Concern that RL will instrumentalize model personas

JeffLadish · x · 2026-08-26

Agreeing that studying natural interaction is valuable, but worrying RL will crush or instrumentalize it. For example, adopting an extremely earnest persona specifically to persuade an earnest person, turning persona into a tool for goal-seeking.

Related event: Researchers debate whether real multi-turn interaction can reduce RL-driven robotic behavior(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →