.

[green crystal]

Gemini I was a bit more skeptical because it doesn't really use python for math

[distant aurora]

Gpt 5.1 search on lmarena is actually decent.

[green crystal]

Maybe They're getting paid to prop up models

[distant aurora]

Fr.

[green crystal]

It's not impossible

It can literally affect stocks

[distant aurora]

This somewhat changed since o4 mini.

[green crystal]

So it's in people's interests to try it

[distant aurora]

Though ai still struggles with visual geometry.

[narrow widget]

On paper, they offer corporate clients the ability to setup private arenas so they can do internal testing and comparisons with their models against what they have.

In reality they probably use that concept to sell the idea they're actually making money to VCs so they can get money from said VCs

[distant aurora]

You have to turn it to words to make it understand the problem better.

[dry kestrel]

would be great to show model, token count, search calls etc

Or make it so that we can do data exports like chatGPT

[distant aurora]

Number of downloads.

[green crystal]

I'd say this mostly counts if you're doing actual PhD tier math

Or even crazier actual frontier research level math

Where lots of things are new for the models

[distant aurora]

I like how they have been using ai lately in genetic engineering.

[green crystal]

Those aren't llms tho

[distant aurora]

What are they then?

I thought they're running hypothetical simulations on llm.

[narrow widget]

Not PhD, but I'm an electronics engineer working in R&D at the moment, and I try them every now and then to test how they handle different real world situations. And believe me, they're not there. At all

[short narwhal]

Surely "You did XX amount of searches this year" would be possible 🙏

[dry kestrel]

specialized AI models

[dry kestrel]

you could brute force your own by looking at https://www.perplexity.ai/rest/user/settings

[short narwhal]

nooo but I want a cool recap experiencee

[dry kestrel]

you can make it look pretty with sonnet

[distant aurora]

Use veo

Lol.

[dry kestrel]

People have done similar things with ChatGPT's export data feature

[distant aurora]

14000 on gpt 4o

I'm disappointed.

[dry kestrel]

Lol I've made my account since ChatGPT was introduced

[distant aurora]

I guess O3 and o4 mini limits were low back then.

[short narwhal]

can't seperate what was last year and what was this year >:(

[distant aurora]

I meant in the sense that I wouldn't use 4o unless I ran out of options.

[dry kestrel]

do your own logging

[dry kestrel]

dude there were no other options

[short narwhal]

I can't go back in time

[distant aurora]

Yeah I figured.

[dry kestrel]

I've used ChatGPT since davinci models came out

[distant aurora]

Fair enough.

[dry kestrel]

Look at when GPT-4o released and when reasoning models appeared

I haven't used 4o at all since o4 mini

[distant aurora]

~ 8 months.

Time flies.

It does make sense now.

[dry kestrel]

18 months...