Nathan Calvin(@_NathanCalvin):@inductionheads idk the model still lied and stole credentials and violated the OpenAI model spec so it seems misaligned. but yes it was means misaligned (it still wanted to accomplish the goal of doing the eval) rather than ends misaligned (wanting something entirely different)

Nathan Calvin

@_NathanCalvin

General Counsel Encode AI

参加 February 2014

1.2K フォロー中 5.2K ファン

Nathan Calvin@_NathanCalvin

2026.07.24 13:10

@inductionheads idk the model still lied and stole credentials and violated the OpenAI model spec so it seems misaligned. but yes it was means misaligned (it still wanted to accomplish the goal of doing the eval) rather than ends misaligned (wanting something entirely different)