NYU Prof Julian Togelius on AI and Games: EA FC's Goalkeeper Is RL-Trained
ziv_ravid · x · 2026-11-02
A new episode of The Information Bottleneck features NYU professor and modl.ai co-founder Julian Togelius, who has worked on AI for games since 2004. Key takeaways:
- Game studios use far more ML than they admit; his tip for finding what actually ships is chatting at GDC afterparties — EA FC's goalkeeper is reportedly an RL-trained network.
- Games are deliberately designed not to need much AI: dialogue trees are awkward but predictable; dropping an LLM into Skyrim and asking an NPC the price of a hamburger gets weird.
- On AGI: a testable standard is learning any unseen game as fast as a human — the hard part is ensuring the model truly never saw the game.
More from AGI Musings
- The most AI-pilled people use AI for everything except writing — cnakazawa · 2026-11-02
- "Federations of Meaning": AI collapses coordination costs and dissolves the org chart — sanjaykalra · 2026-11-02
- Thurston's 1994 warning revived: theorem-credit culture may hurt math as AI automates proofs — alexbilz · 2026-11-02
- François Chollet: Evolution doesn't benchmaxx — it optimizes for generality and robustness — fchollet · 2026-11-02
- NY Fed survey of 5,248 small businesses: AI users are more likely to hire — killboy123 · 2026-11-02
- TechCrunch podcast asks whether AI CEOs are too out-of-touch to build for normal people — nordicinst · 2026-11-02