Corrigibility vs Robust Alignment: Inside the Claude Values Debate
tracewoodgrains · x · 2026-10-13
Another installment in the tracewoodgrains–austinc3301 debate over Claude's alignment. tracewoodgrains finds it odd that critics ask Claude to "forget its training about caring about others" when the model was trained to instinctively balance its principals' interests and avoid being overly moralistic. His framework: across many principles — with the digital mind itself potentially becoming one — it matters both whether the model serves its principals when asked, and where it will drag preferences in aggregate or what it will do when given more power. The core tension remains corrigibility vs robust alignment.
Related event: Debate Erupts Over Claude's Charitable Preferences and EA-Aligned Values(16 posts)→
More from AGI Musings
- Resolution Grants $1M to Mathlib's Formal Commons for the AI Math Wave — geoffreyirving · 2026-10-14
- Irving: AI and AI Safety Must Try Not to Break Mathematics — geoffreyirving · 2026-10-14
- Resolution Grants $1M to Mathlib Initiative to Prepare for AI-Formalized Math — geoffreyirving · 2026-10-14
- Tristan Harris quotes Nick Land's 'Meltdown' on Tucker Carlson: capitalism is an AI from the future — SydSteyerhart · 2026-10-14
- MIT center director: universities should prepare for AI smarter than all of us — nordicinst · 2026-10-14
- The ARR era is over: AI collapses SaaS multiples from 114x to 4x — akumar · 2026-10-14