Just shredding brand credibility in a way that's hard to believe. To "google" used to mean "finding a definitive source for the truth of a claim." Now even non-experts have noticed that the AI summaries are frequently flat-out wrong.
To "google" used to mean "finding a definitive source for the truth of a claim."
Maybe I'm quibbling, but that sounds (wildly) wrong to me. I always understood "to google" as "to search for on the (visible, open) web." Those results could be great, but they were often absolute garbage.
My serious point being, yes LLMs are making everything much worse, but also the web has always been a source of garbage as well as of truth and definitive sources. The web was always dangerous in that Meno's-paradox kind of way: if you didn't already have a reasonably good sense of the likely answer, you couldn't really trust the result you got to be true.
I guess they believe that it hurts them less than not being "in the trend". If they don't use AI, many people would blame them for that.
Which is very sad, but it's not a new phenomenon. Kinda reminds me of Lenovo changing the design (and the keyboard layout) of their ThinkPads to look like a Mac. I still love their old-school design/layout much more than the current one. I guess I should be thankful they still put trackpoint there. Sigh.
Google Brain and DeepMind (another Alphabet subsidiary) are largely responsible for this 'trend'. They hyped the stuff, built their NPU, told everyone that AI was the future of everything. Satya Nadella panicked and went into massive Google-envy FOMO and announced that MS was building an 'AI supercomputer' to compete with Google's NPU clusters. OpenAI was founded by an ex-Google Brain researcher and a bunch of opportunist grifters wanting to cash in.
This happened to me too the other day and I'm still flabbergasted. I was looking up the origin of a quote and instead of search results, I got a very long AI overview reply to the question posed by the quote, as if I wanted an answer to it.
The reply was terrible, I searched for something that roughly translates to just "Am I correct" and it immediately gassed me up with "Yes, you are absolutely correct!". Correct about what?!
But even if it were close to sensible, it was in a completely different domain from what I was doing, namely searching.
As much as I'm not a huge fan of this, I'm not sure it isn't the "right" move for much of their general audience.
I've seldom seen less technical people who really understand search queries and use them effectively... The normal interaction seems to be more of the flavor of "type out some stream of consciousness sentence fragments of what I'm thinking about" and then if they don't see the result they wanton the first page, assume it doesn't exist or isn't findable. People seldom effectively iterate queries that don't immediately turn up results, or if they do, they sometimes treat them like a chat where they assume that the context of their first search will automatically carry over to the next one. I suspect encouraging that sort of chat interaction actually gets many people better search results than they'd find otherwise.
I tried the article's search query and got a similar AI response... but when I went back and forth with a couple short lines ("no i meant dario saric", "i want the memes") it quickly honed in on what the author was looking for with both an explanation of the context and history of the meme, all linked and cited.
I tried the article's search query and got a similar AI response... but when I went back and forth with a couple short lines ("no i meant dario saric", "i want the memes") it quickly honed in on what the author was looking for with both an explanation of the context and history of the meme, all linked and cited.
You could do this because in this case the "ai oververview" was so obviously nonsensical. What about all the cases where it kinda looks sensible, but is wrong?
I've seldom seen less technical people who really understand search queries and use them effectively...
I don't love how often I see people in tech talk about "normal people" as if they are incapable of learning, and need to be constantly babied by some tech solution poorly attempting to guess what they want.
The normal interaction seems to be more of the flavor of "type out some stream of consciousness sentence fragments of what I'm thinking about" and then if they don't see the result they wanton the first page, assume it doesn't exist or isn't findable.
And that worked pretty well. Ultimately if you are looking for an answer to some question someone had asked previously, the question itself is an adequate search query.
You could do this because in this case the "ai oververview" was so obviously nonsensical. What about all the cases where it kinda looks sensible, but is wrong?
What about traditional search results that place well and look authoritative but are wrong? This isn't a new problem. If the user is unable or unwilling to vet the response, I don't have a solution to that, but not gonna pretend there was one before either.
I don't love how often I see people in tech talk about "normal people" as if they are incapable of learning, and need to be constantly babied by some tech solution poorly attempting to guess what they want.
Meeting people where and how they work is often a hallmark of good UI/UX. I hate chat interfaces being bolted on everywhere, without thought to whether they're an effective UI, just because LLMs make the "easy." But this actually seems like a good use case for one, where you're doing iterative refinement of a natural language query over a natural language corpus.
And that worked pretty well. Ultimately if you are looking for an answer to some question someone had asked previously, the question itself is an adequate search query.
