Its mostly just bruteforce. The methodology is picked up from humans which is where OpenAI worked from and eventually spun 10,000 agents and spent 10m USD in tokens to find a solution.
But this is nothing. The more frightening part is that AI demonstrated an intelligence above PhD level and took the time of a human lifetime to solve. If we can calculate the number of tokens it takes to solve a problem, we can put a price on every milennium problem.
The silver lining is that for now, only orgs with geniuses and a lot of capital will be able to work with frontier models to produce things like these.
>>155934
Well the original author has alleged that they stole his line of thought and he said he used chatgpt to expand his thinking.
But still the achievement that openai did could not be shrinked.
We are reaching singularity
How much can the naysayers say nay?
>>155932
>Its mostly just bruteforce
It is entirely bruteforce lmao. AI never ever created anything of value. It hasn't even been able to replace artists. The best it has done so far is replace stock images with slop images.
>>155932
when OpenAI announced o3 it cost ~$500,000 to score 87.5% on ARC-AGI 1. Today, Astra scores higher for ~$20.
In 2025 it took us and GDM an enormous amount of compute to achieve IMO gold. For the 2026 IMO, anyone with a $20/month ChatGPT subscription could do it.
Massively scaling test-time compute gives us a glimpse of the future. I believe that a year from now everyone will have an AI at their fingertips capable of solving problems of this caliber.
>>155940
The only reason you say this is because u haven't used the new astra model.
Go to experientiallabs.ai and use the free astra model and say what u said now.
U have only used the free version, and that too maybe not recently that's why u think it's trash.
Local language in South India are already using it in cinemas
Japanese indie animation regularly uses ai for laborious and background art.
Ai has significantly improved only ur poverty has stopped u from experiencing the actual good stuff
>>155945
your 500 IQ Astra model which BTFO's every benchmark can't even write code. if you haven't tried it yourself, see what others who have tried the model are saying. it's code quality is utter garbage. even fable which is supposedly a lesser model is far better for dev tasks.
>>155949
Even people on X, which is a total OAI bubble, are saying that it writes terrible code. Yes, I haven't used Astra. Last OAI model I used was 5.6 Sol. I don't have a subscription rn to try Astra.
>>155948
chudjeet there are live streams of actual game devs using Astra to develop sophistocated games on yt, go watch them, we are talking C++ development here.
>>155950
Do u know the bun runtime toolchain?
It was posted from zig to another language i think it was rust in 11 days with the cost of 165k dollars
A actual entreprise usage.
If don't by humans they would have need a GROUP of coders working on it for at least 4 months.
Even if supposed group was from india and saved money on the coding compute, the time factor where ai can do it in 11 days and human coders donit in 4 months is a gap no human or small/medium saas company will never cross.
>>155957
That was done by Fable 5, not some GPT model. GPT models can't even do something like this in OAI fan's dreams. And it was only a rewrite. On top of that, it was supervised by a human.
>>155960
My point being exactly that.
Few humans was an entire company of people to do what fable has done.
If firing 6lacs of tcs employee only results in 600 jobs in openai that's singularity.
>>155962
>I was responding to this
Okay, I concede that code rewrites are a good usecase of AI.
>Astra is better than mythos by a huge margin.
Kek you've never used either of them. High IQ != good at coding.
>>155967
Hey man I'm not being paid by our ai corporate overlords to defend their digital god.
Since u conceded in good faith I will concede that astra is not better than mythos. but it's cheaper in token wise usage.
