X · @cwolferesearch
· X / Twitter
Many RSI-style papers use the coding agent to hill climb results. This works and is simple, but I wonder if greater improvements can be unlocked with …
Many RSI-style papers use the coding agent to hill climb results. This works and is simple, but I wonder if greater improvements can be unlocked with coding agent + evolutionary / genetic algos. Much to be explored for designing effective self-improving agent harnesses!alphaXiv: "Frontis-MA1: Training an AI4AI Model to