Paris MarxGoogle’s AI Overviews are providing “tens of millions of wrong answers … every hour — and hundreds of thousands every minute.”
wow, i love the AI future!
https://futurism.com/artificial-intelligence/google-ai-overviews-misinformation
Leander Lindahl 🇪🇺@parismarx "Studies have shown that people tend to trust what an AI tells them without question"
Frank Heijkamp
Tim Finnerty@leanderlindahl @parismarx Do these same people trust all other people without question as well?
Eric LawtonWe have taught people to trust computers because, properly programmed, they are very accurate at deterministic arithmetic problems, like balancing bank accounts, plotting flights to Mars and many more.
That trust carries over to LLMs but can soon be lost.
The tech bros are betting they can fix the accuracy problem before the trust is lost, which seems unlikely.
Haji Mokhtar Stork@parismarx I believe there is a lot of confusion over Google AI. "Google AI" is the umbrella term for the company's research division, but Gemini is the actual model we interact with. News reports often use "Google AI" as a catch-all when in fact Gemini is the consolidated brand replacing previous names like Bard and Duet.
Marianne@MokhtarStork @parismarx the article explains what the tests covered
Toni Aittoniemi@parismarx Russia was SEO optimizing for AI scrapers already in 2022, feeding psychological warfare narratives under the radar. They knew what’s up.
It’s a shame even our enemy knows more about us than the people we seem to worship.
@iveyline@parismarx That's one reason I no longer use Google. AI was the last straw.
leberschnitzel@parismarx I'm running an LLM at home, and when something doesn't work well with it I go "oh ok I won't use it for that".
And I wonder why that concept got lost for companies.
Well I don't actually wonder because it's money, but I wonder how it got that far
And I wonder why that concept got lost for companies.
Well I don't actually wonder because it's money, but I wonder how it got that far
^[[200~Michael@parismarx "A recent analysis conducted by THE AI STARTUP OUMI..." 😵💫
Dave Rahardja@parismarx I’ve met people who were excited that their LLM-based information retrieval systems got an 80% correct-answer rate. Any other tech that showed this dismal of a performance would have gotten laughed out of product review, but they couldn’t wait to push it to production.
Tara L. Campbell@drahardja @parismarx This is what pisses me off to no end. Not only do they get to shirk responsibility, and avoid accountability, they get to push half-assed products to market, making bank on hype and BS alone.
Meanwhile, promising ingenuity in the hands of conscientious and morally-sound people languish, or are lost completely, because there isn't a fat-ego'd, big-talking prick at the helm to sucker wealthy stupid people.
ramblingsteve@parismarx the joy of conflicting business metrics. Theirs is to serve slop at increasing volume, while for users its to make use of the slop at increasing volume. Errors are downstream of the token spend and they own the legislators, so it becomes a you problem for big tech.
Ray McCarthy@parismarx
There was study that showed mostly AI LLMs "shorten" rather than summarise and will likely miss a summary already in the document. They have zero context for what's important. No intelligence, so this is no surprise.
I now use Qwant for search. I've complained to Viber about the Summarise button they add on some versions.
projectmoon@parismarx I'd take it with a sprinkle grain of glue salt.
Arne Hanssen@parismarx As a former IT-collegue of mine used to say: "It's human to error, but to really mess up fast and in a large scale, you need computers!"
Coho@parismarx
Google lens only identifies plants accurately 10% of the time, it's has the worst rating, the best plant Id app only has a 50% accuracy . I wonder how many people have been poisoned eating something AI has misidentified. There are whole forums dedicated to stories about people who've followed AI map directions and ended up in extremely dangerous situations, in the middle of nowhere, and quite a few people have died getting bad directions.
Google lens only identifies plants accurately 10% of the time, it's has the worst rating, the best plant Id app only has a 50% accuracy . I wonder how many people have been poisoned eating something AI has misidentified. There are whole forums dedicated to stories about people who've followed AI map directions and ended up in extremely dangerous situations, in the middle of nowhere, and quite a few people have died getting bad directions.
@parismarx
When it gets to a point they can be in a room together, getting one cat playing with a string toy, often gets the other interested, and they start associating each other with good feelings and fun . I binged watched all seasons of that TV show " my cat from hell" with Jackson Galaxy , and was able to successfully troubleshoot some issues between my adult cat and the kittens I adopted. Jackson , has many episodes on getting cats to accept each other.
When it gets to a point they can be in a room together, getting one cat playing with a string toy, often gets the other interested, and they start associating each other with good feelings and fun . I binged watched all seasons of that TV show " my cat from hell" with Jackson Galaxy , and was able to successfully troubleshoot some issues between my adult cat and the kittens I adopted. Jackson , has many episodes on getting cats to accept each other.
GhostOnTheHalfShellI guess that means that right wing Talking Heads are out of a job? 💩
(((Klefstadmyr)))@parismarx Just wait until the next version comes, it'll be able to provide hundreds of millions, even billions of wrong answers every hour!
McSkinney@parismarx
Lie to me.
Lie to me.
Chris M@parismarx @pluralistic so they can roll out a better version for a fee. Whoever refuses to pay will be condemned to a life of ignorance and misinformation. Without going all H.G. Well's The time Machine: Knowledge schism
killickYou don't understand. You see, the answer is not "misinformation" if it gives the user the answer they want. /s