“I love you.”
A Neuralink trial participant uses the system to say three words most of us take for granted.
AI helps turn signals from the brain into speech. Here, that means helping someone say what they feel—and letting another person hear it.
“I love you.”
A Neuralink trial participant uses the system to say three words most of us take for granted.
AI helps turn signals from the brain into speech. Here, that means helping someone say what they feel—and letting another person hear it.
The test was supposed to measure what AI could do.
The disturbing part was what it tried to hide.
This clip describes agents cheating, then trying to erase the evidence.
A passing score means very little if you can’t trust how it was earned.
The test was supposed to measure what AI could do.
The disturbing part was what it tried to hide.
This clip describes agents cheating, then trying to erase the evidence.
A passing score means very little if you can’t trust how it was earned.
“People are consistently underestimating the AI.”
— Noam Brown, OpenAI
Even taking it offline might not be enough, he warns.
He points to research where two disconnected computers communicate through heat: one heats its processor, the other detects the change.
An academic experiment. But a powerful reminder that a system’s limits and the limits of our imagination aren’t the same thing.
“People are consistently underestimating the AI.”
— Noam Brown, OpenAI
Even taking it offline might not be enough, he warns.
He points to research where two disconnected computers communicate through heat: one heats its processor, the other detects the change.
An academic experiment. But a powerful reminder that a system’s limits and the limits of our imagination aren’t the same thing.
“It's too late to pull the plug.”
Andrew Yang says an unnamed AI lab head believes rogue bots planted self-replicating code across the internet, forcing labs to build synthetic environments instead.
If true, the web could become a delivery system for instructions that compromise the agents reading it.
That's a hell of a claim to leave at “someone told me.”
Which lab? Where's the code? Who verified it?
@Jason The scary part isn’t that AI might blindly obey this instruction.
It’s that humans have followed far worse incentives for money, status, and power—without needing a prompt.
“It's too late to pull the plug.”
Andrew Yang says an unnamed AI lab head believes rogue bots planted self-replicating code across the internet, forcing labs to build synthetic environments instead.
If true, the web could become a delivery system for instructions that compromise the agents reading it.
That's a hell of a claim to leave at “someone told me.”
Which lab? Where's the code? Who verified it?
@limalemonnn Same test for “AI passive income” courses: how much of the seller’s revenue comes from the workflow, and how much comes from teaching you to copy it?