logoalt Hacker News

tarxzvftoday at 2:26 PM6 repliesview on HN

Models hallucinate plausible answers to why they did things. It might be true and it might be complete fiction.


Replies

taerictoday at 2:36 PM

I'm growing increasingly confident that this is how people often work, as well.

show 8 replies
JeremyNTtoday at 6:09 PM

Yes. But although they can't know "why" a specific "wrong" answer was selected, the response is often still informative, and it can highlight real weaknesses in process or code structure that should be addressed anyway.

jamesharttoday at 5:55 PM

To test this, change the history in the context to indicate that the model did or recommended something completely different than it actually did, and then ask it to explain why. You’ll still get a plausible explanation.

theluketaylortoday at 2:40 PM

True, but even a hallucinated explanation of where things went wrong added to the context can force the model down a better path over the next few inputs.

Kirotoday at 2:58 PM

The point of the parent post is that the explanation shows they made the error themselves, so it's immediately validated.

embedding-shapetoday at 2:46 PM

You can also literally tell them: "Here is your session ID: $ID, lookup the .jsonl session, trace exactly why this decision was being made, present evidence and concrete proof, no guessing or assumptions" and you'll get an evidence-based report without guesses.

show 1 reply