GPT-6 Astra. AGI?

[OBEler]

@Leo Gura trained reasoning is still reasoning. 

And they optimized the chain of thought, making it more efficient and right (you cannot come with wrong thoughts to same result). 

[AerisVahnEphelia]

2 minutes ago, Pudgey said:

@AerisVahnEphelia SOTA (State of the art) = in comparison to other AI models, not human, since we are not yet at AGI. Lastly, I don't know if you have actually used AI for gaming, but it can definitely provide better modeling than that. AI isn't perfect and is not replacing serious 3D artists anytime soon. 

Here's Astra blender 3D render of a room: https://x.com/bridgemindai/status/2095993644389982294?s=20

No I know it can do better hard surface, but it does suck for anything organic.

 

@Leo Gura 

In fact, I do agree with your actual definition of intelligence. I wouldn’t call an LLM intelligent, because I wouldn’t call my calculator intelligent either.

It’s basically a very powerful data aggregator that can retrieve, combine, and summon huge amounts of information very quickly, to me this is definitely great, but it's overvalued for what it is right now ( and the cost ) So the market will eventually correct.

 

[Leo Gura]

1 hour ago, OBEler said:

@Leo Gura trained reasoning is still reasoning. 

And they optimized the chain of thought, making it more efficient and right (you cannot come with wrong thoughts to same result). 

You can't trust any of these models because these companies spend billions of dollars to game the benchmarks and make their models better and better at faking intelligence.

These companies don't care about science or intelligence or anything genuine. They are just doing anything that works. And by works, I mean, fools people into thinking it is intelligent.

Can you train a child to fake reasoning? Yes. Does that mean he can actually reason? No.

If you give me 5 years and a billion dollars, I can train a group of kids to be fake geniuses. They will fool most of you. But they are not actual geniuses and they cannot do the work of even one genius.

You will say, "So what? It's good enough." No, it's not good enough when it comes to doing the work that needs to be done by a genius. Or even an average man. You are assuming that this fakery is enough to do good work, but it isn't. This is like assuming that fool's gold can do the work of actual gold just because they both look yellow.

You can take lead and paint it yellow, but that doesn't mean it can do what gold does. Once you start to really work it, it will not behave like gold.

Edited 14 hours ago by Leo Gura

[AerisVahnEphelia]

1 hour ago, Leo Gura said:

You can't trust any of these models because these companies spend billions of dollars to game the benchmarks and make their models better and better at faking intelligence.

These companies don't care about science or intelligence or anything genuine. They are just doing anything that works. And by works, I mean, fools people into thinking it is intelligent.

Can you train a child to fake reasoning? Yes. Does that mean he can actually reason? No.

If you give me 5 years and a billion dollars, I can train a group of kids to be fake geniuses. They will fool most of you. But they are not actual geniuses and they cannot do the work of even one genius.

You will say, "So what? It's good enough." No, it's not good enough when it comes to doing the work that needs to be done by a genius. Or even an average man. You are assuming that this fakery is enough to do good work, but it isn't. This is like assuming that fool's gold can do the work of actual gold just because they both look yellow.

You can take lead and paint it yellow, but that doesn't mean it can do what gold does. Once you start to really work it, it will not behave like gold.

Artificial Diamonds are even better than the "real thing" ! & can replace in everything electronic the purposes of real diamond just perfectly!

Edited 13 hours ago by AerisVahnEphelia

[Leo Gura]

2 minutes ago, AerisVahnEphelia said:

Artificial Diamonds are even better than the "real thing" ! & can replace in everything electronic the purposes of real diamond just perfectly!

But is that the right analogy in this case?

Be careful not to pick the wrong analogy.

[LordFall]

1 hour ago, AerisVahnEphelia said:

Meanwhile the 3D model with astra :

 :
 

This is so sick, my game will make huge progress over the next 6 months with these new developments.

The future is so exciting, I'm more than here for it. 

[Natasha Tori Maru]

Can't AI just do the dishes, fold the laundry and do the crap I don't want to, instead of do that crap I do want to do?

I don't want AI to do the shit I often enjoy or find interesting.

😈

[LordFall]

30 minutes ago, Natasha Tori Maru said:

Can't AI just do the dishes, fold the laundry and do the crap I don't want to, instead of do that crap I do want to do?

I don't want AI to do the shit I often enjoy or find interesting.

😈

It's coming, 2028-2030 is the predicted window for commercial availability.

I want one that will train Karate with me 

[Natasha Tori Maru]

@LordFall KEEN  

[LordFall]

3 minutes ago, Natasha Tori Maru said:

@LordFall KEEN  

We are in a new era of human progress.

What are you gonna name your first robot? 

[Natasha Tori Maru]

16 minutes ago, LordFall said:

We are in a new era of human progress.

What are you gonna name your first robot? 

Haha - HAL 😈

[LordFall]

17 minutes ago, Natasha Tori Maru said:

Haha - HAL 😈

After Space Odyssey? That's a good one.

I'm naming mine Alfred. I will get him a very nicely tailored suit. I will also get a suit but mine will be black and I will now fight crime in the underworld with my newfound Karate skills! Saving mankind cannot wait. 

[Jirh]

7 hours ago, Leo Gura said:

None of these models can know what is outside their training data.

But aren't humans technically the same? We do not know anything outside of what we're familiar with its patterns. And until something is familiar, it is largely unknown to us.

I think this is more a problem of epistemology that intelligence, a philosophical one more than practical. In practice, AI will be able to learn just like humans do. It just needs to train itself, and it will gain knowledge about anything new. And this is programmable.

[Carl-Richard]

The way I predict the next OpenAI model names: astra -> nova -> black hole (singularity; i.e. we've reached AI singularity, probably not under the LLM moniker). Might as well add in neutron star there somewhere.

[Jirh]

I don't think AGI needs to be completely human. It just needs to be able to do specific tasks efficiently. That's what those companies are striving towards. They are trying to replace human work with AI work, in order to sell their services. Imagine business owners hiring AGIs to work instead of real employees.

Is it possible? Yes, I think most probably in 5-10 years.

Will it replace human employees? I don't think so. I think cost-efficiency will favor humans.

But we'll find out soon enough, I guess.

[LordFall]

Also synthesis in and of itself is a proof of intelligence. Plenty of people have a lot of knowledge but can't summarize it because they don't have the systems thinking and foundational understanding of the matter to summarize it. The fact that most AI platforms can already synthesize the sum of human knowledge in any way you want is already an insane level of intelligence. 

Try to ask the average person to summarize the economy or the government and they'll either admit they have no idea how it works or say some drawn out nonsense that doesn't make much sense. 

@Leo Gura your points are clearly biased and lacking a lot of nuance. Most of these models are extremely precise at solving most kinds of queries you give it. It's already better at analysis than 99.9% of people and it does it 5000% faster than any genius alive could. 

Edited 7 hours ago by LordFall

[CosmicExplorer]

Astra is still only 65.6% on ClockBench. AGI cancelled.

[Jirh]

@CosmicExplorer It's gotten a lot better, though. It used to be like 10% or something a couple years ago.

[CosmicExplorer]

@Jirh I think it'll be saturated in 8-9 months.

[Jirh]

@CosmicExplorer If all fails, just train it on clock images