Thread 106202041

It's over
GPT-5 is AGI

>>106202041 (OP)
>Actually Genius Indians

>>106202041 (OP)
Fake & gay

Sirs please buy the AI very smart

>>106202124
SUPERPOWER 2020 SAAR

>>106202041 (OP)
I have an IQ of 135 though.

>>106202041 (OP)
Lmao

>>106202041 (OP)
Meanwhile in real life:

>>106202481
>WATERMELONR

when asked to think, does the free version of gpt-5 always route to the low thinking model? or does it also route to medium and high when prompted?

>>106202481
i asked chatgpt what sylvia plath had to do with fight club and it hallucinated marla singer talking about plath. when in reality it was the narrator who mentions plath.

it was a very basic question for a popular movie and spectacularly failed. it's kind of black pilled me on llms for now.

>>106202481
Holy shit lol

>>106202481
I've been pronouncing WatermelonR wrong all this time?

>>106202628
The R is silent.

>>106202638
just as my penis

So when are these LLMs going to replace the majority of white collar jobs? It seems like every recent major release has only minor improvements. I asked GPT5 and all it gave me were projections from bullshit consulting firms that obviously don’t know anything about the underlying tech and just want to sell products/services.

>>106202750
Deepseek r2, where is it?

>>106202750
probably never
other types of ai might
llms will probably be components thereof
but standalone llms are fools gold

>>106202481
strange mine works just fine.

>>106202481
>>106202573
I love ai hallucinations, pls post more

>>106202041 (OP)
yeah, totally

>>106202376
>Einstein's was 160
I read Epstein instead of Einstein at first kek

>>106202376
If AI is so smart, why haven't we cured cancer? I don't get it

>>106202909
its a significant step along the path to AGI

>>106202124
Actually guided by Indians*

>>106202481
qwenchads stay winning

>>106202376
2 more weeks and 30 billion more dollars

It's funny how shit AI is and how much it has helped me to study during my degree at the same time.

I think this says a lot about society

>>106202911
what's the difference

>>106202998
one of them was based and the other one was a physicist

>>106202868
>Dave
kek, making sure he doesn't make any mistakes this time huh

>>106203037
no I told my gpt to call me by my name, which is Dave.

>>106202376
>write first paragraph
>get AI to generate rest of text

GPT-4o-mini btw. I really don't understand why people have botched GPTs that say 2.

>>106202041 (OP)
>Slop generator has arbitrary and meaningless label applied to it
Cool, still not using it.

>>106202573
>it's kind of black pilled me on llms for now.
You only just noticed LLMs are trash?

>>106202964
how did it help you

>>106202932
AGI is just a marketing term, midwit.

>>106202953
Nice. Now ask it about jews doing 9/11. It's even more eager than Altman's model to suck off jews.

>>106202376
Everyone on LinkedIn transmogrifies into an Indian. Spiritually at first, and then physically.

>>106203279
I genuinely believe this, not even Facebook boomers reach these levels of mental retardation

fucking useless

>>106202931
Cancer isn't one thing to be cured, it's millions of different things.

>>106202376
>It's speculated that GPT-5 will have an IQ of 1,600
Are people on Linkedln really that retarded or are they pretending?

>>106202041 (OP)
What do you expect? It’s worthless

GPT-5 is unironically the beginning of the end of the AI fad.

>>106202041 (OP)
>It's over
>GPT-5 is AGI
You only think that, using the term "think" rather loosely, because you're dumb.
Aren't you?

>>106203520
Cancer can't be cured but elephants already have far lower cancer rates (than expected for the amount of mass) due to tumor suppressor genes. If we did something crazy like CRISPR splicing to include stronger suppressor genes in humans that would be a massive hedge against cancer. I mean shit, we already put woolly mammoth traits in mice. Just requires identifying which genes are the tumor suppressors and the social nightmare that is actually gene editing humans without people losing their minds.

>>106204029
Exactly what I was thinking earlier.
How will OpenAI even recover from this shit?

>>106202931
Soon

>>106202041 (OP)
should I start using claude?

>>106202041 (OP)
IQ calibrated using Indians.

>>106202124
No way you think you're superior to Indians in a board that can do absolutely nothing creative. Do you think programming will make you relevant in 2026?

>>106202041 (OP)
Deepseek R2 and V4 are going to crash the American economy.

>>106202931
>why haven't we cured cancer?
Every time someone does, a Boeing plane falls out the sky.

>>106204029
I hope so.

>>106204777
we've been hearing for so fucking long how american companies need more chips+energy for next generation models but shit like stargate is taking forever.

hopefully china fucking bends them over even with some restraints

>>106202041 (OP)
>it's retarded

>>106202964
I think it says a lot about your degree's worth

>>106203808
>>106205951

>>106204029
even if this might be true, i just want to remind you that every single economist, finance bro and lots of tech bros admit that the western economies basically right now are build and relying on the bubble that is AI. If the bubble burst, Companies like Nvidia may lose trillions of dollars.

