You know how Google’s new feature called AI Overviews is prone to spitting out wildly incorrect answers to search queries? In one instance, AI Overviews told a user to use glue on pizza to make sure the cheese won’t slide off (pssst…please don’t do this.)
Well, according to an interview at The Vergewith Google CEO Sundar Pichai published earlier this week, just before criticism of the outputs really took off, these “hallucinations” are an “inherent feature” of AI large language models (LLM), which is what drives AI Overviews, and this feature “is still an unsolved problem.”
Good lord what is wrong with the people in this thread. The guy is literally owning up to the hard limitations of LLMs. I’m not a fan of him or Google either, but hey kudos for being honest this once. The entire industry would be better off if we didn’t treat LLMs like something they’re not. More of this please!
But it’s not an isolated R&D project. They’re rolling it out in general search. If I have a promising new braking technology, but which still only works well 48% of the time, I’d keep working on it but not put it in production vehicles.
I think @[email protected] has the right idea on how to handle this type of issue. Hopefully they will improve the messaging around this, because I’m getting really tired of explaining to people how what we have is not true AI.
Honestly though this is nothing new for Google, they’ve been providing answers and web results with false information since their inception. You as a user will always need to do some vetting, Google is never going to be able to give you fully accurate information. They’re just sending you to places that may contain more information on the topic you’re searching for. Or at least, that’s how you should use them.
I swear I’m not just trying to start an argument, but I don’t see the disagreement here. You’re saying people here are too negative, but people aren’t shitting on the idea of LLMs, but the over promising of what they can do. You’re tired of explaining that it’s not true AI, but the confusion of caused by Google calling it “AI Overviews.”
You say it’s nothing new and that we’ve always had to vet sources when Google sends us somewhere, which is true, but the Overviews aren’t sending people anywhere, they’re summarizing and trying to give you an answer. They do link to sources for now, but the end goal is clearly that we trust the summary without following the links.
People who are listening to and parsing his comments are not the same people who will be blindly consuming these “AI Overviews.” It’s a problem.
I’m saying most of the time people are correctly complaining about the over-promising that’s happening around AIs. Now here’s an example of a CEO acknowledging the limitations, and yes; perhaps still over-promising on some degree. But the fact that we’re seeing actual acknowledgement of the limitations is a positive thing. Change doesn’t happen overnight, but this is a step in the right direction.
I say all this as someone with a strong distaste of modern google. I actively avoid their services as much as reasonably possible. I’ve tried their AI and found it to be more trouble than it’s worth (what the hell is that control panel). But I can still recognize a positive change when I see one, my distaste of the company doesn’t change that.