If you ask a co-worker to do some research for you, you probably aren’t worried they’ll fabricate it. You’re more likely to worry about things like motivation and prioritization.

If you ask an AI, you won’t worry about its motivation or priorities. But you should probably worry about fabrication and hallucination.

You’ve got to game them, to some extent. In a very different way than you game a person.

-Paul Bloom on using LLMs