everyone’s become quite focused on scale from a safety perspective… calls for audits, licenses and so on. but really we should also focus on generality. single agents that can do a very wide range of tasks will be inherently more challenging to control
AI Safety: Generality as Important as Scale for Control
By
–