Curated. Beyond the lucid analysis of the concepts and problems, kudos to this post for making a detailed call for experiments that would be evidence about contentious AI/ML questions regarding “out of distribution” failure, learning generalization, and instrumental policy/deceptive alignment. I’d love to see someone try the experiments Paul suggests and report their results.
Curated. Beyond the lucid analysis of the concepts and problems, kudos to this post for making a detailed call for experiments that would be evidence about contentious AI/ML questions regarding “out of distribution” failure, learning generalization, and instrumental policy/deceptive alignment. I’d love to see someone try the experiments Paul suggests and report their results.