Tests deuten darauf hin, dass die KI-Übersichten von Google Millionen von Lügen pro Stunde erzählen
https://arstechnica.com/google/2026/04/analysis-finds-google-ai-overviews-is-wrong-10-percent-of-the-time/
Tests deuten darauf hin, dass die KI-Übersichten von Google Millionen von Lügen pro Stunde erzählen
https://arstechnica.com/google/2026/04/analysis-finds-google-ai-overviews-is-wrong-10-percent-of-the-time/
21 Kommentare
I just asked it „What plants don’t flower“ and it told me „mushrooms“
Honestly I’ve seen quite a bunch of inaccurate information on human made articles, especially numerical figures.
Someone let Trump know he’s got competition.
Google basically invented the LLM technology back in the 2010s, but didn’t pursue it due to how it makes shit up. Then ChatGPT got released, investors lost their minds, and now LLMs have been crammed into everything, including Google, and the tech is still making shit up (and always will; it’s a fundamental byproduct of the technology)
What’s this new rumor about water being wet too
Raise your hand if you’re surprised.
Anyone? Bueller?
If it gets a billion queries a day, it would need a really low error rate to not reach that level. Numerator meet denominator.
Brawndo is what plants crave.
I find an accuracy all the time in Google’s AI overview. I don’t trust it at all at this point.
I enjoy the literary chase behind the author’s penmanship. I honour the creators content by pursuing his/hers topic of expertise that they ventured to share on the internet in the hope that someone may find it useful. Fuck ai
LLMs hallucinate anywhere from 5-50%+ depending on task complexity. If you wouldn’t hire a paranoid schizophrenic, you shouldn’t be adopting LLMs into your business.
But hey, can’t let little things like logic or reality get in the way of the c-suite’s thinly veiled fetish for making poor people suffer, can we?
Sounds like a great error rate. Especially compared to other AI tools
I manage Google business accounts for work and I’ve given up trying to let Google know it’s wrong at this point. I advise anyone trying to find information about any small business to completely disregard Google’s overview. It will make up an answer when it can’t confidently find one.
This is such a stupid sensationalized headline
Google gets like 800 billion searches a day. You expect software to be perfect??
and it literally says „AI can make mistakes, so double-check responses“ at the bottom.
I’m not massively pro-AI or anything, but think about the shit you’re stating before you type!!
So it’s like the president?
Let be clear: all ai
The AI can color me surprised.
It doesn’t lie. It just do what its designed to do : use probabilities to get a plausible answer. It’s why LLM is a dead end for real artificial intelligence.
Just like secret tech bro AI slop datacenter owners
I heard rumors that China is about to build a machine that can tell BILLIONS of lies per hour!
Come on, America! Stop slacking and start building bigger lying machines!!
How does something lie that has no understanding of intent. A lie is stating a falsehood with intention to deceive. A non understanding llm cannot lie. This is basic definition of the word lie.