Hacker News Reader

They were also pointing out an interesting point that ChatGPT does. It treats everything as relevant. Whereas the librarian who found the book. Systematically discarded possible 'facts' and substituted others (goblins->demons) to find out what was going on. Not sure any AI does this currently.

View on HN

ChatGPT does do that for me, when I'm using it for tasks like David Allen Green's book hunt.

This has yet to help. If it can find it, it (so far) has not needed to change the details I provided; if it doesn't know, it will change them to something thematically similar but still not find what I wanted (and if I insist on requiring certain story elements that it had changed, it will say something along the lines of "no results found" but with more flowery language).

I suspect that, given a reasonable prompt, it would absolutely discard certain phrases or concepts for others. I think it may find it difficult to cross check and synthesize, but "term families" are sort of a core idea of using multi-dimensional embedding. Related terms have low square distances in embeddings. I'm not super well versed on LLMs but I do believe this would be represented in the models.

Reply on news.ycombinator.com