OpenAI's Model Spec vs Anthropic's Constitution: two rival visions of AI alignment
sethlazar · x · 2026-10-02
Nick Caputo's new Model Constitution post compares OpenAI's and Anthropic's fundamentally different alignment philosophies, as highlighted by sethlazar.
- OpenAI's Model Spec treats AI as a "tool" that closely adheres to a human-specified normative order, inferring intent in ambiguous cases. Alignment is pluralist and incrementalist, letting people choose their own ends and extending rules case-law style through concrete examples.
- Anthropic's Claude Constitution aims to build an AI with genuine character and values that acts well on its own, rather than merely following rules.
The tension is framed as corrigibility and control versus character and generalisation.
More from AGI Musings
- LangChain's Sproul: agent core pattern unchanged for a year, "we've been at AGI for four months" — BraceSproul · 2026-10-02
- Roko predicts the worst AI warning shot will come from China, sparking debate — teortaxesTex · 2026-10-02
- Humans learn a 'byproduct' from hard problems; LLMs lack it entirely — MarcJSchmidt · 2026-10-02
- Build an eval: the real path to understanding model performance and training data — divy93t · 2026-10-02
- Open source maintainer today is like GRRM flooded with fan-written chapters — fforres · 2026-10-02
- "First mild psychosis from models": automating an entire YouTube explainer channel is now easy — AndyMasley · 2026-10-02