Industrial AI Podcast Calling America - World Models Are Back
Mar 11, 2026
Jakub Tomczak, professor and AI researcher specializing in neurosymbolic methods and world models, discusses the resurgence of world models. He contrasts symbolic and physics‑based approaches with pure data‑centric models. He explores Europe’s strengths, the role of domain expertise and simulators, and why smarter, domain-aware methods may outpace blind scaling.
AI Snips
Chapters
Transcript
Episode notes
Neurosymbolic Approaches Can Beat Pure Scaling
- Neurosymbolic AI combines symbolic domain knowledge with data‑driven methods to improve trustworthiness and sample efficiency.
- Jakub sees neurosymbolic approaches as underutilized and timely given limits of pure scaling.
Europe Should Double Down On Domain And Math
- Europe should leverage strong math and domain expertise to focus on neurosymbolic and simulation applications instead of copying big LLM investments.
- Jakub highlights Europe's uniform education and existing researchers as strategic advantages.
Domain Knowledge Trumps Model Size For Physical Problems
- Strong domain knowledge (biology, chemistry, physics) plus math matter more than chasing parameter counts.
- Jakub stresses domain understanding for simulators and physics‑based models over billion‑parameter LLM arms races.