Never Trust, Always Verify

Transcript

In our course, students get an AI answer with three neat sources at the bottom. One checks out. One says the opposite. One doesn't exist.

The one that says the opposite is a real National Archives page. The answer says the Declaration of Independence was signed in 1783. The page it links to says 1776.

A few years ago, the warning sign was an answer with no sources. Now most chatbots give you sources, and that's the new problem. An answer with links looks like it's been checked, so nobody clicks.

But a link is a claim too. It only counts once you've opened it and found the sentence on the page. So there are two things to check about any source. Does it open? And does it actually say the thing?

To be fair, chatbots have gotten a lot better at this. Most links open now. That's exactly why people stop checking. It's right often enough that you stop looking for the time it isn't.

That's all "always verify" really means. A source you haven't opened isn't a source yet.