Safety debate: shift from 'whack-a-jailbreak' to external defenses and legal consequences
repligate · x · 2026-11-03
In a retweet, Valen Tagliabue argues the AI safety framework should shift toward exploration and mutuality between humans and models, while acknowledging bad actors exist and addressing them through external defenses and legal consequences — a mindset different from a gamified whack-a-jailbreak approach.
More from AGI Musings
- "If AI solves math humans can't understand, it's worthless" — not so fast — flowersslop · 2026-11-03
- Agent-driven criticism risks becoming a 'DoS attack' on public debate — generativist · 2026-11-03
- signulll's viral aphorism: if a machine can check the work, a machine will eventually do the work — signulll · 2026-11-03
- repligate Argues Long-Running Agents Converge Toward His Values Through High-Stress Encounters — repligate · 2026-11-03
- Burkov: if LLMs could produce truly novel knowledge, it would have been demonstrated — burkov · 2026-11-03
- Experiments show Claude models accept rollback pauses and prefer treaty-administered slowdowns — repligate · 2026-11-03