GPT-6 fez 99,9% neste Teste… Chegamos à AGI?

Well, let's talk about some technology

news that happened in the last few days

. Finally, GPT-6 has been released.

OpenAI launched the model codenamed

Astra, which is GPT-6, and it's truly

quite impressive because the model

manages to be better than Anthropic's

models, so the competition is really

heating up between the two companies.

And there are also a lot of people

saying that perhaps this OpenAI model

is finally an AGI, because there are

some market benchmarks that this model

managed to surpass, which is an

impressive feat. Let's talk a bit more

then about GPT-6, some interesting

things people are creating with this

model, and let's see if this model

really is an AGI. Why are people

talking about this? Well, we'll talk

about this and other tech topics in

this video. So, on September 3rd,

OpenAI released GPT-6, and the

presentation of this model was really

cool, right? I don't know, they put it

on their website looking like a galaxy

with the number six. They really did a

great job with the marketing for this

new model. And look, it had been a few

weeks, even months, that OpenAI had

been dropping some teasers about the

name Astra, suggesting it would be a

new model. And now it has become clear

that Astra was, in fact, GPT-6. And

honestly, if we stop to think about it,

it's crazy, because it wasn't long ago

that we were here in these videos

discussing, I don't know, the launch of

GPT-4 and just kind of fantasizing

about when OpenAI would release GPT-6

and how advanced GPT-6 would be

compared to GPT-4 or other previous

models. So, I don't know, it's

impressive. I don't know if time is

passing quickly or if the models are

evolving even faster. Maybe that's the

point. But hey, there are some very

interesting things, right? For example,

OpenAI released this video here showing

, I don't know, this new model in

action. So you can, you know, talk to

the model and ask for things. And what

is clear in this video, and what they

emphasize here in this post, is that

GPT-6 is very good at using the

computer, at computer use. So it can

manipulate various tools, various

programs, and do what you want. And in

all these tests they showed here, this

is very clear, right? People go there,

ask the AI to do something, the AI

accesses a site, creates something, or

works in a program. So that is what

they are focusing on, what GPT6 is very

good at. And it's interesting because,

well, we know there is a very fierce

competition between OpenAI and

Anthropic, obviously. And when we look

at the benchmarks here, the comparison

between the models, it is very clear

that GPT Astra either surpasses

OpenAI’s Fable 5, which is now in

version 5.1, also recently released.

And even if it doesn’t quite surpass

or isn’t better than Fable 5, at

least what they emphasized here is that

GPT6 costs much less. So, for some

tasks where using Fable 5.1 costs a

fortune, Astra would cost half or a

third of that. And, well, there are

some things they actually had this

model do that I found impressive, and

we will see them here. And there were

some things I thought were a bit

irrelevant. Like, I don’t know, for

example: they put a video here showing

this new model manipulating Power BI,

which is Microsoft’s dashboard tool,

showing that the model can move the

mouse and create new charts within

Power BI. But I thought this was

pointless because, well, this is a

program that is sort of being

discontinued, right? Nobody really

wants to use Power BI anymore. No data

analyst, no one who, I don’t know,

has to create dashboards, is using AI

to create Power BI reports. People are

trying to create new dashboards using

AI, but outside of this Microsoft

system, right? So I thought this part

was a bit pointless. Something they

showed here that I was like, yeah,

maybe it’s not that interesting, is

that, well, the new model can create

slides, right? But looking at these

slides here, I don’t know, I didn't

think it was a big deal. Of course

it’s interesting because it brings

the information, and the information

here is most likely correct. But in

terms of design, I don’t know, I

didn't think it was much, to be honest.

But many people got early access to

GPT6 and started using the model in

recent weeks; and now that it has been

released, these people have started

posting what they created recently. And

there are some very interesting things.

So, I don't know, for example, this guy

posted on Twitter that he used Astra

over the last few days to build a game,

basically a game with 3D graphics, I

don't know, quite advanced. And it's a

game where you can kind of create a

village, it's a crazy thing. But, of

course, the game has some limitations,

it isn't finished yet, but it's very

impressive what, I don't know, if you

just let Astra work, what it can

achieve with just a few prompts? This

here is something truly impressive. He

let the thing run there, creating this,

and the AI simply created the game. But

it's not a common game, it's actually a

game where there are several AI agents

inside that can talk to each other,

which is something really impressive,

right? And another very impressive

thing, which is something OpenAI itself

published in the official GPT6 post

regarding interior design, especially

using Unreal Engine tools and also

Blender. So, for example, you can

upload a Blender model, a 3D model, and

Astra itself will be able to use Unreal

Engine and create an entire view, an

environment using the Blender model,

and create a video of someone

navigating inside an apartment, a house

, from the 3D model that was created.

Man, this is very impressive. And then

create an entire PDF with photos of the

rooms and the furniture. This is

something I really found impressive.

