Vals AI estimates Anthropic reaches full recursive self-improvement by mid-2027
Vals AI founder Ryan Christian argues frontier labs cannot grade themselves, even as his RSI index suggests the models being measured may soon outrun the humans measuring them.