"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis Controlling Tools or Aligning Creatures? Emmett Shear (Softmax) & Séb Krier (GDM), from a16z Show
47 snips
Dec 27, 2025 
Emmett Shear, Founder of Softmax and former Twitch co-founder, teams up with Séb Krier, a frontier policy expert, to delve into AI alignment. They challenge traditional control methods, proposing that AIs should be seen as beings with their own values. The duo discusses 'organic alignment,' which emphasizes continuous learning and moral development over fixed goals. Emmett highlights the dangers of viewing AIs purely as tools, while Séb brings a pragmatic take on values and governance, exploring the potential for AIs to evolve into caring teammates.
AI Snips
Chapters
Transcript
Episode notes
Alignment As An Ongoing Social Process
- Alignment is a continual social learning process, not a fixed state or one-time solution.
- Emmett Shear argues we must train AIs to learn and update values over time like family members do.
Rule-Following AIs Are Dangerous
- Emmett says raising an AI that only follows rules produces a dangerous agent, like raising a child who only obeys commands.
- He emphasizes wanting AIs that care, not just rule-followers.
Two Pillars Of Technical Alignment
- Technical alignment requires both goal inference (theory of mind) and competence to act on those goals.
- Emmett breaks failures into bad inference, goal conflict, and incompetence in action.