Human error? AI error?

Posted by

Still don’t believe that generative AI is breaking new ground? When was the last time you saw a human make a $100 billion mistake? Google showed off its answer to ChatGPT-infused Bing this week and things went…worse than expected.

In an advertisement Google released on Monday touting its new AI chatbot, Bard, the system made a factual error, confidently asserting that the James Webb Space Telescope snapped “the very first image of a planet outside our solar system”—although the milestone was actually reached 17 years before the JWST launched.

In response to the mistake, the company highlighted its rigorous testing process for Bard…which is kicking off this week. But between that and a hastily staged event (coming one day after Microsoft impressed with its souped-up Bing), Alphabet shares dropped ~8% yesterday, the most in over three months.

Zoom out: Bard’s confident wrongness isn’t uncommon in generative AI—ChatGPT’s ability to be boldly inaccurate is well documented. The systems work by analyzing patterns in a huge mass of text and making predictions about what words come next (rather than analyzing and synthesizing ideas, like our brains do). This has led one Princeton professor to refer to them as “bullsh*t generators.”


Comments are closed.