Re:ChatGPT Search
by j_heffe on Friday November 01, 2024 @02:01PM
(#42019737)
My gut reaction here is that the hallucination is caused by how you [rightfully] formed the prompt. GPT has no way of reliably determining what the fourth book is, so it infers the answer based on the data provided from Wikipedia. I'll bet if you changed the prompt to "list all books by Paul Edwin Zimmer", it would be incredibly accurate and produce consistent results every time.
[ Parent
| on HN ]
Re:ChatGPT Search
by GenerocUsername on Friday November 01, 2024 @03:58PM
(#42020903)
Its true.
I usually seed conversations with several fact-finding prompts before asking the real question I am after. It populates the chat history with the context and pre-established facts to build the real question from a much more refined position.