Comment by raincole
11 hours ago
I have to admit, nowadays Google AI Overview's accuracy is so good that I often don't check the links. It's scary that it got from 'practically useless' to 'the actual google search' in less than two years.
I really don't know where the internet is heading to and how any content site can survive.
It's because the AI overview is most of the time directly summarising the search results rather than synthesizing an answer from internal model knowledge. Which is why it can hyperlink the sources for the facts now. Even a very dumb lightweight model can extract relevant text from articles
I just can't see how this is sustainable since they are stealing from the sources who are now getting defunded.
> I just can't see how this is sustainable since they are stealing from the sources who are now getting defunded.
Yeah, that's why I said I don't know where the internet is heading to.
You can see the fall in real time - half the sources are also dubious AI slop now and that number’s only growing :-/
At work the conversation is that simultaneously everyone is using LLMs now, yet we receive virtually no traffic through them. The LLMs scrape our data, provide an answer to the user, and we see nothing from it.
3 replies →
> I have to admit, nowadays Google AI Overview's accuracy is so good that I often don't check the links. It's scary that it got from 'practically useless' to 'the actual google search' in less than two years.
It says things I know to be false fairly regularly. I don't keep a log or anything, but it's left an impression that it's far from reliable.
For my anecdote, I don't frequently deign to look at the overview at all... but every time I have, it has been completely and totally wrong. There's probably some selection bias going on in when I choose to try looking at it again, but still notable that the frequency is that high.
Today I searched something and almost pasted the output into an internet forum discussion I was having. But I decided to check the wikipedia source just to make sure. The AI summary was not quoted directly from wikipedia, and it got some major aspects wrong in its summary. Lesson learned.
It is simply summarizing the top few search results. If they are false then the summary will be false.
> I have to admit, nowadays Google AI Overview's accuracy is so good that I often don't check the links.
You would know how?
The links contradict or do not support the overviews often in my experience.
You should be checking the links more often, IMO. I've seen it respond a number of times with content that is not supported by the citations.
While trying to find an example by going back through my history though, the search "linux shebang argument splitting" comes back from the AI with:
> On Linux and most Unix-like systems, the shebang line (e.g., #!/bin/bash ...) does not perform argument splitting by default. The entire string after the interpreter path is passed as a single argument to the interpreter.
(that's correct) …followed by:
> To pass multiple arguments portably on modern systems, the env command with the -S (split string) option is the standard solution.
(`env -S` isn't portable. IDK if a subset of it is portable, or not. I tend to avoid it, as it is just too complex, but let's call "is portable" opinion.)
(edited out a bit about the splitting on Linux; I think I had a different output earlier saying it would split the args into "-S" and "the rest", but this one was fine.)
> Note: The -S option is a modern extension and may not be available
But this, … which is it.
We must be using different Google.com.
Sometimes I use a completely meaningless combination of keywords by mistake, and AI Overview will happily make up a story telling me what I am looking for.
Of what relevance is this to this discussion?
It is scary but also exciting. As long as there are humans making informed decisions, there will be demand for quality sources of information. But to keep up with AI, content sites will need to raise their standards. Less intrusive ads, less superficial stuff, more in-depth articles with complex yet easily navigable structure, with layers of citations, diagrams, data, and impeccable accuracy. News articles with the technical depth of today's dissertations.
For AI to steal and summarize without attribution. These sites you talk about exists today but are dying because of AI.
Well, I hope you take this story as a caution that you shouldn't do that in any way that can seriously compromise your career/health/finances.
Try searching for something niche. You'll get a confidently wrong and often condescending answer.
The ai summary has been wrong so many times for me. Not that I ever trusted it anyway.
I think content sites will need to rely on supporters (ala patreon or substack). It's shitty but it's what the internet has come to
I have seen it be utterly wrong so many times recently I'm considering permanently hiding it. For instance, googling for "Amiga twin stick games" it listed a number of old, top-down, very much single axis games like Alien Breed as examples.
Really? I’ve noticed that the AI overview is full of glaring issues repeatedly. It’s akin to trusting the first Reddit post that is found by Google.
It will cycle.
Without the content site the AI overview will become useless
I know people love to hate on the AI overviews, and I'm a person who generally hates both google and AI. But--I see them as basically good and ideal. After all most of the time I am googling something like trivial, like a simple fact. And for the last decade when I have to click into sites for the information it's some SEO spam-ridden garbage site. So I am very glad to not have to interact with those anymore.
Of course Google gets little credit for this since it was their own malfeasance that led to all the SEO spam anyway (and the horrible expertsexchange-quality tech information, and stupid recipe sites that put life stories first)... but at least there now there is a backpressure against some of the spammy crap.
I am also convinced that the people here reporting that the overviews are always wrong are... basically lying? Or more likely applying some serious negative bias to the pattern they're reporting. The overviews are wrong sometimes, yes, but surely it is like 10% of the time, not always. Probably they're biased because they're generally mad at google, or AI being shoved in their face in general, and I get that... but you don't make the case against google/AI stronger by misrepresenting it; it is a stronger argument if it's accurate and resonates with everyones' experiences.
> -I see them as basically good and ideal. After all most of the time I am googling something like trivial, like a simple fact. And for the last decade when I have to click into sites for the information it's some SEO spam-ridden garbage site.
What good is it if the overviews lie some percentage of the time (your own guess is 10%) and you have to search to verify that they aren't making shit up anyway. Also, those SEO spam-ridden garbage sites google feeds you whenever you bother to look past the undependable AI summaries are mostly written by AI these days and prone to the same problem of lying which only makes fact checking google's auto-bullshitter even harder.
> I am also convinced that the people here reporting that the overviews are always wrong are... basically lying?
https://en.wikipedia.org/wiki/Availability_heuristic
No one remembers when AI Overview gets the answer right (it's expected to do so after all) but everyone has their favorite examples of "oh stupid AI."
That's incomplete, because another "nobody remembers" is when the hallucination differs from reality, but the reader doesn't promptly detect the problem and remember where they got it from.
Think about the urban legends in the style of "the average person eats X spiders per year." It's extremely unlikely that Rumor Patient Zero is in a position to realize it's wrong, or that they will inform the next person that it came from an LLM summary.
Uh, really? In my experience, at least a quarter of the info it gives me is usually manufactured or incorrect in some critical way.
In fact, if you switch to "Pro" mode, it frequently says the complete opposite of what it claimed in "Fast" mode while still being ~10-20% wrong. (Not to say it's not useful — there's no better way to aggregate and synthesize obscure information — but it should definitely not be relied on a source of anything other than links for detailed followup.)