Re:Rats have an imagination, new research suggests
by Davidzheng on Sunday November 12, 2023 @03:59PM
(#38244028)
I don't think this is correct. It could (at least theoretically) be able to verify chains of pure reasoning bounded by a fixed length.
[ Parent
| on HN ]
Re:Rats have an imagination, new research suggests
by chongli on Sunday November 12, 2023 @04:13PM
(#38244120)
If I ask ChatGPT "is the sky blue?" it replies "Yes, the sky is often seen as blue during the day, particularly on clear days." How does it verify this? It cannot. There is no chain of reasoning by which the LLM can verify this statement. All it can do is defer to others. In other words, the utterances of ChatGPT are entirely hearsay.