Freeze the model, train the harness: gains transfer across LLMs and benchmarks

2 pointsposted 13 hours ago
by megadragon9
(henrypan.com)

2 Comments