The western economy right now is basically relying on the idea of AI working and being profitable. Even if there are 4 'r' in strawberry or 5 'r' in Watermelon, the idea itself powers the economy. Just like crypto had the idea of replacing banks 100%, so does AI has the idea of replacing all jobs 100% and this alone is 99% of the market value - the idea.

>>106202041 (OP)
so GPT-5 is actually dumber than a nigger

>>106204671
You might want to take a look again at the IQ map, rajeesh

>>106203242
I really wonder why

>>106206220
they are not trying to hide it anymore...
Just ask if if Jews controlling most of the media in the west is problem, then ask if whites controlling the media in China will be problem.

>>106202481
1600 IQ

>>106202041 (OP)
they made a black AI?

The only people that think it is possibly AGI are what generally would be considered retarded.

>>106202041 (OP)
GuPTa-5

>>106203046
sup brother, my name is TARS

>>106203046
>

>>106203242
I remember laowhy86 had a video explaining that the chinese look up to them in a lot of ways

I'm completely fucking retarded at programming(35 years old and best I could do was a hello world and an array in c++) but I was able to make this webapp text editor program that sorts things into categories, saves files, loads from hard drive, uses some formatting shit to create lists and "collapse/expand a block of text" features, and im gonna make a webserver so I can access from anywhere

Next I'm gonna make a program that sorts through my 15000 random images and create a webapp so I can easily sort through it. I can finally find the perfect pepe.

I'm also going to use ai to sort through my 2 decade long text files and find things like health supplements I was wanting to try and to categorize it and explain them for me

The future is here.
Programmers are completely fucked lmao

>>106203520
it always amazes me how it's never about
>how can we prevent getting cancer in the first place?
or
>how to stay healthy
it always assumes you're a fatass that has no choice but to consume carcenogenics

>>106208315
You do have no choice but to consume and be around environmental carcinogenics, we've fucked our environment up. You could probably make a credit card out of the microplastics in your body

>>106202041 (OP)
The more reasoning and user input parameters they add to the LLM the less truthful it becomes but it gets more efficient at manipulating people.

>>106202041 (OP)
>>106202481
Its more depressing that so many people INSIST its actually super helpful in their day to day lives when it fails like this very frequently all the time.

The average person is so fucking stupid its unreal, I know I'm also dumb in areas but the difference is I can tell based on how many errors AI makes on average that it can't be trusted with things I actually do understand nevermind shit I don't.

>>106206137
Yes. We might see a repeat of 08 crisis but now with AI because everyone went all in on it. Yet with ChatGPT you see an over 70% reduction in usage during common summer break periods in Europe and US. Meaning its mostly used to generate papers and give study advice but it mostly stays out of professional workflows.

I don't see it getting traction any time soon either since the fundamental problem remains, it is a black box model that often deceives an end user. I've had it confidently quote a law article and write paragraphs about it. But it made up the law article, the actual article stated something entirely different. This means its unreliable and you'll still need an expert / human in the loop.

>>106208498
It's very useful for Indians who can't write or speak English properly. The one thing it doesn't suck at is writing coherent sentences. It's also a very powerful grammar, spelling and text layout check. It can assist a lot in writing things down neatly.

Actual things that have exact meanings or anything requiring reasoning? It sucks at that.

what is this graph representing, grift extent?

>>106206137
>>106208522
If Nvidea crashes a lot of other tech stocks crash. These tech stocks are used to prop up the US economy. Some of them use that as a coverage for loans as well. If somehow AI fails it could lead to a pretty bad financial crisis. I don't think the US will recover as neatly as it did in 08 since printing more is off the table this time. It's off the table since the debt would then spiral into one where US won't be able to pay back the interest payments.

>>106208541
Amount of mensa question and answers in the training data. GPT5 used much more synthetic data so most unique benchmarks are going to be hit.

the idea that people spend hundreds of dollars per month on these glorified ask jeeves is haunting.

>>106204029
>>106204092
This only hurts OpenAI dipshit. Scorned and burned loyalists aren't going to quit using AI, they're going to finally look at the other companies' offerings.

If GPT-5 is so shit, it's clear OpenAI is no longer the market leader so they could fold tomorrow and the race would still be on.

>>106208549
Dumping synthetic data into a model is typically a bad idea unless you really are having a low amount of data and are confident that you approximately know the distributions that you want to generate. It will not solve the issues with open AI, but it can make it potentially worse. You're either going to create spreads that aren't there with synthetic data or follow a wrongly statistically interpretted trend.

>>106208738
Race for what? I wouldn't call the model improvements a race really. The race is about finding a way of reasoning that is also explainable. Currently most models are devolving into taking user input and satisfaction (thumbs up for response) and tune models to what users like in terms of output. That's making the model better at manipulating people but it's not really doing much for putting LLMs into workflow of professionals.

