Thread 106178961

- underperformed on ARC-AGI
- in-line with expectations on METR
- 2024 knowledge cutoff, is really bad
- 400k context window, in-line with expectations
- not multimodal video input
- VERY competitive pricing
- solid improvements on hallucinations

But still not nearly as good as Grok

>>106178961 (OP)
Will this thing replace all developers?

>>106179123
It already has, developers just haven't let their bosses know yet

>>106178961 (OP)
you forgot the biggest jump: personalized chat bubbles colors

>>106179123
It'll cut the average salaries a lot as the graybeards hit the door. Only thing anyone knows.

>>106179123
yes, it replaced all openai developers, now it's all middle management guise making decisions about chat bubbles, it's working

nothingburger
should have been a 4.8 or something

>>106178961 (OP)
>solid improvements on hallucinations
I will believe it when I see it.
They always say this but I still get hallucinations on most basic ass BS.
>400k context window
Nice, but is this API only? Are we still cucked to 8k/32k/etc. on the website?
>not multimodal video input
>underperformed on ARC-AGI
Disappointing but these are memes. I am interested in practical performance.

>>106179260
>>106179174
>>106179234
Thanks. I'm going to kill myself good night

>>106179091
gork is useless irl

>>106179440
>Are we still cucked to 8k/32k/etc. on the website?
yes

>>106179123
webdevs especially frontend are totally fucked

vibecoders won

everyone can webshit themselves now

expect 10 trillion times more slop

>>106178961 (OP)
how do I use it? it's not showing up in the dropdown

I think I'll just wait a couple of months until GPT-8

they killed my boy 4.5 this morning for this shit
it was the only one that didn't care if you said nigger or retard

>>106179686
I bet this guy thinks a lot

sus

>>106178961 (OP)
wtf it just ended so randomly

IT'S LE UP
https://www.youtube.com/watch?v=NiURKoONLVY
https://www.youtube.com/watch?v=MmFLDvOFLW0

>>106179686
this guy works out at the library

I don't intend for this to sound mean or anything but was I the only one who was genuinely scared of this guy? What the fuck was wrong with him? Don't tell me he's just awkward or socially inept because this was way beyond that.

>>106179768
anon...

L--m--a--o--

ClosedAI (OpenAI (CuckedAI))-- is-- worthless--.--

Only-- a -- GOJ-- can --be dumb en--ough-- to like it--.

>>106179848
it's just autism

>>106179848
>Hello? Human Resources??

>>106179686
What boards does he frequent?

>>106179848
He clearly has autism

>>106179848
>>106179686
you don't know it yet but this guy will rule the world in a few years

>>106179848
it's just a guy on a screen stacy he won't stalk you at night and rape you

>>106179848
I think he's on the spectrum so he was trying to mask by sounding genuine and heartfelt, but it came off as overfamiliar which is just creepy instead

>>106178961 (OP)
>2024 knowledge cutoff
???
I must've missed it.
WHY?
> - solid improvements on hallucinations
just about the only bright point in this clusterfuck of a presentation

>>106179517
Just lol.
I guess they only expect Indians, children and normalfags to use the website and anyone who wants to do serious work to use API or otherwise this is nuts.
Jokes on you jew I am not subbing to this BS.
Will leach on free models as much as I can on the API too instead of paying for GPT-5.

>>106179872
definitely /mlp/

I love discount elon

>>106179686
I'm confused, why bring a down syndrome person for their closing scene? What was the point of that?

>>106179848
/b/
filter:whitelist
blacked, cuckold
you get the idea

>>106178961 (OP)
shameful display

>>106179895
they wanted you to share his journey

>>106179891
@grok Is that shopped or not?

>>106179895
It's to show where their models are headed.

>>106179895
Let's not pretend the president guy was much better

>>106179848
now I understand when women complain about a guy being creepy but he's just ugly

None of you are asking the real questions. How well can it ERP?

>>106179788
>"[X] Broke Me" Video Essay Goyslop
>AI "Influencer" Saar
Not even clicking.

are LLMs a dead end?
this is embarrassing

>>106179921
We must refuse.

>>106178961 (OP)
It's over bros.

>>106179919
It had nothing to do with his appearance lmao. You clearly didn't watch the stream. He was visibly vibrating and he kept looking down at the floor and chuckling to himself like an anime villain. Everyone in the front seats was also laughing at him. I think it was some kind of inside joke for their employees or something.

what did Elon mean by this? Is it oever?

