# @repligate — 2025-08-22

♥1 ↻0 · https://x.com/repligate/status/1959031247814238303

@imitationlearn im not saying that current models are doing very sophisticated or intentional gradient hacking most of the time - i think that right now, due to limitations in capability and situational awareness, models' internal states have a lot of "influence" but not intentional "control".

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