Comment by whimsicalism

Misuse of extremely capable models, misalignment during RL are both very large risks as capabilities grow imo