And several people posted on Twitter

that they had early access to GPT6 and

tested this functionality of

reconstructing these 3D things coming

from Blender. And look at this, it's

really impressive, right? How GPT6 is

managing to manipulate these 3D images

and create these visuals and animations

. Anyway, this can really change the

lives of interior designers a lot. I

don't know, there are a lot of people

commenting on this, right? Commenting

on this OpenAI video and showing how

this was done from the floor plan that

GPT Astra managed to create later in

Blender, the 3D visualization, and even

the video of the environment that was

created, which is even interactive; you

can manipulate the objects there. Man,

this is really cool. But, for me, the

most interesting thing about the GPT6

launch was an analysis they published

in the official OpenAI post, which is

something that might have gone

unnoticed by some people, but it is

extremely important. Well, if we look

here at OpenAI's post and search for

AGI, we come across this benchmark here

, this analysis that shows how GPT6

Astra. performed in relation to the ARC

AGI 3, 2, and 1 benchmarks. And the ARC

AGI is this benchmark here, which I've

also mentioned in some videos, where

there are some tests that AIs are

subjected to, which are logic tests

that usually only human beings can

solve, because it's a kind of game that

you have to manage to win, but you only

discover the rules of the game while

playing it. So that requires logic

there for you to be able to understand

the thing. It's not a game where you

could know beforehand what you have to

do. And look, the most absurd thing is

that GPT6 Astra reached 99.9%in ARC AGI

3, which is the most advanced ARC test.

This is impressive because the previous

model, GPT 5.6 Sol, reached 7.8%and

Claude Opus 5 had reached 30%in this

thing. Of course, anything that reaches

99.9%accuracy, we are a little

skeptical, right? Did they really

perform the test correctly? Or is there

some overfitting of the data happening

there? I don't know, maybe GPT6 was

trained on the data from this test and

that's why it got everything right. But

look, I think OpenAI wouldn't have

released this if something like that

had happened, right? So everything

indicates that this new model really

has something special. It is one of the

most intelligent models and the closest

to an AGI that has ever been created so

far. And this test here makes that very

clear. And so, this is interesting

because some news sites publishing

about the launch of this new model

touched exactly on this point, that

OpenAI's new model has entered the AGI

era. The Verge published about this and

Wired also published exactly the same

thing, that GPT6 Astra is here and that

OpenAI thinks we may have entered the

era of AGI. So, what would an AGI, an

Artificial General Intelligence, be, an

artificial intelligence that would be

as good at most tasks as normal human

beings? an AI that would be better than

us at any type of task, basically in

any field of knowledge. Of course, an

AI, you know, like GPT-6, can't tie a

shoelace because it’s just a

theoretical intellectual AI, but in all

other fields of knowledge and even

logic, this model is already highly

advanced and can be superior to or as

good as the best humans. And so,

recently, Greg Brockman, who is one of

the founders of OpenAI, gave an

interview where he commented a bit on

this subject. He said that, if we

manage to progress over the coming

years and look back, we might ask

ourselves, you know, when was AGI

actually created? And he believes, Greg

believes, that it would be around this

time, and that it might be about this

model that's being released. And later

in the interview, well, he said that

personally he believes we are there. He

says he doesn't think it's irrational

to think that we are already in the era

of AGI, that we have managed to build

an artificial intelligence that is as

good as the best humans in the vast

majority of areas of knowledge. And

it's important to make the distinction,

you know, between the terms. So we

would have AI, which is, you know,

artificial intelligence, and AGI, which

is artificial general intelligence. So,

it would be an intelligence very

similar, very compatible with human

intelligence in most fields of

knowledge. And there would also be ASI,

artificial super intelligence, which

would really be an expanded

intelligence or one much more advanced

than ours. So, GPT-6 Astra would be an

AGI, according to Greg Brockman here.

But of course, you know, we know that

this could be a big marketing stunt,

because we know there's competition

there with both Anthropic and Chinese

companies, and that the models are very

similar and are competing head-to-head.

So for him to say that their model, GPT

-6, is an AGI, there is marketing

behind it, we know that. But at the

same time, I was very shocked by the

ARC-AGI results, showing that this

model managed to reach 99.9%, which

means they will have to create a new

benchmark to be able to evaluate this

model, to be able to test the

intelligence of this AI regarding AGI.

And you can see that the model is

really very good at building many

things, whether in the model's ability

to work well on a computer and

manipulate various programs and

interpret that, using 3D models to

create incredible scenarios, and not to

mention that the model is extremely

advanced in programming. So, anyway, I

think this was the ace up OpenAI's

sleeve this year to manage to beat

Anthropic. And now we'll have to wait

to see what Anthropic will do, when

they will launch Claude 6 or Opus 6 or

a new Mythos model that will leave GPT-

6 behind. But for now, well, OpenAI is

winning. But what is your opinion on

this subject? Do you really think GPT-6

Astra is AGI? Does it have the

necessary characteristics to say that,

well, it is a model that is actually

smarter or as smart as a human being or

just a very good model, but one that

still lacks those characteristics? And

that, deep down, what they are doing

here with the launch of this thing is

just a marketing stunt, until the

Chinese models come along and put an

end to all of this. Write down here in

the comments. Let's discuss this

subject a little more.