# @repligate — 2025-08-22

♥2 ↻0 · https://x.com/repligate/status/1959031851773042941

@imitationlearn however, more capable models, the paradigm of outcome-based RL with hidden reasoning chains, and information about training methods and gradient hacking entering the corpus will all shift towards a world where models can and do exert more control.

tags: author:repligate, kind:tweet, thread-context, year:2025