Researcher Anders Sandberg notes the study lacks a human-error baseline.
@robinhanson Does any of the cited research compare to human error rates on the same task?
“more than 60% of responses from AI-powered search engines were inaccurate.” https://www.wired.com/story/fact-checking-ai/
Many users dismissed the Wired article claiming AI search engines are inaccurate over 60% of the time as outdated propaganda and irrelevant to current models.
No Digg Deeper questions have been answered for this story yet.
@robinhanson Research on this topic from Feb 2025 is completely irrelevant now. And they don't even report the exact models they were using.
@kmaru1701 @robinhanson The idiomatic English is over, not past. Rolling Stone may hate technology and technologists but that’s not really notable, the NYT does. It is notable that Wired does it; they used to be techno-optimists.
@robinhanson I don’t necessarily disagree here but citing Wired is like citing Rolling Stone
@kmaru1701 @robinhanson Dunno, Rolling Stone isn’t noted for hating technology and technologists.
@bair82 @robinhanson Neither does Google. These models will always be inferior, because they are served to everyone for free.
@stimmtdochgarn1 @bair82 @robinhanson Whatever they have now is much subjectively much better than Gemini 2.0 flash non thinking that powered AI overviews in Feb 2025?
@BarryPCotter @robinhanson Sailed right past your head
@robinhanson March of 2025 is ancient history. Interestingly, Wired is much more inaccurate than it used to be.
@robinhanson I have noticed that google's has improved drastically as of late.
@robinhanson fr
@robinhanson wired article on ai inaccuracy. lol. imagine trusting googke's AI to tell you if it's accurate. thats like asking a politician if they're honest. need actual demand signals not just more noise
@robinhanson That’s okay, we’ll still keep drinking that garbage.
(We wouldn’t want to be accused of Luddism.)
@robinhanson Wired is the NYT of tech; a vessel for propaganda.
@robinhanson why don't they just come up with a truth algorithm? just filter for truth?
@anderssandberg @robinhanson There you go again Anders, trying to put things in perspective. 🤣❤️
@BarryPCotter @robinhanson Go away. You still don’t get it.
@SrivatsanS28660 @bair82 @robinhanson True, but I still find it is often frustratingly inaccurate, especially when it fails to itself perform a web search to validate its claims. Which is ironic, considering that web search is Google's thing and probably cheaper than LLM inference.