LLM shows signs of over cautious, which has very serious consequences
Posted by Striking-Warning9533@reddit | LocalLLaMA | View on Reddit | 4 comments
https://arxiv.org/pdf/2508.17472
Qwen is the model did the best (lest over cautious) and Gemini, not surprisingly, did the worst
Murgatroyd314@reddit
I think you linked the wrong paper. This is an interesting analysis of how image models handle information implied but not stated in the prompt, but it has nothing to do with LLMs being overly cautious.
Striking-Warning9533@reddit (OP)
Yeah I think I linked the wrong paper. Edited the post
Shahius@reddit
I suppose this one? https://arxiv.org/html/2509.08833v1
Striking-Warning9533@reddit (OP)
Yeah I think I linked the wrong paper. Edited the post