Christopher King comments on GPT-4 aligning with acasual decision theory when instructed to play games, but includes a CDT explanation that’s incorrect if they differ

[Christopher King]

Maybe. It just seemed weird that when they coincided, it gave perfect reasoning.