Thomas Fel on X: "A science of post-training can be built on interpretability. Models know when they reward hack, and we see it, yet training ignores it. We propose Mechanistic Shaping: using the model’s own concepts to write training rules. In vision, DPO & RL, and its effects on monitoring 🧵 https://t.co/ts4IY825Lf" / X
x.com · saved by 1 readers
A science of post-training can be built on interpretability. Models know when they reward hack, and we see it, yet training ignores it. We propose Mechanistic Shaping: using the model’s own concepts to write training rules. In vision, DPO & RL, and its effects on monitoring 🧵