Google is very keen on its AI Overview feature, rolling it out across the world and spending a lot of effort to minimize embarrassing moments like supposedly telling people to eat rocks. But apparently, those improvements don't always include things like the current year.
A viral Reddit post this week showed an AI Overview declaring it is still 2024 to question “Is it 2025?” Google soon claimed to have fixed the issue, according to an Android Authority report, but I found that to be only sometimes true.
Most of the time, the AI Overview provides accurate answers. It actually appears to be a manual fix because it's phrased as "Yes, according to the provided information, the current year is 2025," even though I had not provided any information. Around a tenth of the time, though, I get "No, it is not 2025. The current year is 2024."
That's odd enough, of course, but that's nothing compared to when I apparently caused a time paradox and the AI Overview said, "No, it is not 2025. The current year is 2025. According to a calendar, we are in the year 2025."
Overlooked AI Overview
Firstly, it's not 2025; it's 2025 is a good reminder that generative AI is often just a fancy autocomplete tool that doesn't have to make sense for the algorithm to function. Also, "a calendar" suggests that others disagree for some reason.
It's hardly a catastrophic bug, but it does bring back memories of AI Overview's other awkward moments. Most famously, the AI Overview briefly told people to “eat one small rock per day” for digestion and “put glue on pizza to make the cheese stick better.” When a product bills itself as an amalgamation of reliable sources but can’t reliably tell what year it is, it makes you wonder how much else it’s getting wrong.
Google calls these kinds of results edge cases. And most of the time, AI Overview gives you something sensible. It’s usually correct. It’s just that the times it’s not are sometimes so bizarre you might forget about how good its record is.
AI search is very much on the rise at Google. Search is also a central preoccupation for many AI developers, such as Perplexity and OpenAI. But since AI doesn’t quite understand the difference between fact and fiction, you'd do well to clarify and confirm whatever it tells you.