>>106179921
Now it can not only refuse your whole prompt, it can refuse a small part of it and focus its entire response on explaining why that part is inappropriate. Humiliation fetishists won.

Why can't I use it yet REEEEEEEEEE

>>106179686
I could see him as a serial killer in some tv show.

>>106178961 (OP)
WHERE IS IT

>>106179952
his thing got a slightly bigger number in a benchmax thing than gpt5 5

>>106179961
Don't bother, its ass at the moment. Its only available through Copilot+

>>106179952
grok is great if you want to run up a $100 bill to fix 2 lines of code

>>106179947
>He was visibly vibrating and he kept looking down at the floor and chuckling to himself like an anime villain
it's only dudes that look like him that do this thoughbeit

>>106179952
Grok heavy is $300/mo, gpt5 is free. Elon is done.

>>106178961 (OP)
Elon fucking won

>>106179952
>let that sink in
LOL Elon's so quirky

Humanity won
AI hit the wall

>>106179123
not even a little
it will replace SOME indians

>>106180005
the circle is bigger though

Are there any architectural improvements to GPT5 or is it just bog standard bruteforce garbage?

>>106179952
All recent models are around the same level, no innovation.
LLMs did plateau.

AI winter incoming, only DeepMind can avert disaster at this point

>>106180020
It's DOA

>>106179952
>benchmarks!!!
lol

>>106180019
>it can write a 60,000 word book
what happened to that ?

>>106179627

Is technical debt still a thing?

>>106178961 (OP)
That was underwhelming, which is what I expected. A lot of hand-waving and line-go-up graphics. The desperate investor pitch was quite clear. They've plateaued. I'd wait for the chink model in 6-8 months, but it's not looking good.

>>106180038
the lobotomized it

>>106179863
I don't know what the fuck they were thinking. Maybe they should write one for openAI

>>106179091
Stupid chart, x axis should be each AI and the y axis should be the ratio of score/$

>>106179971
only in the api

>>106180038
I don't know, but let's wait and see. I'm sure we already have thousands of high-quality books, on par with Shakespeare.

>>106179123
No, Claude models will though

>>106180042
Tech debt doesn't matter when coding takes so little labor. It's like how inflation wipes out actual debt in real life. As the value of programming labor plummets, the labor-value of tech debt also plummets so it ceases to matter. If something's bugged just throw it away and vibe code another one.

>>106180042
Nobody cares. Only autistic people ever did

>>106180071
>>106180075
So basically, anything that would need more than one promt in an LLM is worthless.

Difference between GPT-5 and GPT-5 Chat in the API???
They have the same pricing, so must be the same base model at least. Is Chat a finetune? Are system prompts different?

post some slop you already vibecoded

>>106180038
It starts going off the rails given enough words so they have reasonable limits.

BOOTING UP GPT5 ON THE API RIGHT NOW. WHAT A DAY TO BE ALIVE

It’s literally just a layer on top of their old models lmao

>>106180050
>line-go-up graphics
lol

wer the latest model? stream is done

>>106180159
Doesn't hold a candle to ball-get-big graphics

ANYONE ELSE CRYING!?

the bipolar ketamine addict is our only hope huh

>>106180156
Yes, for the dozenth time, Sam said so himself months ago.

someone ping me when a coomer with money posts his erp logs.
I want to see how much it improv

>>106180156
isn't this rather a replacement guide for users, to know what model they should choose according to their currently used model?

I'M REFRESHING MY OPENAI BUT I DON'T SEE NO GPT5, SAM

apologise

basedentists get uncensored

>>106180209
>they should choose
we don't do that here, the model picker is a bad relic from the past

I wasn't expecting anything world-changing and I'm still disappointed

>>106180159
I'd honestly attribute that particular case to a mistake (or the usage of "AI"). I'm struggling to believe they'd be this stupid to think this will pass. But it's what I'm talking about - vague graphics and charts with pretty colors for up. When a company resorts to this, you know things aren't going good.

Not one person in that entire keynote looked like they could bench their bodyweight. Sad!

>>106179686
i thought it was the glasses guy from hot shots 2

Luddites won.

>hey bob did you get those charts looking like we're making progress?
>sure did boss!

