HN user
Vitalik recently wrote about vibe-coding in Lean and assembly language, using proof verification in Lean, saying "if done right, this has potential to both output extremely efficient code, and be far more secure than the way programming has been done before."
The article says Kimi is better at some things than Fable. That's probably not true of Sonnet.
It'd be interesting to see how that works in various languages, because there may be some selection bias. Most English-speaking countries have decent protections for all but the most abhorrent speech, and people who don't find their speech constrained may not bother with anonymous platforms. In non-democratic countries with more restrictions on political speech, maybe we'd find anonymous platforms full of pro-democracy dissidents.
Maybe something that would help these platforms is some kind of distributed or federated moderation, so you can avoid the icky stuff yourself even if it's still out there.
Yep at this point we just have to do whatever seems least risky. It's not like we have a pristine planet. We have to balance the unknown consequences of phytoplankton (and other mitigations) against the known and unknown consequences of a >50% higher CO2 level and >1.4C higher global temperature.
768GB next quarter. The chip supporting 1.5TB won't come out until 2028 or 2029.
What do you mean by "clean-rooming your prompts?"
If the government provided funding to independent organizations that made public-goods AIs, that could be great, as long as the government had no editorial control.
It sounds like he's imagining AIs only being trained and provided by governments, which could get pretty dystopian.
As a biological human I am glad that Google is applying AI to biodefense.
But if Anthropic had just bought the authors' books instead of pirating them, then they would have paid nothing.
Anthropic paid a large settlement for the copyrighted data they pirated. So far, US courts have found that it's perfectly fine to train AIs on copyrighted data for which you have legal access.
A lot of mortgages were trash but the securities were a new and complicated financial product with a plausible story. A few people figured out the problems, but Michael Burry for example did it by combing through thousands of pages of prospectuses.
SpaceX bonds are just plain ol' corporate bonds, the same stuff these investors have been analyzing for a very long time.
Some people actually do track their cars, though.
I'm old enough to remember they weren't just buying and selling corporate bonds like most of the bond market. They were a hedge fund that used massive leverage to exploit tiny arbitrage opportunities between correlated securities.
There may be important factors that the bond market isn't aware of, just like most of them didn't know about the problems with mortgage-backed securities in 2008. But anything you and I are aware of, they probably are too.
Institutional bond traders are pretty sophisticated. They're aware of all the possibilities you mentioned, and pricing the bonds low anyway.
At this point I can only ask that you actually read my above comments more closely and think about them. I have exactly the level of security I described in my last comment, with complete anonymity, all with one design. If your claim is that this level of security is insufficient, my answer is that it's more secure than anything actually being used for age verification, by a significant margin.
The reason we keep switching back and forth is that every time you make objection A, and I explain how the system addresses that, you switch to objection B, ignoring that I already explained how the system addresses that as well.
The only requirement I dropped is absolute perfect security, which no system will have and certainly not the nonsense "show your ID to the camera" stuff they're actually using.
I showed how to eliminate proxies, and without that, the only hole is the credentials themselves getting passed around. If that happens at a small scale between people who know each other, it's a minor problem. If a credential gets posted online and widely shared, it's easy to notice and revoke it.
How? The websites don't see anything consistent from one session to the next. They can't tell when the credential is renewed, and don't see a credential id.
You can fix the leaked token problem if your prover also proves that (a) the private token id is not on the public revocation list, and (b) the token has not yet expired. Use short expirations and auto-renew, this is just to keep the revocation list from growing forever.
Attackers could still compromise the system with proxies, but you can fix that by (a) passing in a random sessionid from the server so proofs can't be replayed, (b) also passing in the server's public key, so a MITM attack will result in proof the server can't verify, and (c) as you mention, using secure hardware on the client, and encrypting communications between that hardware and the server. The secure hardware doesn't have to preclude general-purpose computers; it can work like a yubikey or hardware wallet, just plug into USB or bluetooth.
Without proxies, a leaked key has a minor impact unless it's widely distributed online, in which case it's easy to notice and add to the revocation list.
Tracking clients can be prevented if the client generates a new public key for each session.
Requiring hardware is in one sense a downside, and strong protections for access would have to be part of the law. But giving everyone secure cryptographic hardware that can do key management and zero-knowledge proofs would be a huge improvement for everyone's privacy and security, so it might be a good trade.
Ideally, but in practice only if the bits are uncorrelated and each divides the population in half.
But there are a couple in production that are topping the benchmarks.
