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 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.

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.

> 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.

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.

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.

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.

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.

It will cycle.

Without the content site the AI overview will become useless

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.)