Tag
This paper evaluates LLM performance on statics problems, finding that while text-only questions are handled well, accuracy drops with diagrams and multi-step reasoning, suggesting difficulties in applying visual information consistently.