Emad Mostaque
speaker
1,351 appearances
1 recordings
1 series
first heard Dec 2024
last heard Dec 2024
Emad Mostaque’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsNo recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.
Appearances
I say those are probably the areas that I'm most concerned about versus very large scale model emergent race away behavior.
But if there was something in race away behavior, I think probably less concerned about the labs versus maybe some of the decentralized stuff, because that can actually go very quickly.
Like we've had one of the first agents from hyperbolic that can provision its own computer now.
And I think we're moving to decentralized autonomous organisms effectively.
Actually, there's probably one exception to that, which is that if you look at OpenAI, Anthropic and others, and the relationships with Andrew and the defense industrial complex, that's probably not going to create aligned AI.
Like many listeners may remember the Slaughterbots mini documentary about drones, swarms basically killing the bad guys.
It w anyone the government said was a bad guy at the end.
Just look at Andreel's latest commercial.
And then if you think about entire teams at the top AGI labs now being Colonel Defense, you're probably going to have models that are aligned to command, control.
and more extreme things versus models that are aligned to being helpful, humane and others.
So I think that is again something that's somewhat inevitable, but
potentially dangerous, particularly when they're used in offensive capability perspectives.
Like we're starting to see the first AI pen attacks and other things like that, utilizing generative AI.
And they're getting very good very quickly.
But even if we take a step back and we think about the first principles of this.
All wars are based on the lie that humans aren't humans.
And any reasonable intelligence can determine that.
Especially an O one level intelligence.
So in order to get an intelligence to kill humans, or to authorize that type of thing, or to aid that, you basically have to teach it to lie.
Because again, just a step back would say that this is ridiculous.
Showing 81–100 of 1,351 · page 5 of 68
← Previous
Next →