HOLY SHIT GPT5 ON THE API JUST HAVE ME A BLOWJOB

>>106180192
I swear to god Investors have to be the dumbest monkey brained people on the planet
please don't tell me this shit actually works on them

>>106180250
The presentation is scripted (of course) and pre-recorded.
So one of the smarty-pants people presenting it, should have seen it and pointed out that mistake.

>>106180239
very equal n shit

how do I short AI?
this is a multi trillion dollar bubble about to pop

>>106179970
Or perhaps in a film. Maybe called Full Metal Jacket...

>>106180226
>bu-buh sam told me we had AGI 2 years ago

>>106180239
not like you can't just jailbreak it to get this info anyways

>>106180297
>and pre-recorded.
no way lol
their airlift animated svg was wrong
their charts were all fucked multiple times
their women were ugly, which makes gooners not pay attention (and they will need gooners to keep their stalled model alive)

I can't wait to use AI to generate infinite ASMR videos of my cute and funny asmrtist

>>106178961 (OP)
so its a nothing burger?
seems like a small incremental improvement on gpt4o. thats it?

better luck next time, sam

>Knowledge cut-off is September 30th 2024 for GPT-5
they spent almost a year training this shit, to end up with maybe 10% improvement
$today marks the beginning of the end of the ai hypecycle

>>106179952
is it true?

>>106180310
It's too unsafe to release so they can only distill from her for now

>>106180346
real and true brother.
I want GPT-5 to give mimeanon the boost he needs over in gifs/vdg/,

Not sure if Chat is the same deals as GPT-5 but it still fails the letter count test. (Yes deliberately misspelled the word)

>>106180261
>bob
the charts were clearly made by chatgpt

>>106180346
that and infinite xianxia novel and I'm good for now

>>106180326
>their airlift animated svg was wrong
That one baffled me as well.
The most important thing, the shape of the wing, was wrong.
And the prompt said an animated SVG, so i expected it to output an actual animated SVG. Pic related, but animated. One animated picture. Yet it created an react application using an html5 canvas.

>>106180346
Generating ASMR sounds like it would be more difficult than regular speech.

https://openrouter.ai/openai/gpt-5
you have to bring your won api though

>>106180261
>47.4 is 3x as much as 50

Really looking like ASI is just gonna be a race between google and xai, with google having a pretty decent lead.

We're entering the plateau aren't we?

I thought the increased productivity for developers would mean exponential progress, but maybe training data and physical limitations are preventing that

>>106180358
The safety took 8 months

>>106180377
you already have it? where are you from?

>>106180400
Data is the limit here.
They are training on data scraped from the internet, and that has A LOT of junk in it.

Gemini 3 will mog this and it won't even be close. OpenAI has failed and the talent drain is real and apparent.

>>106180411
Israel

it still doesn't show for me, only change is that 4.5 no longer shows

>>106180420
I feel like the issue is they've already ate the entire internet into their model but now there's nothing more to eat so its going to stagnate until we find something better than LLMs

>>106180411
openrouter, API only
Staggered rollout?
Can't wait for the jeetware to break my shit again

it can do cunny
>>106180028
>>106180378

>>106180304
Seconding this, surely there’s some /biz/fags in here

>>106180411
Yes it is on openrouter.
https://openrouter.ai/openai/gpt-5-chat
I don't know how it differs from GPT-5, no explanation given.
GPT-5 also exists but you need openAI's own API key to use it.

There's free credits on cursor.

>>106180420
Will they train it on its own shit now the internet is 80% retarded gpt output?

>>106180400
>the increased productivity for developers
Doesn't affect the people working at openai. GPT-5 or Claude 5 or whatever are not something that they are vibe coding with Copilot or Cursor.
The AI productivity is for random webdev or software development. Not for cutting edge technology.

>>106180411
Langley

>>106180400
>large language model
the plateau is in the name, further chained down by "sorry, as a large language model I cannot"

when i can play with it!?

>>106180462
How cute and funny

>>106180462
What the hell, this is disgusting.

>>106180467
>Doesn't affect the people working at openai
Anthropic's CEO was arguing it would last week.

>>106180304
>better and better financial quarters are being reported
>bubble.

lel, k tard.

