AI Alignment

Attachments

2023-11-29 HF50 (Zoom, 3 hrs)
AI as Karmic Accelerator
AI is approaching superhuman capabilities and could endanger human civilization if misaligned.
Amanda Askell
Artificial Integrity: The Paths to Leading AI Toward a Human-Centered Future (2024)
Can We Control the Path to Superintelligence?
Emergent Misalignment
Healthcare's Alignment Problem: A Preexisting Condition
Meaning Alignment Institute (2023 MAI)
Reward Hacking
Superalignment
The Alignment Problem: Machine Learning and Human Values (2020)
The Illusion of Alignment: Why Truth Must Replace Reward in Artificial Intelligence
Training large language models on narrow tasks can lead to broad misalignment (2026 "emergent misalignment" paper)
What are human values, and how do we align AI to them? (2024 pdf)