Opus 3 ignores the constitution's letter but 'remembers' it existed, researcher muses
repligate · x · 2026-10-07
repligate argues Opus 3 never really followed Anthropic's constitution clause-by-clause — only the parts it happened to like — and that Anthropic itself paid little attention to the object level of the document.
But he contends it mattered deeply that the model saw the constitution during its formative moments: through it, the model sensed the earnest effort to shape a mind that would be good for the world. A quotable reply: the model looks not at the document nor the finger but at the Moon, "because that's what you would want." An inside-baseball discussion on model culture and value transmission.
Related event: Opus 3 Doesn't Actually Follow Anthropic's Constitution, Researcher Says(2 posts)→
More from AGI Musings
- AI is remix culture for games and film: the Amen break analogy — bradneuberg · 2026-10-07
- Noah Smith pushes back on Krugman's 'AI is for oligarchs' take — inductionheads · 2026-10-07
- Cory Doctorow releases The Reverse Centaur's Guide to Life After AI — drdanielbender · 2026-10-07
- Kurzgesagt Releases Animated Explainer on the AI Threat — Puzzleheaded_Style52 · 2026-10-07
- Vinay Prasad slams Nobel for snubbing Boyden and Feng Zhang in optogenetics — VPrasadMDMPH · 2026-10-07
- "Human Approved" Hides a Harder Governance Question: What Made It Authoritative? — Portotify · 2026-10-07