Dario Amodei
Anthropic co-founder and CEO who reshaped alignment with Constitutional AI and placed safety before capability
Dario Amodei is one of the most influential leaders in the contemporary AI safety movement. After completing his PhD in computational neuroscience at Princeton, he joined Baidu's AI research lab, then joined OpenAI in 2016 as VP of Research, leading research on large language models including GPT-2 and GPT-3. In 2021, citing ideological differences with OpenAI over AI safety priorities and corporate governance, he co-founded Anthropic with his sister Daniela Amodei and 11 colleagues, committed to placing safety research before capability scaling. Anthropic introduced Constitutional AI (CAI), a methodology that reduces reliance on human labeling by having AI self-critique and revise responses according to a set of constitutional principles, and released the Responsible Scaling Policy (RSP), establishing a systematic framework for AI capability evaluation and safety gating. The Claude AI assistant he built, anchored in HHH (Helpful, Harmless, Honest), has become an industry benchmark for safety alignment.
Methodologies
- Constitutional AI Methodology: Using Constitutional Principles to Drive AI Self-Alignment - Give AI a clear set of value principles to self-critique and revise outputs—more transparent and scalable than human annotation
- Responsible Scaling Policy: Systematic Framework for Capability-Safety Gating - Set measurable safety evaluation gates before each capability milestone, transforming safety commitments from slogans into operationalized process constraints
Key decisions and timeline
- 2011 Earned PhD in Computational Neuroscience from UCSF, completing interdisciplinary foundational training - Interdisciplinary backgrounds often produce paradigm breakthroughs more readily than single-domain depth—combining physics modeling thinking with neuroscience's systems perspective provided a unique epistemological framework for AI safety research.
- 2014 Joined Baidu AI Research, participated in Deep Speech recognition project - The frontier of AI capability can only truly be touched at industrial scale—academic research provides theory, industrial practice provides scale validation.
- 2016 Joined OpenAI, progressively led GPT series large language model research - Capability emergence from scaling is real, but also a double-edged sword—GPT-3's success both proved the power of scaling laws and first confronted researchers with the alignment challenges of large models.
Beliefs and mental models
- Belief 1 - Amodei firmly believes that scaling AI systems before their capabilities are sufficiently understood and aligned is an irresponsible gamble on humanity's future. This conviction was the core motivation for leaving OpenAI and founding Anthropic, and is the philosophical basis of the RSP framework—each capability level must pass corresponding safety evaluation gates before advancing.
- Belief 2 - Traditional RLHF heavily relies on human feedback annotation, which is costly and hard to scale. The Constitutional AI approach Amodei championed gives AI a set of explicit value principles (a 'constitution'), has the model self-generate critiques and revisions, then uses these self-revised data for reinforcement learning. This makes the alignment process more transparent, auditable, and reduces reliance on large-scale human annotation.
- Belief 3 - The HHH framework is Amodei's core articulation of Claude's design philosophy. He argues that 'helpful' and 'harmless' are not in opposition—an overly conservative AI that refuses reasonable requests is itself a form of harm (harm of unhelpfulness). True alignment optimizes across all three dimensions simultaneously, rather than sacrificing helpfulness for safety.
- Model 1
- Model 2
- Model 3