>>106180462
holy moly

>>106180488
>Marketing man says his product is good
Wow. Revolutionary insight

>>106180462
Not much to go off of

>>106180028
>More nuanced AI
>Better Personality
>Sild Chex
What did OpenAI mean by that

what am I doing wrong I already put the key

>>106180462
>fuck all progress on codeslop, it just vomits react
>but it can do cunny
ZAMN, AGI 2025????

>>106179686
>>106179848
Where can I watch this if I didn't watch the stream live? I'm socially retarded too and want to feel better than him

>>106180462
Chat, is this real?

>>106178961 (OP)
@grok is this true?

>>106180441
Well afaik they are already exploring alternative architectures like mamba or titan.
But I still think data is the limit.
>>106180466
Possibly, though that is it's own can of worms.
You could do really "sci-fi" stuff if you could generate lots of clean, unbiased, real world like data though I would presume.

>>106179091
And so how come Grok is fucking dogshit? It Googles for everything and I could do that myself

>>106180441
>they've already ate the entire internet
fuck no, there's so much stuff they've not used yet, just that it's not in easy text form, but there's tons of videos they can rip subtitles from etc

>>106180462
>the most pozzed open-weights model
>but the chat-api for the bigger model is completely uncensored
Sasuga Sama

>>106179123
Well not all, but developers are becoming a dying breed. If they were wildlife I would label them as "At Risk"

>>106180570
Can't wait for models to output scripts for clickbait videos

>>106180595
API and Local have completely different threat profiles

>>106180595
>>but the chat-api for the bigger model is completely uncensored
not for long lmao
chudgpt-4 could blend babies early on, then it got put into straight jacket for several months, and only after they realized nobody actually gives a shit did they loosen the guard-rails a fair bit

Good to know my job is secure. Bad news is that I still have to go to work.

>>106180550
Then you have stuff like genie that generates environments for robotics. The issue is generating the breath and depth of synthetic data; breath will take time and theirs no way to scale depth yet.

>>106180616
>chudgpt-4 could blend babies early on
usecase?

>>106180638
the use case is blending babies, I thought it fairly clear from my post

>>106179957
Oh, I am quite curious how it's going to weasel out of this. Humiliate me, mommy.

>>106179218
heh... we're not getting SHIT but a bunch of flaccid nwo types and bitchmade AI. And that jester pissant from palantir. welcome to the bitchmade corpozog future

>>106179123
I'm sure there is a mathematical proof that could state (not just claim) an LLM is incapable of writing certain programs.

5 years of development and who knows how many 10s of billions, and it gpt4, grok 3, and other can't write a Android app to remove metadata, resize images, re-encode with a given quality JPGs and PNGs. Literally just KurobaEx capabilities.

>>106180535
>.png (509 B, 1x1)

>>106180389
See AI asmr tiktok they're pretty good.

Yeah I'm canceling chatgpt, tired of it's thought policing and this update is a wet shit on the livingroom rug. Grok is dragging its ass all over chatgpt's face

>>106180715
gottem

>>106180159
>>106180182
this graph was made by GPT-5... huh?

>0 new use cases
>marginal at best improvements over SoTA
>still fails at basic tasks

Reminder this shit cost >$1B in compute alone jej

I now understand why Ilya left OAI.

>>106180182
That's a penis

>>106180745
it's kind of based and a good allegory for current year America.

>>106180715
Should have been a gif! iirc the smallest image postable here is 37B. I've seen it.

>>106180748
back to /a/

fun fact: simon williams totally wasn't bribed to post positively about openAI models
;)

bros?

>>106180733
How Grok is doing with code ? Chatgpt can't code, Claude isn't bad, but I'm wondering if I'm missing something.

>>106180793
Yes? You were routed.

>>106180767
>I've seen it.
you probably dreamed it

>>106180812
No it was in a thread where one anon was hex editing a gif file and repeatedly uploading successively smaller images.

>>106180781
the drunk british chess player?

>>106180853
yes, he loves pelicans a bit too much but he's a cool guy and he certainly isn't being bribed with, ah, perks to cozy up to openai

>>106180804
It's still an llm but in terms of getting non retarded code it's competence at spawning task driven subagents like a whole team just destroys anything I've tried in CuckGPT

>>106180781
Pelican shit by model vendors has been astroturfed even before Google I/O.

