mayfer: turn-based RL can't close AGI's gap — realtime perception is required

mayfer · x · 2026-10-13

mayfer argues turn-based RL can never solve the tail end of broken software; perceiving action results in realtime is a requirement. He sees "digital optimus" and thinky interaction models as more serious long-term AGI projects than chatbot coding RL environments — less superintelligent on jagged domains but needed for generality. He also laments "system 1" being co-opted by decision models when it means realtime perception and reaction.

Original post →

More from AGI Musings

AGI Musings channel →