Dwarkesh Podcast Dario Amodei — "We are near the end of the exponential"
7506 snips
Feb 13, 2026 
Dario Amodei, AI researcher and CEO of Anthropic, renowned for work on large-scale models and AI risk. He discusses why scaling and task-specific RL may generalize, how AI could diffuse through the economy, timelines for near-AGI-like capabilities, Anthropic’s compute and profitability choices, and governance and geopolitical risks tied to powerful models.
AI Snips
Chapters
Books
Transcript
Episode notes
Big Blob Of Compute Explains Progress
- Large-scale compute, broad diverse data, long training and scalable objectives are the main drivers of AI progress.
- Dario frames pre-training and RL as the same “big blob of compute” pathway to generalization across tasks.
Pretraining Across Wide Data Enables Generalization
- Pre-training on a broad distribution yields surprising generalization not seen in narrow corpora.
- In-context learning gives models rapid short-term adaptation that complements long pre-training.
Focus RL On Diverse Tasks, Not Specific Skills
- Design RL environments to provide broad task diversity rather than teaching every specific skill.
- Train on many tasks so models generalize to novel, unobserved situations.