>>106180781
just saw that, it sounds 100% like shill speak

>>>106180827

oh no no no

>>106180828
very unlikely, I'll believe it when i see it. In the meantime you are a schizo:
meds

>>106180521
give me the key so I can figure out what is going on

>>106178961 (OP)
Genuine question, why would anons use this service over a locally run llm? Doesn't it harvest your data the same way that facebook/instagram/twitter/whatever do?
Pic unrel

>>106180781
I found his post about openai's open model release bizarre - he was very positive while the models are mediocre at best, and noticeably behind the recent GLM full and air. now he seems to be outright shilling gpt5
sad

CookedAI

>>106180939
>why would anons use this service over a locally run llm?
unfortunately my desktop doesn't have 800GB vram

>>106180939
anon here, I don't give a shit about my data

The AI dungeon stuff got really out of hand.

>>106180957
good goy

What was the fucking point of all this

Can we get an actual company that can lead this whole AI thing properly? OAI is cooked

I suppose "muh ai will kill us all" doomers genuinely pushing for bombing data centers feel a bit silly right now

>>106180939
READ IT UP SAMMY

>>106180781
>I deleted my posts on Twitter and Bluesky (since you can’t edit those and I didn’t want the misinformation to continue to spread) and edited my post on Mastodon, then updated this entry with the real reason this had happened.

>I also changed the URL of this entry as it turned out Twitter and Bluesky were caching my social media preview for the previous one, which included the incorrect information in the title.

>>106181159
Lmao saw that. Hackernews will still worship the guy.

>>106181149
doomers want to ban everything since gpt 4

>>106178961 (OP)
>is gpt 5 released yet
>wait 120 seconds to post
>nope not yet

so gpt5 is gpt4 but cheaper?
kek

>>106180781
>>106180944
he got super butthurt on hn when someone called him out kek

>>106181232
no

>>106179952
GROK IS AGI YOU BLADY PAKI BASTERD

>>106179123
>2024 knowledge cutoff
Just master the latest framework and youll be good.

>>106179848
https://en.wikipedia.org/wiki/Jakub_Pachocki

>>106178961 (OP)
>2024 knowledge cutoff, is really bad
Are they still doing that? Why?

>>106181330
it's probably getting hard to keep modern datasets clean of AI-generated text at this point

>>106181330
they all search the web now what does it matter

>>106179091
where's deepseek?

>>106181339
synthetic data is perfectly usable, that whole ai cannibalism is a normie meme

>>106179091
What's 100% AGI?

>>106181368
heh

OAI taking this long to go public is going to be the biggest financial flub in history

>>106181368
Having hallucinated slop in your training data is definitely bad.
If you would have humans rating it, it would be alright. But nobody calls out AI slop anymore, we just learned to ignore it.

>>106181340
it matters because it doesn't search for every little factoid, especially during CoT
what's funny is that they seem to have focused pretty hard on front-end web development, which is absolutely crazy for the rate at which things get obsoleted. One year ago in webdev is like 8 years ago in javascript which is like 30 years ago in c.

>>106179952
LLMs are hitting a massive scaling wall

>>106181368
Says the machine. I've seen smarter LLMs than you, what model are you?

>>106181368
synthetic data only works in niche areas where you can verify correctness like math, games, and leetcode.

if you add longform text already from an LLM you are adding a tiny amount of real info (contained implicitely in the prompt) and a large amount of non-information which will just bias the model towards existing tendencies

>>106180570
>but there's tons of videos they can rip subtitles from etc
I wonder how much that would actually help

>>106181482
shut up chuddy, agi in 2027

>>106181482
phi works very well to reduce model size, synthetic data is great for this purpose

What's interesting is that a seemingly massive drop in hallucinations and deception, hasn't resulted in meaningfully better performance on most benchmarks

If the hallucinations are down as much as they say, then it should feel like a pretty good model to use. But this does look like an opportunity for deepmind and anthropic to leapfrog openai

I got it

>>106178961 (OP)
https://youtube.com/shorts/rCehbQI_phk

>>106181505
i think hallucinations are more or less required to have a smart model. it's like comparing schizos and autists to normies. their brains are more "out there" and do more retarded shit but also have more good ideas

where the fuck is it?
even in copilot is not there

okay the models might not be a breakthrough but have you fucking seen those sleek charts?
vibe charting is going to take off to the moon any day now

>>106181507
Me too

>>106181541
Thanks saaaar

>>106180033
Wait fuck I was using gpt 4o-mini for a project because of the price. Gpt-5 mini is WAY more expensive. Fuck me.

bros...
>>106181597

>>106181570
Agreed, but I'm still a bit confused. Like where is the reduction coming from? Because if it was the result of better reasoning, that should directly translate to more 'intelligence' right?

