Large language models (LLMs) have improved so quickly that the benchmarks themselves have evolved, adding more complex problems in an effort to challenge the latest models. Yet LLMs haven’t improved across all domains, and one task remains far outside their grasp: They have no idea how to play video games.
While a few have managed to beat a few games (for example, Gemini 2.5 Pro beat Pokemon Blue ...
The narrative presents a compelling case for the limitations of LLMs in video games, contrasting their coding prowess with their gaming ineptitude. The strongest version of this argument—steelman—acknowledges that LLMs excel in structured, rule-based domains like coding but falter in the open-ended, diverse world of video games. This isn’t just an LLM problem; even specialized game AI struggles with generalization. The pattern scan reveals no overt manipulation, but there’s an implicit tension: ...