LLMs Can Now Read and Write Neural Network Weights
Jia-Bin Huang · 26:04 · Watch on YouTube
A general model the video calls Astra can read network weights as text, patch one localized failure, or write a small policy from a plain-language goal, without Adam-style gradient steps. The large score jumps, above...