For the optimistic case, read this piece by RethinkX:
https://www.rethinkx.com/blog/rethinkx/the-disruption-of-lab...
(Fwiw, >20 years ago RethinkX correctly projected the exponential cost declines of solar and batteries, when everybody else was drawing straight lines.)
Since they've reduced the max RAM on current Studios from 512GB to 96GB, I'm not holding my breath.
I already described in detail how to do revocations despite the website not being able to identify the device.
The whole point of zero knowledge is that you can prove you have possession of information with particular properties, without having to actually reveal the information. I explained how to use this in my first reply to you. You're just repeating your claim without bothering to understand the first thing about the technology I'm talking about.
Age verification itself is a change to regulations. Obviously to make a private system, our regulations would have to change in a way that allows that, instead of in a way that destroys all privacy as they're attempting now.
Yeah, but it's everything getting fuzzed, including the parts you care about.
Well it would be anonymous because the client public key is randomly generated every time, and the zk proof doesn't include the identifier.
I don't think this is about becoming a great musician. Maybe that comes later. As the title suggests, this is just about becoming a musician in the first place.
As a piano player who's been noodling around for a few years, trying to learn to write original stuff and not making much progress, something like this is probably what I need.
Thinking about this further: it looks to me like we can fully prevent proxy attacks if we assume secure hardware handling the cryptography on the client side, and send an encrypted sessionid to the client after verification.
The reason is that:
- We send a random number from the server at the beginning which is associated with the session, so a proxy can't just replay the encrypted proof.
- We pass the server key into the proof, so a MITM attack would give the wrong server key to the client, and the real server wouldn't be able to verify the proof.
Then with each request the secure hardware can provide encrypt(sessionid, nonce).
I'm not sure what my "attitude" is but I'm being pragmatic. This is not a binary situation, where it's either perfectly secure or useless. If our society is not willing to do what you and I prefer and leave things entirely open, then perhaps it's good enough to make things more difficult for teens to access, rather than accept pervasive surveillance to make it impossible. If people think it will improve society enough if most teens stay off certain sites, then we can do that and maintain anonymity.
I'll note that you skipped over my point that even with a "perfect" system, teens could still pay foreign porn sites etc directly. And that using a proxy would require installing an untrusted app on the phone, which would be relatively easy for parents to monitor and could be prevented entirely on iPhone. And that we can probably fix proxies with secure hardware anyway.
And no, the police idea that I do not support would not require surprise inspections. It just requires careless teenagers to occasionally reveal their identities online, with enough evidence to convince a judge to issue a warrant. It's dumb to make a federal case out of this, but not as dumb as losing all privacy and anonymity online. And, as I mentioned, this is not something actually required to make the idea workable.
I'm not going to keep repeating myself so I think I'm done here unless you have a point I haven't addressed in previous comments.
So as I've mentioned elsewhere, that depends on how much of a stickler we insist on being.
If we're ok with "mostly fix it but if a few teenagers get through it's not the end of the world," then there are a few simple measures that could help a lot:
- Keep an eye out for any credentials posted online, and put those on the revocation list.
- Keep expirations short (and auto-renew).
- Keep the credentials in phone secure enclaves and USB hardware "wallets."
- Consider including private information like name/dob/ssn or credit card number in the credentials, so users have good reason not to share. (We could consider making USB hardware optional if we do this.)
Given secure hardware it might be possible to prevent proxies entirely, the same way we prevent other MITM attacks.
Failing that, we could start by making it illegal to run proxies. Installing a proxy on your phone would mean getting an app from a criminal, not checked by an app store, and giving the criminal a way to pay you. I wouldn't expect this to happen much. Installing on a computer, using a VPN, taking payment via anonymous cryptocurrency, sure, if the VPN isn't compromised. But I wouldn't expect all that many people to do all this. Generating the proofs is a bit expensive so you wouldn't have huge capacity per person.
Criminals in foreign countries could do it with stolen credentials, and they'd only need one. But our teenagers would have to pay a foreign company for the service, and for porn at least they could just pay a foreign porn site directly. For phones, the teenager would have to install an app to use the proxy, which is another dodgy untrusted app (on android, and not possible at all on iphone), and it's easier for parents to check what apps are on the phone than to check what websites the kid visits. And social media gets less appealing if a lot of your friends aren't on it.
If we want to lock things down harder we could go with criminal penalties for intentionally sharing your credentials, which I do not support, but would still be better than pervasive surveillance of everything we do online.
Requiring everyone to have secure cryptographic hardware would in one sense be annoying, but less so if we use it for other things too.