likely all it takes to make this work is for major labs to set up an rl gym where you extrude models that solve tasks with energy penalty, and hill climb on it.

i have always assumed this capability was just a matter of training and that no special architecture is required.

Is this basically the bitter lesson applied to the very act of creating model weights? Absolutely wild, and the implications for RSI are noteworthy.

>Qwen3.5-0.8B

I wonder what the biggest dense model in the same model family, the surprisingly capable Qwen3.5-27B, would be able to do if given the same task.

Will weights programming be the norm for creating new models? To what degree do we still need a full training run vs modifying the weights directly?

How is this such a big story on HN. It seems to me that this is just another engineering problem, not particularly surprising

what would be the viability of a model changing its weights and using genetic algorithms to try to achieve something novel, even without understanding exactly which weights need changing?

So, how soon before OpenAI's researchers are using Astra to modify Astra's own weights to make them better?

Completely shook. Is this the start of it? At the very least one (not sure if the first?) example of strictly superhuman capabilities.

That felt like LLMs are just a set of numbers and people pay to multiplication of these numbers.

Could be an avenue for synthetic continual learning? A powerful model could update its own weights manually given new training data.