HN Explorer

7 comments

[ 0.33 ms ] story [ 7.1 ms ] thread
LoganDark
> In response, he started developing his own AI study tool, concluding: "If you can't beat them, join them."

I feel like a lot of people would be very offended by this approach. "Can't stop climate change? Start causing it instead!"

(I don't personally equate LLM use to climate change, but datacenters are an incredibly big problem right now.)

kleinicke
Yeah, China is currently gaining speed to reach the level of US AI Models. So it's clear they won't want to stop now. Also they would need to establish some trust with Trump to create some reliable regulation there. I guess it's very hard to trust or rely on Trump for anyone.
verdverm
The part of co-regulation that is rarely raised is verification, how do we verify the other is holding to their side of the agreement?

With the USSR and nuclear weapons, we went into each others sites.

I find it very unlikely America would let CCP operatives into our data centers with the access they need to verify.

chrisjj
I guess the idea is it would work like bio weapons. I.e. on trust.

This is unfortunate, since OpenAI and Anthropic have already demonstrated that to trust them would be suicidal.

verdverm
bio/chem weapons is a good analogy that I'm less familiar with, but likely sits in a closer place than nuclear weapons

the problem with "ai weapons" is that they are "invisible" in the physical world, whereas chemistry, biology, and nuclear weapons require IRL activity

chrisjj
[delayed]
verdverm
that was a missile that has existed for decades, you see the explosive part of the weapon, not the Ai in a data center prioritizing targets

lots of human failures in that incident, really out dated Intel that should have been flagged

closer to your intent is Ukraine's Ai enabled midstike campaign against Ruzzia