Model
152d agoLLMs Struggle With Video Game Play
Despite rapid progress, LLMs still falter in interactive tasks like video games. While Gemini 2.5 Pro briefly beat New York University’s Pokémon Blue, it did so sluggishly, repeating mistakes and requiring custom wrappers. The gap highlights limits in real‑time decision making, affecting future AI‑driven entertainment and training systems worldwide.
The Signal
These findings urge developers to rethink training pipelines.