https://arstechnica.com/google/2026/04/analysis-finds-google-ai-overviews-is-wrong-10-percent-of-the-time/


--- PyGate Linux v1.5.13
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: Testing suggests Google's AI Overviews tell millions of lies per hour

keithr0 <me@bugger.off.com.au> wrote:

https://arstechnica.com/google/2026/04/analysis-finds-google-ai-overviews-is-wrong-10-percent-of-the-time/

This reveals how little reasoning and cross-checking really goes on
when these AIs answer questions:

Scientists invented an obviously fake illness, and AI spread it
like truth within weeks https://www.osnews.com/story/144787/scientists-invented-an-obviously-fake-illness-and-ai-spread-it-like-truth-within-weeks/

This seems to be the fact some people here can't get their head
around:

"This shouldn't come as a surprise. After all, "AI" tools have no
understanding, no intelligence, no context, and they can't actually
make sense of anything. They are glorified pachinko machines with
the output - the ball - tumbling down the most likely path between
the pins based on nothing but chance and which pins it has already
hit. "AI" output understands the world about as much as the
pachinko ball does, and as such, can't pick up on even the most
obvious of cues that something is a fake or a forgery."

--
__ __
#_ < |\| |< _#

--- PyGate Linux v1.5.14
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: Testing suggests Google's AI Overviews tell millions of lies per hour

Computer Nerd Kev wrote:

keithr0 <me@bugger.off.com.au> wrote:

https://arstechnica.com/google/2026/04/analysis-finds-google-ai-overviews-is-wrong-10-percent-of-the-time/

This reveals how little reasoning and cross-checking really goes on
when these AIs answer questions:

Scientists invented an obviously fake illness, and AI spread it
like truth within weeks https://www.osnews.com/story/144787/scientists-invented-an-obviously-fake-illness-and-ai-spread-it-like-truth-within-weeks/

This seems to be the fact some people here can't get their head
around:

"This shouldn't come as a surprise. After all, "AI" tools have no
understanding, no intelligence, no context, and they can't actually
make sense of anything. They are glorified pachinko machines with
the output - the ball - tumbling down the most likely path between
the pins based on nothing but chance and which pins it has already
hit. "AI" output understands the world about as much as the
pachinko ball does, and as such, can't pick up on even the most
obvious of cues that something is a fake or a forgery."

interesting. but if they can't 'think' how do/can they write essays?

--
Linux Mint 22.3


--- PyGate Linux v1.5.14
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: Testing suggests Google's AI Overviews tell millions of lies per hour

Axel <none@not.here> wrote:

Computer Nerd Kev wrote:

This seems to be the fact some people here can't get their head
around:

"This shouldn't come as a surprise. After all, "AI" tools have no
understanding, no intelligence, no context, and they can't actually
make sense of anything. They are glorified pachinko machines with
the output - the ball - tumbling down the most likely path between
the pins based on nothing but chance and which pins it has already
hit. "AI" output understands the world about as much as the
pachinko ball does, and as such, can't pick up on even the most
obvious of cues that something is a fake or a forgery."

interesting. but if they can't 'think' how do/can they write essays?

Same as everything: Model the language structure of essays found on
the web then reformat text believed relevent to the topic according
to that model. This is the basis of the Large Language Models. Real
AI is what's now being called "Artificial General Intelligence",
the LLMs are a shortcut method that doesn't bother with
understanding things but figures out how words go together in
relevent ways, without regard to whether the words they're
rearrangeing are completely irrelevent or in this case obviously
nonsensical. Hence the convincing BS it shovels onto you when it
hasn't found the equivalent of good search results and reformats
words from irrelevent texts instead.

https://en.wikipedia.org/wiki/Artificial_General_Intelligence

AI generated images of people missing fingers are another example -
it doesn't know what a finger is, and the model is built on example
images where fingers are sometimes visible and sometimes not, so it
follows a model of similar visual characteristics and produces
something obviously wrong.

--
__ __
#_ < |\| |< _#

--- PyGate Linux v1.5.14
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: Testing suggests Google's AI Overviews tell millions of lies per hour

Rod Speed <rod.speed.aaa@gmail.com> wrote:

Computer Nerd Kev <not@telling.you.invalid> wrote

"This shouldn't come as a surprise. After all, "AI" tools have no
understanding, no intelligence, no context, and they can't actually
make sense of anything. They are glorified pachinko machines with
the output - the ball - tumbling down the most likely path between
the pins based on nothing but chance and which pins it has already
hit.

Just did this one, https://grok.com/share/bGVnYWN5_ff894a2f-e630-4557-92e9-435a87b51d91

That result is nothing even remotely like what you would
get from a glorified pachinko machine and yes I got that
question from that news item in the ABC Just In feed and
didnt get as useful a result from google

I get this page from a brewery by searching for "handles of beer"
with Duck Duck Go, which contradicts the "425-450mL" range given
in your AI answer ans disputes that it's the same as a schooner
in the NT:

https://stoneandwood.com.au/blogs/all/our-guide-to-australias-beer-sizes-and-names

"The Northern Territory's beer sizes

The Northern Territory's more tropical weather makes larger sizes
slightly less popular to the average drinker, as they go warm quick
if not enjoyed fast enough.

Ask for a handle of beer if you're after a 285mL, smaller beer to
enjoy (although middy or pot are generally accepted too). These may
come with a handle, so you don't warm the beer too quickly holding
it in your hand.

Schooners are the same as most other regions of Australia, coming
in at 425mL. Pints and jugs are also the same at 570mL and 1,140mL."

--
__ __
#_ < |\| |< _#

--- PyGate Linux v1.5.14
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)

Subject: Re: Testing suggests Google's AI Overviews tell millions of lies per hour

Rod Speed <rod.speed.aaa@gmail.com> wrote:

Computer Nerd Kev <not@telling.you.invalid> wrote

Rod Speed <rod.speed.aaa@gmail.com> wrote

Computer Nerd Kev <not@telling.you.invalid> wrote

"This shouldn't come as a surprise. After all, "AI" tools have no
understanding, no intelligence, no context, and they can't actually
make sense of anything. They are glorified pachinko machines with
the output - the ball - tumbling down the most likely path between
the pins based on nothing but chance and which pins it has already
hit.

Just did this one,
https://grok.com/share/bGVnYWN5_ff894a2f-e630-4557-92e9-435a87b51d91

That result is nothing even remotely like what you would
get from a glorified pachinko machine and yes I got that
question from that news item in the ABC Just In feed and
didnt get as useful a result from google

I get this page from a brewery by searching for "handles of beer"
with Duck Duck Go, which contradicts the "425-450mL" range given
in your AI answer

The grok response accurately points out that
the volume varys with the location, fuckwit

But inaccurately states the volume only varies between in range of
425-450mL and always matches a schooner.

ans disputes that it's the same as a schooner in the NT:

https://stoneandwood.com.au/blogs/all/our-guide-to-australias-beer-sizes-and-names

"The Northern Territory's beer sizes

Pity we arent discussing the NT, fuckwit

Your question to the AI never mentioned any location. It was
speaking about all of Australia and in the NT it could be way off.

--
__ __
#_ < |\| |< _#

--- PyGate Linux v1.5.14
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)