Your AI did not "go rogue"it's just a program.

[em0nm4stodon@infosec.exchange]

@TheEvilSkeleton It's an error or a bug, computer programs can't hallucinate or lie.

[nieuemma@mastodon.de]

@TheEvilSkeleton @Em0nM4stodon I would probably say it "produced incorrect output"

[orb2069@mastodon.online]

@Jalella

Wait, you think the guys with nearly limitless bandwidth and compute - Not to mention a demonstrated lack of morals and a failure to grasp consent - might be running bot scripts on social media to try and astroturf their viewpoints on the unsuspecting?

Gee, I wonder where Musk got his crack team to take control of twitter from? How he managed to spin up Grok on almost no notice?

@Em0nM4stodon

[aydakikio@mastodon.social]

@Em0nM4stodon
Most of the time, I assume vibe coding is kinda cheating if you're not familiar with that concept or tool or language
The hallucinations in AI models will be solved sooner or later......the key thing to solve that is to formulate the doubt as a mathematical equation

[bedifferent@mastodon.social]

@Em0nM4stodon Late 1970s had my first permanent job w/a major computer joint. Worked in DC when Kathleen Graham was WaPo.

This is the gist of what happened. It was decades ago, so memory is yellowing.

WaPo published article w/headline , Oops, Computer made Mistake, Thousands not Paid! Friend&I saw the line. We wrote a letter to the ed correcting them saying computers don't make mistakes, people do.Our cmpy wrote code?

Published next day w/our names and where we worked. Not fired.

[felipeb@hachyderm.io]

@Em0nM4stodon

[cav@infosec.exchange]

@Em0nM4stodon agreed

[supermoosie@mastodon.au]

@aydakikio @Em0nM4stodon

No, the crappy output won't be solved.

[markc568@beige.party]

@StealthyTango @Em0nM4stodon The computer did this illegal thing. A computer can’t commit a crime or be charged as such, but we’re *so* sorry. Pinkie promise it won’t happen again!

[budududuroiu@hachyderm.io]

@Em0nM4stodon sigh.. it's not code... It's perfectly useful, when talking about alignment, to refer to it's behaviour

[berryfigapple@mastodon.social]

@Em0nM4stodon i ask it statistics, about songs, what's the word for this definition, etc. It is my reference librarian.
Yes, it makes mistakes. So?
I believe in not believing. So, that's life.

[tsyum@thepit.social]

@Em0nM4stodon yep! I believe the EU just ruled that Google is responsible for their so-called AI search results, so it could be interesting to see the ramifications of this court case. Let’s keep on the pressure

[imalcolm@mastodon.social]

@SuperMoosie

Exactly. Random errors (so-called "hallucinations", which is just bullshit anthropomorphism by slop-merchants and the AI-brain-pilled) are absolutely inherent to how LLMs function. Their weaknesses are absolutely baked in, and cannot be "solved".

@aydakikio @Em0nM4stodon

[donchacale@mastodon.social]

@Em0nM4stodon
robot vomit

[ghouul@social.vivaldi.net]

@kekekuku @Em0nM4stodon *To the wrong computer program.* Computers have been able to automatically gen. text for a very long time now, LLMs and image generation models do the same, but use a ton of ripped off content from a thousand places as training for this text generation. Or should I say text regurgitation? Ultimately, we already have search engines, which are not perfect, disinfo is common online. However, at least the search engine won't automatically gen. dangerously wrong medical advice, and make it sound accurate and authoritative, just because I was wondering if maltodextrin was a safe additive!

I can understand wanting to gen. a bunch of nonsense text, there are much older programs which do just that; and to your point the use of LLMs as a search engine certainly qualifies as a mistake by the human user. However, it is also an error of Microsoft to have included an AI overview, before the other listed websites on their search engine. If the AI is not meant to be used as search engine, then why search engine shaped?! Why market LLMs as a search engine replacer, if you aren't supposed to use them to look for accurate information of a very broad range of topics? AI is a paper tiger, useful at first glance, but the use cases dissolve under any pressure.