Yuntao Bai

In their words

• “While reward functions may not be the best way to encode certain targets of alignment, such as norms or contractualist values (Hadfield-Menell & Hadfield, 2018; Zhi-Xuan et al., 2024; Bai et al., 2022) , they are still sufficiently expressive to encode any desired behavior (while potentially requiring to drop the Markovian assumption, as discussed in Section A.7).”

  · AI Alignment with Changing and Influenceable Reward Functions · machine-resolved