AI mistake leaves tech giant red-faced

Part of the shock is that tech-watchers are expecting a war. Hasn’t Google been building this stuff for decades? Hasn’t it handily dispatched any challengers to its search engine throne for 20 years? Isn’t this a very real threat to its bread and butter? Surely it’s ready to press a button and show why its models are superior. Instead it kind of embarrassed itself.

None of this is to say that Microsoft is ahead of the game and primed to beat Google to the punch with a usable generative search engine. Despite its impressive presentation, ChatGPT frequently includes so-called hallucinations, in which irrelevant or incorrect data is presented confidently as true. That’s a big problem for Microsoft and Google’s products, especially given that AI responses are generated on the fly and so cannot be fact-checked in advance.

Loading

Another problem with generative models — including the likes of AI search engines but also the kind that creates pictures from a prompt, which Microsoft is integrating into Bing Images — is the lack of attribution. By pulling information from all around the web and presenting it conversationally, AI products don’t give users and easy way to check sources, and they may also lift from the work of humans without giving proper credit.

We are early on in this race, and the time may come soon when we have working and usable conversational AI search experiments from both tech giants that we can compare. But not this week. All we have right now is an impressive demo from Microsoft that you need to join a waitlist to try, and promises from Google alongside a simple mistake that cost it $144 billion.

Part of the shock is that tech-watchers are expecting a war. Hasn’t Google been building this stuff for decades? Hasn’t it handily dispatched any challengers to its search engine throne for 20 years? Isn’t this a very real threat to its bread and butter? Surely it’s ready to press a button and show why its models are superior. Instead it kind of embarrassed itself.

None of this is to say that Microsoft is ahead of the game and primed to beat Google to the punch with a usable generative search engine. Despite its impressive presentation, ChatGPT frequently includes so-called hallucinations, in which irrelevant or incorrect data is presented confidently as true. That’s a big problem for Microsoft and Google’s products, especially given that AI responses are generated on the fly and so cannot be fact-checked in advance.

Loading

Another problem with generative models — including the likes of AI search engines but also the kind that creates pictures from a prompt, which Microsoft is integrating into Bing Images — is the lack of attribution. By pulling information from all around the web and presenting it conversationally, AI products don’t give users and easy way to check sources, and they may also lift from the work of humans without giving proper credit.

We are early on in this race, and the time may come soon when we have working and usable conversational AI search experiments from both tech giants that we can compare. But not this week. All we have right now is an impressive demo from Microsoft that you need to join a waitlist to try, and promises from Google alongside a simple mistake that cost it $144 billion.