Anthropic CEO proposes ASI as a model committee with a separately trained ethicist model
robleclerc · x · 2026-09-16
Rob LeClerc proposes that ASI need not be a single model or MoE — it could be a committee of models including a separately trained conscience/ethicist model that sets the range of permissible directions and checks outputs for consistency. He invokes Alasdair MacIntyre's "first principles are argued from, not to" to explain how such an ethics model would ground the reasoning models.
More from AGI Musings
- After welcoming Dario at Dreamforce, Benioff's feed draws jab: if you truly believe in 10% extinction risk, why sell AI into B2B SaaS — SumitGup · 2026-09-16
- Census study: most AI-exposed college majors see employment odds fall 5 points, starting pay down 13% — asusarla · 2026-09-16
- Andrew McAfee on 'Geek Doctrine': how Silicon Valley ran circles around incumbents — amcafee · 2026-09-16
- Scott Alexander on AI skeptics' endless shell game of moving goalposts — teortaxesTex · 2026-09-16
- callcongress.ai: ex-OpenAI/Anthropic researchers urge public to lobby on AI risk — eli_lifland · 2026-09-16
- New study: companies adopting AI hire MORE entry-level workers, not fewer — chris_j_paxton · 2026-09-16