Dario Amodei
CEO & Co-founder · Anthropic
Argues that scaling laws are an empirical fact, not a slogan — and that the safety case has to be built into the training loop, not bolted on after capabilities arrive.
Context Jamming · Executive Insight
Frontier-AI leaders, timestamped
One conversation, mapped. Jump to the exact moment each executive makes their case — the player, the profiles, and the timeline stay in sync, so you can read the room while you watch it.
Bret Kerr · ACRA Insight · June 2026
Cold open — the infinite now vs. continual learning
Timeline0:00
The executives, on the record
Demis Hassabis
CEO & Co-founder · Google DeepMind
Frames AGI as a scientific instrument — the through-line from AlphaFold to general agents is using learned models to compress search over very large problem spaces.
Ilya Sutskever
Co-founder & Chief Scientist · Safe Superintelligence
Holds that next-token prediction, pushed far enough, forces a world model — and that the hard part now is not capability but reliable generalization out of distribution.
Jared Kaplan
Chief Scientist & Co-founder · Anthropic
The physicist who wrote the scaling laws down: predictable loss curves let you budget compute like an engineering quantity rather than guessing.
Chris Olah
Interpretability Lead & Co-founder · Anthropic
Treats neural networks as objects to be reverse-engineered — features and circuits as the unit of explanation for what a model has actually learned.
Timestamp index
- 0:00Cold open — the infinite now vs. continual learning · Dario Amodei
- 0:22Amodei: scaling, and what the context window can’t remember · Dario Amodei
- 0:52Hassabis: memory, retrieval, and general agents · Demis Hassabis
- 1:24Sutskever: prediction, compression, in-context learning · Ilya Sutskever
- 1:56Kaplan: scaling laws meet a finite context · Jared Kaplan
- 2:26Olah: a context window from the inside · Chris Olah
Source video: youtu.be/M89fL14fK8c