Follow

Google's AI search features are handy a lot of times, but this is a good reminder of why not to trust a neural net to parse info for you. Here Google claims only 6,142 humans were alive in 22AD, a figure that immediately seems impossibly low. Looking at the table it's pulling info from, we can see the AI is making 2 major errors here--first, it's pulling a figure from 2000AD, not "2000 years ago"--second, it's missed the column label that indicates this is 6,142 millions of humans, or 6.1B.

Also a good example of how neural nets are good at making associations, not necessarily... comprehending. The search AI matched up the pieces from the prompt to info out on the web, so it assumed that must be the answer! But it had no way of comprehending the broader contexts that made it actually a very wrong answer.

Sign in to participate in the conversation
Mastodon for Kansas City

Hey Kansas City, be friendly!