baturinsky comments on Cap Model Size for AI Safety