To the degree it works ok today is a pretty dramatic change from how it was, and presumably based on quite a bit of engineering by google, which I'd assume was already using language models behind the scenes, even if it was exposing them in the traditional search result format.
carlana | 10 hours ago
Just shredding brand credibility in a way that's hard to believe. To "google" used to mean "finding a definitive source for the truth of a claim." Now even non-experts have noticed that the AI summaries are frequently flat-out wrong.
telemachus | an hour ago
Maybe I'm quibbling, but that sounds (wildly) wrong to me. I always understood "to google" as "to search for on the (visible, open) web." Those results could be great, but they were often absolute garbage.
My serious point being, yes LLMs are making everything much worse, but also the web has always been a source of garbage as well as of truth and definitive sources. The web was always dangerous in that Meno's-paradox kind of way: if you didn't already have a reasonably good sense of the likely answer, you couldn't really trust the result you got to be true.
[OP] dimonomid | 9 hours ago
I guess they believe that it hurts them less than not being "in the trend". If they don't use AI, many people would blame them for that.
Which is very sad, but it's not a new phenomenon. Kinda reminds me of Lenovo changing the design (and the keyboard layout) of their ThinkPads to look like a Mac. I still love their old-school design/layout much more than the current one. I guess I should be thankful they still put trackpoint there. Sigh.
david_chisnall | 9 hours ago
Google Brain and DeepMind (another Alphabet subsidiary) are largely responsible for this 'trend'. They hyped the stuff, built their NPU, told everyone that AI was the future of everything. Satya Nadella panicked and went into massive Google-envy FOMO and announced that MS was building an 'AI supercomputer' to compete with Google's NPU clusters. OpenAI was founded by an ex-Google Brain researcher and a bunch of opportunist grifters wanting to cash in.
marginalia | 7 hours ago
Not gonna lie every time I see Google updating their search product, as someone operating in the same market, it has me like that sickos meme.
rski | 9 hours ago
This happened to me too the other day and I'm still flabbergasted. I was looking up the origin of a quote and instead of search results, I got a very long AI overview reply to the question posed by the quote, as if I wanted an answer to it.
The reply was terrible, I searched for something that roughly translates to just "Am I correct" and it immediately gassed me up with "Yes, you are absolutely correct!". Correct about what?! But even if it were close to sensible, it was in a completely different domain from what I was doing, namely searching.
abeyer | an hour ago
As much as I'm not a huge fan of this, I'm not sure it isn't the "right" move for much of their general audience.
I've seldom seen less technical people who really understand search queries and use them effectively... The normal interaction seems to be more of the flavor of "type out some stream of consciousness sentence fragments of what I'm thinking about" and then if they don't see the result they wanton the first page, assume it doesn't exist or isn't findable. People seldom effectively iterate queries that don't immediately turn up results, or if they do, they sometimes treat them like a chat where they assume that the context of their first search will automatically carry over to the next one. I suspect encouraging that sort of chat interaction actually gets many people better search results than they'd find otherwise.
I tried the article's search query and got a similar AI response... but when I went back and forth with a couple short lines ("no i meant dario saric", "i want the memes") it quickly honed in on what the author was looking for with both an explanation of the context and history of the meme, all linked and cited.
hjvt | an hour ago
You could do this because in this case the "ai oververview" was so obviously nonsensical. What about all the cases where it kinda looks sensible, but is wrong?
I don't love how often I see people in tech talk about "normal people" as if they are incapable of learning, and need to be constantly babied by some tech solution poorly attempting to guess what they want.
And that worked pretty well. Ultimately if you are looking for an answer to some question someone had asked previously, the question itself is an adequate search query.
abeyer | an hour ago
What about traditional search results that place well and look authoritative but are wrong? This isn't a new problem. If the user is unable or unwilling to vet the response, I don't have a solution to that, but not gonna pretend there was one before either.
Meeting people where and how they work is often a hallmark of good UI/UX. I hate chat interfaces being bolted on everywhere, without thought to whether they're an effective UI, just because LLMs make the "easy." But this actually seems like a good use case for one, where you're doing iterative refinement of a natural language query over a natural language corpus.
To the degree it works ok today is a pretty dramatic change from how it was, and presumably based on quite a bit of engineering by google, which I'd assume was already using language models behind the scenes, even if it was exposing them in the traditional search result format.
telemachus | an hour ago
In related news, for the moment,
-aiin a search on Google still seems to get you results without LLM summaries or responses.Also in related news, probably as a result of this post, Google seems to have taught their LLM the joke: https://imgur.com/a/DL8BgEm.