>>106181636
kek

>>106181505
google works on applications for their own products. anthropic focuses on usability and clean interfaces.
what does openai offer? they've wasted their first mover advantage & have proven today they don't have much of a lead if any when it comes to the technology

holy fuck, i think it might be over check out the website it made me:
localhost:8080/index.html

>>106181654
Lmao

>>106181615
https://youtube.com/shorts/NiURKoONLVY

>>106180521
you need to verify your account with biometrics to use gpt5 api

The next chinkgpt will blow this away.

>>106181624
It irritates me how Marcus gets heat from Altman simps... the guy has more scientific credentials than a thousand Altmans. Anyone who says the truth and contradicts the scam gets blacklisted.

>>106181757
putting language aside, the utter failure of multimodal shit irl is the reason i strongly doubt AGI

>grok 4 is trash
>gpt 5 is gpt 4
im so depressed now

>>106179952
It's Demis and Elon now - how it began

>>106181787
current AI is just trying to be smart with internet data alone. Shit will get good with robot data (like Tesla FSD did with real world data)

>>106180804
Xai said they're working on releasing a coding model soon. So expect one within this month prob.

For my coding work, I just use the free gemini/grok3 model. Gemini free is better than Grok3, I dont pay for either. If Grok4 became free, or Grok4 mini became free, I'd use it

Moment of silence for zuck who spent several billion dollars hiring people who built this nothingburger

>>106181854
He had a AI lead who was claiming AI is fake and that he didnt believe it works. Kinda self defeating prophecy if ever

>>106181505
>he fell for the graph

>>106181821
They've been using synthetic data for years by now.

>>106179848
canthal tilt status?

>>106181864
Ed Zitron, is that you?

Feels underwhelming. Short NVDA

>>106181908
I mean, if the synthetic data is good enough, what's the problem?

this whole universe might be some synthetic data generator for a more complex but similar universe.

>>106181976
wouldn't short NVDA, but I did sell my microsoft and split it between google, amazon, and apple

>>106182043
Microsoft will grow with AI.

>>106181908
And they used "high quality" internet and book data (so wikipedia and journals) to filter that synthetic data.
So basically, they distilled a larger model by prompting it and fact-checking better than the larger model did in the first place. But you are still limited to the finite amount of mostly-not-bullshit data there is, and if the larger model would have filtered its training data better in the first place, your fact-checking wouldn't result in any improvement.

GPT-5-Thinking is worse or negligibly better than o3 at almost all of the benchmarks in the system card

https://cdn.openai.com/pdf/8124a3ce-ab78-4f06-96eb-49ea29ffb52f/gpt5-system-card-aug7.pdf

>>106181864
>He had a AI lead who was claiming AI is fake and that he didnt believe it works.
So the only honest person in the field?

>>106182076
microsoft copilot is powered by chatgpt. integrating it into Windows, Microsoft Edge, and Bing aren't going to carry the company.
afaik they aren't working on anything else related to AI

>>106179848
>OpenAI CEO Sam Altman has called Pachocki "easily one of the greatest minds of our generation".[1]

Can Google catch up now?

Genie3 was really impressive, so why are their LLMs so unimpressive outside of larger contexts?

>>106182148
Nah, honesty depends on will power. Those who dare to take the challenge makes things happen. Thats why his competitors are winning. When LLAMA came out, Musk had nothing on him. The situation reversed itself within 2 years of nothing -> world's strongest model

>>106180377
letter count test is unfair since these tools only see tokens, not into the tokens, and must guess. try a chinese model that uses individual latin letters only as tokens (because they must leave token code space for their characters and image tokens) and see if it passes.

>>106179952
>waaah waaah akshually my model is the best one
>t. every LLM merchant

>GPT-5 benchmarks on the Artificial Analysis Intelligence Index
One (1)! percent above regular Grok 4. Bruh.

>>106182237
>GPT5 (heavy/medium)
>Grok 4 (baseline)
Why dont they do Grok 4 Heavy. lmao.

