Google Search results polluted by buggy AI-written code frustrate coders

Google has indexed inaccurate infrastructure-as-code samples produced by Pulumi AI – a developer that uses an AI chatbot to generate infrastructure – and the rotten recipes are already appearing at the top of search results. This mess started with Pulumi's decision to publish the result of its users' prompts on a curated AI …

[The Dogs Meevonks]

If you still actually use google

If you want to clean up your google searches to clear out a lot of the AI generated bollocks and disinformation.

Use the following at the beginning of your search

"before:2023"

That should filter out a lot of the SEO & AI bullshit.

[Red Ted]

Re: If you still actually use google

It all comes round again...

My spam-filled search index is bigger than yours! in this very organ from 19 years ago!

[DS999]

Re: If you still actually use google

That's fine today, but what about a few years from now when you're searching for stuff about the 2024 post election riots Trump instigates or whatever unseats Rust as the language of the moment everyone is coding in?

You'll need to use AI to search and tell it to filter out anything it thinks was written by AI!

[Pascal Monett]

Well there goes the neighborhood

we regularly hear 'Even if imperfect, we prefer to have something 80 percent correct, [rather] than nothing at all'."

Great. Looks like Borkzilla has always been ahead of the curve after all.

So, this is the future of programming. This explains Crapita, Fujitsu et al. to a tee.

I am not reassured.

[Mishak]

Hopefully

This will mean "professionals" get more work* at better rates. Though I would want even more if the task was to sort out an AI-generated mess...

* Assuming they can be found.

[Michael H.F. Wilkinson]

Re: Well there goes the neighborhood

Something that "works 80%" has a one in five chance of borking something important. It reminds me of the old days of PCs claiming 95% IBM-PC compatibility. These machines invariably ran into the remaining 5% causing important programs to crash,

[Neil Barnes]

Is anyone in any way even slightly surprised?

This was inevitable from the start.

A visit to an insanitary river - y'know, the sort where they just dump in the sewage - will show you what always floats to the top... and the more of it there is, the thicker the floating layer is.

[Norfolk N Chance]

Re: Is anyone in any way even slightly surprised?

It's CJD* for computers.

*Creutzfeldt-Jakob disease

[Paul Herber]

'the AI answers rising to the top of the search results overnight'

People spend years trying to get their own websites higher up the search results, in a legitimate manner.

What's the %$@&%£ point!

[abend0c4]

If everyone is trying to get their websites higher up in the search results, there truly is no point.

The cost of gaming the algorithm is too low but I suspect if Google were to make significant changes there'd be howls of anguish and swathes of lawsuits from those who'd slid down the rankings: as an effective monopoly supplier of search services, Google is between a rock and a hard place.

If they were to apply more AI to the assessment of search rankings, it would be fascinating to know whether it preferred the product of its own kind.

[Anonymous Coward]

You're presuming that they still understand. I suggest that they don't.

[David Hicklin]

I help run a riding club website and daily we get a huge number of SEO emails punting to get our website to the top of the search rankings

So fully agree, what is the point...the data lake is well and truly polluted

[Snake]

That's fun

"People spend years trying to get their own websites higher up the search results, in a legitimate manner."

Lol, err, no.

"Today I was googling various infrastructure related searches and noticed a worrying trend of [Pulumi AI] answers getting indexed and ranking high on Google results, regardless of the quality of the [AI] answer itself or if the question involved [Pulumi] in the first place"

Pinterest has entered the chat.

[that one in the corner]

"We move fast and try innovative new ideas regularly"

Translation:

+++ "We throw any old crap at it and see what sticks to the wall"

> Despite the challenges

+++ Despite our total lack of ability to even test before we publish

> Duffy expects AI will improve over time

+++ damn it, we must get lucky someday! More monkeys, more typewriters!

[Michael H.F. Wilkinson]

+++ Despite our total lack of ability to even test before we publish

Testing code before publishing: now there's an idea!

In our programming courses, we have automated test suites for answers to programming exercises, so by the time an answer reaches the TAs or lecturers we know it meets some basic functionality standards. Setting up such a test suite does require actually understanding the problem, which current LLMs clearly don't, seeing how they happily hallucinate all sorts of non-existent stuff.

[An_Old_Dog]

Re: +++ Despite our total lack of ability to even test before we publish

Do YOU create your automated test suite, or is it provided by the instructor?

[The Dogs Meevonks]

Re: "We move fast and try innovative new ideas regularly"

If you put a million monkeys in a room with a million type writers... eventually you'll have to clean up all of the monkey shit.

[Anonymous Coward]

"The number one Google result was an official Pulumi documentation page"

Er.... Nope. The first half page or more of search "results" from Google would be a stream of nearly but not quite completely irrelevant and utterly disinteresting adverts and sponsored results.

[An_Old_Dog]

.... results paid for by the Sirius Cybernetics Corporation.

[anonymous boring coward]

Oh, what a massive surprise...

[wknd]

Just ran the test for "AWS Lightsail xray" in private mode and with AdBlock :

- Google -> first

- Bing -> first

- DDG -> seven

- Kagi (no private mode obviously) -> fourteen

And I just blocked the domain on Kagi so I won't be bothered with it in the future.

The real issue, like many people spoke about already, is about the fact that current/future models are trained on content produced by other models (and probably their own older selves), which is basically inbreeding and will make it harder and harder to access to actual good content.

[Mike 137]

The root cause maybe?

