Anthropic and US Government, there can be only one right in this situation.

I think David Sacks is right, if you are saying you are building nuclear bombs, then prepared to be regulated like one.

There is no eating it while having it

I agree completely. If these things are so dangerous that they turn every person into an advanced persistent thread actor, capable of causing untold cyber destruction (oh, and they can make bio weapons etc), then they should be treated like the weapons they are.

> Anthropic and US Government, there can be only one right in this situation.

And just to be clear, that's a maximum of one right in this situation.

Yeah, LLMs are a national security issue on par with spellcheck.

LLMs are piloting EM-proof kamakazi drones and destroying logistics networks today.

The real threat isn't the drones, it's the ability to generate a target bank. Historically militaries that are not just carpet bombing have been bottlenecked by target selection, humans can only review and authorize so many strikes. Now the AI will select the targets and the bottleneck moves to how many bombs you can build.

https://en.wikipedia.org/wiki/AI-assisted_targeting_in_the_G...

That video makes a lot of claims that sound reasonable, but doesn't provide data to back it up.

For example, in the first 30 seconds, he says that at the beginning of the Iran War, AI was used to strike 900 targets in 12 hours. Which he calls "unprecedented" but then never backs up.

For context, in the 1991 Gulf War "Operation Desert Storm" the U.S. struck about 1500 targets[1][2][3] in 12 hours.

[1] https://www.mitchellaerospacepower.org/app/uploads/2021/02/a... [2] https://www.youtube.com/watch?v=zxRgfBXn6Mg&t=401s [3] https://gulfwar.org/gulf-war-1991-timeline-desert-storm/ -

And gps guided missiles were doing that since the 80s. Humans are already really good at killing each other. Yeah it sucks the tech will be used for that.

But it changes little.

iirc consumer grade GPS chips purposely become less accurate if they find themselves moving at high speed.

Drones do not need to mov at high speed to be effective, as cam feeds from FPVs in the Russo-Ukrainian war have demonstrated

You're slightly off the mark here.

They are NOT "em-proof" --- what they are is electronic warfare immune.

https://www.newscientist.com/article/2529849-fully-autonomou... Published this year, but talking about a trial 2 years ago.

Blocking any leading edge AI model changes nothing. We (humans) have a long history of determined attackers finding creative and unexpected solutions.

What the AI we have, the stuff that is already PUBLICLY AVAILABLE, is good enough to shrink the time for developing one of those creative solutions into a working tool/weapon.

Shh...you'll burst the bubble of the folks who think that LLMs are toy stochastic parrots...

Can you share any of these serious thinkers?

Right now it's basically this easy: 1. apex domain 2. ???? 3. critical PII exposure

There is /so/ much stuff on the internet that just needed someone to spend enough time on it.

Nobody with even a modicum of understanding of how LLMs work believes any of this. These 'serious thinkers' are just grifters preying on the feeble minded.

Hinton and Bengio don't understand how LLMs work?

They do, but they're not above grifting. Clearly people like you are their mark.