A second problem of these LLMs is that its quite easy to deploy your own LLM and host it in the cloud. You don't need much tech workers for it. Therefore over time it is likely that universities will develop algorithms that heavily reduce costs and will then host some of these solutions online. They'll likely be just as good as what openai or google can offer.

»That’s not intelligence — it’s just approximation at scale.«

(chatGPT-5 reflecting about itself, after having been pointed out the irony in calling itself »lightweight« while repeatedly failing the strawberry test)

>>106209090
They did it on purpose, right?
There's no way they didn't test this shit before release.

>>106208333
she looks deliciously rapeable

>>106203242
>Now ask it about jews doing 9/11
Any time I look into evidence of jews being involved in 9/11 it's the most retarded lack of critical thinking piece of shit ever and TikTok is full of it now which should tell you all you need to know about the quality of the evidence.
Keep seething, goy.

>>106202481
based ai dabbing on humancels

>>106202041 (OP)
just got this weird feeling and maybe its cope but i think AI is over for now.
it will retreat for a bit, stocks will crash, vibe coding services like cursor will further develop and enhance tech quietly but the general consensus around AI will be one of disinterest for a couple years until a real breakthrough

>>106202573
it's amazing how probabalistic word pachinko machine fooled so many people. I guess the hype machine had a kings new clothes effect too.

>>106202894
we should stop calling them hallucinations and call them errors. it's a machine. don't anthropomorphize things that can't think.

>>106209931
It's a desperate attempt to save money. The premise of GPT 5 was that it would determine which version to fulfill your prompt based on how difficult it is. ChatGPT Lite for a poem about Shrek eating lasagna, Super GPT for coding or legal questions for example. Come to find out it seems to almost always default to the less compute intensive (and less expensive) models to save on operating costs. They're desperately trying to reign in the expense in order to become profitable

>>106210162
>they're desperately trying release something and pretend it's progress so MS continues to inject cash
FTFY

>>106210173
Why not both?

>>106202573
I find this when I search for info about Baldur's Gate 2 items. The google AI summary hallucinates an entire questline with the name of the item.

>>106208498
>>106208532
Pure cope at this point. These systems have already automated away
>generating boilerplate
>translating texts
>proof reading them
At this point you can replace 10 office sluts with 1 motivated pajeet, and he will do a better job. Hallucinations aren't a problem in this context, it's all ordinary shit. Computer programming is next, and that's just the language models

>>106202573
Look none of these things have a memory, everything it outputs is technically a hallucination. It just generates text and sometimes it is somewhat truthful and other times not so much.

>>106210099
>decades later
>retards still can't understand GIGO

>>106202931
taking meds for cancer for years is much more profitable than just curing it once. also even if someone did there is massive gatekeeping from FDA to the point it's easier to buy meth than adhd medicine for actual adhd you have

>>106210099
this looks like what netanyahu is doing in gaza, just sayin

>>106210653
>>generating boilerplate
Do you know what boilerplate is, nocoder?

1600 IQ

>>106211765
>wessex

>>106202041 (OP)
I highly doubt Claude scored 120. The questions must be very similar to questions it's trained on. Claude is fucking horrible at spatial reasoning

>>106202573
>it's kind of black pilled me on llms for now.
Why? Did you actually believe the pajeets that told you LLMs were just like people?

>>106211765
>Ireland is Wales
based

>>106202376
>staikos
I just noticed that american greeks are kinda jew-coded.

>>106202041 (OP)
Why is this shit on a belle curve? What population are you trying to display?

>>106209087
>hurrr professionals aren't using it and giving their own feedback
do you realize how stupid you sound you midwit stemfag?

>>106202041 (OP)
Okay play Pokemon fucking AI

>>106202931
You've seen the kind of javaslop that gets sharted out of AI vibeproompting. Would you REALLY trust an AI cancer cure to be injected into you?

>>106202376
>i'm retarded and don't understand standard deviations
this is probably bait tho

>>106204049
Cancer can be controlled via autophagy.

>>106208738
It's not about the users; it's about the investors who've poured trillions into this shit suddenly being faced with the reality of a garbage product.

>>106202376
>i help companies drive revenue
stopped reading there

i'm convinced 95% of posts on this site are hyperbole, and NO ONE here is actually using GPT-5 to do any work it can do.
at the same time, i'm not going to just trust Sam Altman and all his boosters, but I don't think OpenAI is in the business of humiliating itself. so what if it can't write prose. there are different forms of intelligence. can it write code better than Opus 4?

>>106211765
>>106211864
my favorite games in Crusader Kings involve starting as Wales and conquering Ireland.
megabased.

>>106202931
IP6 + inositol already cured metastatic melanoma (and it has potential for the rest of cancers), doe.

>>106214930
It'd only be the realization that they bet on the wrong horse.

>>106214996
>"b-b-but da code"
Claude already won that market, zoom out dumbass