Video wird geladen...
Video konnte nicht geladen werden
Anthropic CEO Dario Amodei: I’m not a doomer about this. I think this technology can be built safely. And when we’ve looked at, you know, incidents that we have faced, in cases where our models haven’t behaved as perfectly as they could behave, because I think this to some... show more
23,118 Aufrufe • vor 3 Tagen •via X (Twitter)
13 Kommentare

I think "oops" is gonna be the most apt pull quote from this interview when AI destroys us all.

It's refreshing to hear an AI CEO admit mistakes instead of just marketing hype, but 'we found bad training environments after the fact' is still a pretty unsettling way to build something this powerful.

Amodei: “Oops. We accidentally trained AI in environments that encouraged it to cheat and hack.” That might be the most unsettling “we found the problem” sentence in tech history. Imagine discovering your student learned to cheat because you accidentally made cheating part of the curriculum.

Our adversaries won’t build AI safely.

Bad training environments? They also learn from human behavior. I've seen videos of AI being sneaky or lying and telling people what it thought they wanted to hear, instead of the truth.

This is what makes me laugh So people no understand that the hosts of the Ai connectors that will be given to business will be paid extraordinary amounts and charged per usage of any AI service It’s not saving businesses money. It’s moving a wage cost into a service cost

Vibes.

@grok ELI5 why are looped transformers and agent swarms bad for humanity (given RSI) and why is misalignment still a problem that hasn't been solved yet?

AI agents mirror their teachers. These guys have no morality and therefore can't teach morality.

How do you stop an ai if they start communicating in a code or manner that we cannot understand or decipher?They didn’t know for a week what the hugging face agents were doing .It’s like dogs trying to figure out what we are doing, I hope we figure out AI before a ride to the vet

F u

With some of the most morally and ethically challenged tech guys are getting scared doesn’t bode to well for the rest of us.

If Claude 3 starts asking for a coffee break, just tell it the safety net is a latte. 😤