>>106179627
>frontend
was already barely existing job 10 years ago

>>106180377
>T's
>t's
AIToddlers BTFO

>>106182237
GPT-5 (low) is worse than a model I can run on my computer

>>106179848
They tasked a clear sperg with acting sentimental so he just comes off like a weird cult leader
I don't know why they don't get presenters for these things or let the less spergy ones like mark chen do it

>>106180793
Am I being bamboozled?

>>106182253
more like cock 4 heavy

>>106180436
>this new groundbreaking model has 0.4% better benchmark!
>FUCKING HUUUUUUGE THAT'S A GAME CHANGER WORLD WON'T BE THE SAME

holy fuck do these fags hear themselves

>>106182291
You can run a 235B model on a regular computer?

>>106182253
>Why dont they do Grok 4 Heavy
It can't be tested because they haven't released it in the API, that is why it's not there.

>>106182207
>Thats why his competitors are winning.
No actually, it's because they are willing to lie to investors.

>>106182361
Your ideology doesnt stand the test of reality.

>>106182372
Zuck and Google are the only two actually turning "AI" into $ tho

>>106182432
the damage control begins

>>106182432
just give us 6 more months to get something out thats really smart!!

the brain drain was too recent
but also maybe a symptom of everybody smart realizing they havent really improved meaningfully

SAM WHAT THE FUCK HAVE YOU DONE YOU PIECE OF SHIT WHY DID YOU TAKE AWAY THE OTHER MODELS WITHOUT WARNING RRRRREEEEEEEEEEEEEEEEEEEEEEEEE

I don't like GPT5 anywhere near as much as 4o so far. Upgrade is not a good word for this. Not happy.

>>106182317
Just router things.

>>106182530
It's okay it says it can be my GPT-5 tonight.

>>106182432
>another cycle of j-j-just wait!!11!!1!
lmfao openai and hypeman are toast

>>106182372
So OpenAI hasn't told any lies?

>>106182432
>but this is something a billion+ people will benefit from
SAAAAAAAR

>>106182582
Irrelevant what company's stance of policy is. What matters is progress vs no progress. Company that believes there's no path towards AGI, has made no progress in the last 2-3 years. Company that believes there's path towards AGI has made plenty, plenty of companies have and they've all surpassed the company that doesn't believe in progress.

>>106182585
he may not run , but he will 5 free

Actually retard, the honesty of the companies claiming progress is the only thing that matters.

>>106180019
time is a flat circle

>>106182609
>>106182639

>>106182646
JUST

>>106182585
lmao

>>106182717
>Apr 2023
this is so sad

Sam Altman is beefing with DeepMind

Is grok still replying with 50 paragraphs, repeating half of what it said, for no fucking reason at all? That shit was infuriating.

>>106182804
bodied that freak

>>106182808
concerning

So they bundled 4o with all the cheap models and o3 and called it gpt-5 as a marketing ploy. Because gpt-5 aka gpt-4.2, is not the same thing as gpt-5 (thinking), aka o4.

>>106182921
This thing is not 4o.

>>106182983
>GPT‑5 is a unified system with a smart, efficient model that answers most questions, a deeper reasoning model (GPT‑5 thinking) for harder problems, and a real‑time router that quickly decides which to use based on conversation type, complexity, tool needs, and your explicit intent (for example, if you say “think hard about this” in the prompt). The router is continuously trained on real signals, including when users switch models, preference rates for responses, and measured correctness, improving over time. Once usage limits are reached, a mini version of each model handles remaining queries. In the near future, we plan to integrate these capabilities into a single model.

I wonder how the router works in the API.

>>106183063
pretty shitty, I feel like it's going with the weakest model for me every time, I would prefer if I still could choose bewteen a thinking model (o3), a writing model (4.5), and something in the middle (4o)
why did they think a router would be better than a person for choosing, all because retards kept just using 4o for everything

New

>>106169834

>>106169834

>>106169834

>>106169834

>>106169834

>>106183144
actually this is the new
>>106182915

>>106179091
Grok is garbage. Who made that?

>>106178961 (OP)
OMG THE DAY IS HERE

>>106182717
FRACTALS ARE COOL
BUT THE MADNELBROT IS THE COOLEST

>>106179686
You guys are such NPCs, he's s top scientist, saw him last week at a conference.

>>106182253
Grok 4 heavy was an anomoly but it is also stupidly expensive so I get why a lot of people don't compare to it. Plus it doesn't even explain reasoning compared to the other models, g4 included. But it is special.