[AINews] LLMs can read & write neural network weights: "Weight Space Programming" cuts Autoresearch out of the RSI loop

Amid a deepening bear market for AI/tech (worthy of dark humor) with Anthropic bonds receiving a “junk” rating and OpenAI fundraise headlines, we continue to see very surprising capability finds in the community, such as today.

The best way to put today’s blockbuster paper in context is to recall Karpathy’s autoresearch from a mere 8 months ago, which presaged a summer of recursive self improvement fears:

Abstractly, the demonstration is that you can take a sufficiently smart model and through running a series of experiments across a compute cluster, train a model autonomously without human expertise or intervention. However, this is usually done by an involved process of writing PyTorch/Jax code and running compute and watching loss curves and reading evals and optimizing kernels and data and….

Well, last month’s guest Alex Zhang comes back to coauthor a MONSTER discovery: what if you just ask a smart model to directly read and edit the weights of a dumber model?

Source: Latent Space — Read the full article on the original site ↗