Inception Labs CEO Stefano Ermon: training an AI model is fundamentally compression
No Priors · youtube · 2026-10-05
The No Priors podcast released an explainer video, "How AI Models Learn," featuring Inception Labs CEO and Stanford professor Stefano Ermon on what it actually means to "train" an AI model.
Ermon's core point: training is fundamentally compression. The model searches for the most efficient way to compress the data, and it's precisely in optimizing that compression that it captures the underlying structure — the better the compression, the deeper the grasp of the data's regularities. It reframes "learning" as an information-theoretic problem.
More from Research
- Petri Alignment Evals May Be Detectable: Simulated Worlds Are Suspiciously Responsive to the Subject Model — a_karvonen · 2026-10-05
- LLM-guided evolutionary search for algorithms: keeping 'weaker' candidates lifts solution quality 0.81→0.99 — bravo_abad · 2026-10-05
- Ian Osband's 'Planning to Learn': One-Line Horizon Loss Beats Both Policy Gradient and Cross-Entropy — IanOsband · 2026-10-05
- Countervailing Technologies: how distillation-style tools can diffuse AI power concentration — DavideCrapis · 2026-10-05
- Galbot humanoid to play autonomous tennis at China Open; Paxton on what robot sports reveal — chris_j_paxton · 2026-10-05
- Policy gradient gets 4% vs 62% for cross-entropy on ImageNet, argues LLM post-training blame is misplaced — IanOsband · 2026-10-05