How to prevent AI from lying to you

Recently I found myself having to disprove something that Google’s AI made up. A reader of my Cord Cutter Weekly newsletter wanted to know why they couldn’t use the Apple TV app on their Roku after signing up for the service through the Roku Channel. This is in fact a notable downside with Roku’s subscription marketplace—you have to use Roku’s own apps and menu system to access the content—but Google’s AI overview was claiming otherwise. Turns out Google was basing its answer on a discussion thread from a random coding website I’d never heard of, and when I replicated the query myself, the source link didn’t even work. Google also cited a separate CNet article that didn’t back up the reader’s claim at all. Stories like this are why I never think of tools like ChatGPT or Gemini as trusted advisors, even if that’s how AI companies want you to see them. In my mind, they’re more like unreliable witnesses or interrogation suspects. They’re capable of provide useful information, but you might need to grill them to get it. This story first appeared in Advisorator, Jared’s weekly tech advice newsletter. Sign up to get free tech tips every Tuesday. Ask for sources (and check them) Every ma…