Why you have to be so careful accepting answers from AI

Interesting piece today in the FT magazine by Tim Harford. He relays an anecdote about encountering a fellow runner on his way to the start of the London marathon in Greenwich. This other guy had used Chat GPT to advise him how to get to the start. He'd been told to use the Elizabeth Line to go to Liverpool St and thence to Greenwich. But there is no train from Liverpool St to Greenwich.

The point of the article was why this guy had elected to use Chat GPT rather than Google Maps, which would have given him the correct information. The reason is that Chat GPT wrapped up the wrong advice in a nice, chatty package in perfect English, complete with a rationale for avoiding the Circle Line as it would be busy, and so forth. So this guy was suckered by the slick presentation and the human-seeming style of communication into trusting it, when it was actually talking crap.