Is SpaceX Coming For Your Phone Bill? | The Brainstorm 153

Brett (0:00)

If the world is gonna get hacked, it's actually more likely to get hacked by open weight models than closed.

Nick (0:05)

What about quantum encryption? Is that crackable with AI?

Sam (0:10)

All right, SpaceX coming for the jugular of AT&T, T-Mobile and Verizon, picking up Nextel's old spectrum, and according to Musk, the last piece of the puzzle for Direct2Cell, and then OpenAI absolutely demolishing mathematicians worldwide with a massive release of proofs, 722 proofs out there. But I think the more interesting question that we'll dive into is whether or not this is a risk for crypto and how this is kind of butting up against the fear for quantum. Brett, maybe we'll start there. And in all of these proofs, one, how significant do you think they were? Two, notably nothing on cryptography. Is that because if they release it into the wild, it is too dangerous and everyone's Bitcoin wallets then immediately go to zero?

Or is there something else at play here?

Brett (1:12)

I think there's one, it's definitely a demonstration of capability in that these are the sorts of proofs that working mathematicians work towards as multi-year projects that they then publish on and get recognized for. So the mathematical community is some excited, some very upset that OpenAI is speed running all of mathematics and seems to be able to do so with the people they have on staff and some number of hours of compute time with their super powerful models.

And I think that everybody noticed that across all of these proofs, many of which are, hey, these are interesting and could lead us to other kind of discoveries in formal logic, some of which have potential commercial applicability. There is this notable lack of advancement in cryptography. One way to interpret that is that they recognize these models are very powerful and probably in collaboration with the US government have said, hey, we're going to withhold any advances we have in this particular area so that we, the US and US enterprises can harden our defenses against a crack in cryptography. Another interpretation would be, well, there's so much money at stake with cryptographic architecture, that these are areas that there's already been commercial incentive for people to go after and try to probe and try to advance the frontier. And so, it's not as easy pickings for an AI model to unpack it. And so, I think it's, you don't necessarily know which is the truth. I do think it's accurate to assume that formal logical mathematics is going to just fall to this super intelligent, call it beast, that we're building. And that does at some point begin to put some of these advanced cryptographic techniques under risk. And I would assert, actually, there's greater risk from AI than there is from quantum computing, to kind of like certainly penetrating cryptocurrencies broadly.

But the risk is not necessarily in the cryptographic primitives, but in everything else that tries to secure your blockchain, and that all these organizations have multiple risk vectors. And this is just an indication of power and how risky it all could get.

Nick (3:46)

Isn't the risk actually that it's not open AI that releases something that is able to crack cryptography, but that the rest of the field catches up to where open AI is and the frontier is in 6 to 12 months?

Brett (4:02)

Yeah, I mean, I think that you can see what is, there is what's publicly available now, which across Anthropic and Open AI, you can pay nosebleed prices for best in the world intelligence. Then you know that behind the scenes, they have something more powerful that they're able to work on, and that they're driving forward to uncover these vulnerabilities, and then you have what you could run in your own Cloud implementation, and do whatever you want with without anybody being able to stop you, and that trails the frontier by six or 12 months, and I think it's reasonable to assert that maybe the industry as a whole lags the frontier labs, maybe it catches up, I would tend towards lags, but this capability will be available.

The OpenAI and Anthropics approach is one, we're going to be careful about what we release to the public, because we're going to selectively provide infrastructure providers with this protection service, so that when the nefarious actors have access to this intelligence, we have a more hardened infrastructure. And yes, I think if the world is going to get hacked, it's actually more likely to get hacked by open-weight models than closed, because those will be where the nefarious actors will operate.

Sam (5:28)

And great, as we said in previous episodes, we're looking for the Frontier Labs to step in and say, well, great, here's our cybersecurity package pay up.