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.”
He’s not owning up. He’s dismissing.
Pichai acknowledges the problem (how could you not, it’s obvious the problem exists), but in a way that basically goes “Yeah, but it’s cool, nothing is prefect”, shrug, smile.
The part of the problem he’s pointedly not acknowledging is this; if the answers generated by this system are so unreliable that we have to double check them every time, using traditional research methods, what is the point of having the AI there in the first place?
More ad impressions.
That’s fair, but that doesn’t appear to be the rationale by most commenters here. I think your point of view is much more constructive and opens up some interesting discussion topics rather than circle jerking over “google CEO bad”.
I think the reason Google didn’t release their AI before OpenAI did is precisely for this type of issue. OpenAI is now forcing their hand, because often it’s not about the best product that wins, but rather the one that got to market first. I feel like what we’re seeing now is less about these companies trying to release a product that they strongly believe has user value, but rather it’s about these companies creating some sort of foothold such that by the time they figure out what the actual product is they still have some capacity to sell it.
I seriously wonder how the industry would look if OpenAI had not played their hand so soon. Obviously the technology would’ve arrived regardless, but perhaps the packaging would’ve been different enough that the miss-understanding around the technology wouldn’t be so widespread.
Going back to my original point, it feels like most commenters here are still miss-understanding the technology, because there isn’t anything to fix. Making LLMs smarter is impossible, it’s inherently “dumb”.