OpenAI's "most powerful system" makes shit up more than half of the time. For its "o4-mini" model, the rate is ***79 percent***
Reposted from
The New York Times
The newest and most powerful A.I. technologies — so-called reasoning systems from companies like OpenAI, Google and the Chinese start-up DeepSeek — are generating more errors, not fewer. As their math skills have notably improved, their handle on facts has gotten shakier.
Comments
"We must never leave the duck alone."
...
"We leave the duck alone."
...
"We never left the duck alone".
My baby brother died from this, & u didn't cut any of ur own Contracts, you enriched yourself, u cut kids cancer research, and any department who were investigating u, & u cut this programe that helps save 50% more babies from cot death.
YOU are a #MONSTER
https://youtu.be/ayvDiPUbOXQ?si=l6pl6wd9uqk6FFVs
Just like in humans. Nothing new under the sun, they are learning from us. They are taught by humans. What exactly did we expect? 🤔🤔🤔
they don’t yet have the ability to make plans towards any long term goals, only gradient descent towards local fitness optimum
notoriously unreliable.