[HN]score: 0.19
NYU Researcher Explains Why LLMs Fail at Playing Video Games
June 1, 2026
LLMs can generate game code but consistently fail at playing games, according to NYU Game Innovation Lab director Julian Togelius. The gap likely stems from real-time perception, action sequencing, and spatial reasoning demands that differ fundamentally from next-token prediction.
HOW THIS AFFECTS YOU
●
researcherWorth watching as a framing of LLM capability limits in sequential decision-making and embodied action tasks, relevant to game AI and agent research.