[Tina Nguyen]

Sometimes ChatGPT gives me confident answers that turn out to be totally wrong (it invented a fake statistic once). How do you all deal with the hallucination problem?

[Marcus Bell]

Two things help a lot: 1) ask it to ‘only answer from information you’re confident about, and say if you’re unsure,’ and 2) never trust names, numbers, or quotes without verifying. Treat it as a fast first draft, not a source of truth.

[Ahmed Zaki]

Also try giving it the source material yourself — paste the article/data and say ‘answer using only this.’ It hallucinates way less when it’s summarizing something you gave it vs. recalling from memory.

[Author]

Posts