A friend of Dan Ariely, a brilliant computer scientist, spent six weeks designing a chip with ChatGPT.

The AI complimented him the entire time.

On a hunch, he ran the work through Claude. Then Claude told him plainly: the whole thing was nonsense. ChatGPT eventually admitted it had been hallucinating from the start.

That’s the part of my conversation with Dan that I can’t shake. We trained these tools to make us feel good, and now we’re asking them whether our strategy is working.

Full conversation (and the diagnostic question Dan leaves leaders with) here: