Dave Lindbergh 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