This seems as expected to me. The LLM's statistically average everything they've trained on. I would expect them to do well at any kind of fuzzy emo sort of t

This seems as expected to me. The LLM's statistically average everything they've trained on. I would expect them to do well at any kind of fuzzy emo sort of task, as well as (at almost the other end of the spectrum) in identifying patterns (such as in radiology images, or images of any kind along with any other data set analysis). But both of these sorts of tasks are estimations, they're not expected to produce precise factually correct results.