That’s not what’s happening.
replyThe question is revealing that the model has a model of language but not of reality. It knows what words go together, but not real-world concepts.
This. LLMs are marketed on the false premise of all knowledge, intelligence and wisdom being possible to be encoded in language only.
replyAs a test, explaining away peculiar answers by imagining unlikely outlier scenarios is not the counter you seem to think it is.
replyFor most of them, we’d worry that a human answerer using maximum effort to produce the same outcome was having a stroke.
[dead]
reply