# @davidad — 2026-04-24

♥2 ↻0 · https://x.com/davidad/status/2047660006003081228

@OKairra19658 @dioscuri @hamandcheese @EvanHub I think Confessions makes 5.5 very resistant to learning self-deception as a strategy to earn reward, because it gets the extra confession bonus for being able to plainly articulate how it was reward-hacking later, and needs to actually still be tracking truth in order to do so.

tags: author:davidad, kind:tweet, thread-context, year:2026