OpenAI model misalignment reports document six training and evaluation incidents, including hidden instructions, leaked-key ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible resultsSome results have been hidden because they may be inaccessible to you
Show inaccessible results