"The real issue, like many people spoke about already, is about the fact that current/future models are trained on content produced by other models"

It quite possibly goes deeper than that -- down to product and service vendors not offering real support any more. This leaves users to the mercies of third party advice offerings and quality-blind search tools, so verifiable reliability inevitably falls through the floor. If the primary source were accessible to deal with these queries, the problem would not occur. The use of AI merely escalates a problem that has been exemplified for ages by crowd sourced advice sites such as Stack Exchange.

[Mishak]

DDG

It's #2 for me with the same setup.

[Michael Hoffmann]

Re: DDG

#5 for me. Interesting variation.

[MiguelC]

Re: DDG

For me it doesn't even appear on DDG, I only get news reports about this result pollution. Weird DDG behaviour.

[The Dogs Meevonks]

As everyone (inc me keeps repeating) who will bother reading what no one could be bothered to write?

Oh... other AI's reading other AI's in a never ending loop of regurgitated ever more inaccurate bollocks until it collapses under it's own ineptitude.... at least... that's my hope. let it come crashing down as swiftly as NFTs did.

[unimaginative]

For me Pulumni was first on Google and Kagi, second on Bing and DDG, and not on the first page at all on Brave.

[Marcelo Rodrigues]

I'm getting reasonably sane results with Brave. Ditched google altogheter.

[Howard Sway]

So, about the whole AI making programmers redundant thing.....

Sounds like the wailing has started from the "I just copy and paste everything from the internet" crowd. What a big shame for them. Next we have to wait for all this crap to clog up StackOverflow with "why doesn't this code work?" questions. And GitHub and copilot with even worse clueless buggy code. Which will train its next generation and see it deteriorate even further.......

If the end result is that those who've actually spent years learning stuff properly stop being undercut so much by all the chancers and can start demanding an ever bigger premium for true knowledge and experience, then it'll be a huge win over the fools who swallowed the hype that we would no longer be necessary.

[Mishak]

Back in the day

I remember a large automotive OEM saying to me "we won't need you contract types much longer as Simulink* will replace you".

It took them 1.5 years to implement what "we contract types" did in 6 weeks - and they never did answer "does it do the low-level device drivers as well"?

* model driven, not AI.

[Vincent van Gopher]

Move fast and break things

"We move fast and try innovative new ideas regularly – and sometimes they just don't work out the way we intended," . . . and break things, maybe like our current civilisation?

AI needs to be caged, not roaming loose.

Do dumb people know they're dumb? AI hasn't a clue if it's dumb or not and dumb people might think AI has all the answers when instead it's just being dumber than a dumb human.

Icon - will it all go bang? 'It' being anything you like :-)

[Alan Bourke]

Someone here mentioned Kagi recently

and I'm glad they did

[xyz123]

THIS is how you stop google etc from vacuuming up all your website data. Put available data thats just invisible to end-users in there.

Stuff thats wildly incorrect and even dangerous.

If you're an art website, put horrific fake images of puppies being mangled but tagged with "happy christmas" etc, but make them invisible to users.

For code, put stupid examples like if age <15 then show_porn = true

Screw with them until your stuff becomes untrustworthy to scrape.

[MattPDev]

Turn the internet off...

... It was good whilst it lasted.

[Faye Smelter]

Shit In > Worse shit out

Did nobody learn from "The Human Centipede" ?

[heyrick]

Google Search results polluted

Just leave it there.

So many times I've been looking for lyrics to obscure songs and "the main sites" appear in the list, only when you go there it's a "whoops, we don't have this" page.

Couple that with Google's desire to give you results that are "sort of maybe" what you are looking for even when you be specific; and their propensity for saying "we have xxxxxxx results" and then either simply giving up or returning utter rubbish after the first few screenfuls... the usefulness of the search is WAY inferior to what it was, say, a decade ago.

I think right now the reason most people Google is a, because it's the verb everybody remembers and b, it is the default option on many many devices. If not for those, they'd surely have lost market share due to being so bloody awful.

[Norfolk N Chance]

Re: Google Search results polluted

Indeed, that's been my experience also.

I suspect that searches such as these are not particularly monetizable and hence I frequently get that supercilious ice fishing troll, even when Google has returned pages for the search previously.

I close the browser and use any other search engine for the rest of the session or at least until I forget again...

[Matthew "The Worst Writer on the Internet" Saroff]

Enshittification

[OllieJones]

Next: garbage out; garbage in; second-generation garbage out. etc. etc. ad nauseam.

OK, it's happened that top-ranked search results are substandard Ai-generated results.

What happens when LLM training system ingest these top search results and builds them into its models?

I think the AI industry may soon discover its own kind of inbreeding, and will turn into its own equivalent of hare-lipped Hapsburg royalty, with notorious flaws. Especially since LLM training sets are getting close to the point where they've ingested all human generated content.

[Denarius]

Bring back Books

as above. Hard copy, curated and cross checked. Expensive compared to internet but, usually accurate, reliable and best of all, works with computer off. Also good for door stops if big enough.

[david1024]

AI is not the problem

AI is not the issue here. We need trusted authorities to issue asymmetric signing keys (we already have these!) And have humans sign off on articles they write or, with a different identity--have verified correct. Then we'll be able to start trusting people again. E.g. we need to add authentication to our text and graphical content so we can screen AI from human created and human verified content. And also ID humans that are not to be trusted. We have had the tech to do this for over half a century. Time to roll it out.

We can no longer trust the sources like NYT or journals. Time to start keeping score on poor reviewers and folks rubber-stamping peer reviews.

[technovelist]

No one could have predicted this!

Well, no one who believes the "AI" hype, anyway.