Sabitlenmiş Tweet
David Wong
36K posts

David Wong
@cryptodavidw
security @zksecurityXYZ & AI @zkaol, author of Real-World Cryptography, prev: architect @Mina, sec lead Libra (@Facebook), crypto @NCCGroup
New York Katılım Haziran 2011
3K Takip Edilen17.9K Takipçiler

@dankrad are you saying everyone has a different definition? I would have thought that there are some common definition people agree on
English

When you use it for a while on a project, it is clear that even Fable misses things that are obvious to you and you need to drive it in the right direction. Not clear to me what it is, but this has so far I have always felt this with new models after the first impression wears off.
Don't get me wrong, it already has superhuman performance for a lot of tasks, but it still feels like there is something missing, although it is hard to pinpoint what exactly. Something happens in the human brain as you get deep into a project and get a good understanding, that models just don't as far as I can observe.
Will be interesting if (and when) that gap closes.
English

@cryptodavidw Making your politics your identity does not produce interesting content. Mostly predictable biased advocacy.
English

Ok we have to agree that Meta might have gotten back in the LLM game
Big Tech Alert@BigTechAlert
🚫 @elonmusk is no longer following @alexandr_wang
English

@cryptodavidw I don’t think ZK inference is THAT important.
English

@ZamDmytro I like the fact that it's a nice benchmark that teams can work towards to improve ZK, there's huge room for improvement there. Also, I love the idea of self-willed NFTs :P cryptologie.net/posts/whats-ne…
English

I have been working in zkAI field for roughly a year, and to be honest it still feels like "let us put ZK to yet another thing" to me.
In principle, you have two modes:
- model is public, inputs are private. This is imho the most interesting case. For example, you can make biometric model public and try to make some "zk biometrics login" stuff. However, once the model is public, you have loads of different problems that you cannot tackle. For example, one might easily do adversarial attacks when weights are known. And in general neural networks cannot be treated as one-way functions. And problems just go on. If you come up with some interesting use-case, that would be awesome
- model is private, inputs are public. I think this just kills the idea of ZK because (a) you do not hide users' inputs (like, why do we need ZK in the first place then?) and (b) some centralized entity must store this model. So the only "OK"ish use-case is verification of LLM outputs, but let's be honest: it's unlikely the prover's efficiency will be adequate for the modern scale of architectures
English
David Wong retweetledi

love that RH is talking more and more about crypto
i) helps onboard net new capital
ii) helps Solana by giving us credible competition to sharpen against
this will be a ton of fun
(I will die before I let another dude named vlad stand against the ottomans)
Robinhood@RobinhoodApp
We're in a very crypto time of our lives.
English
David Wong retweetledi

@WhaleInsider @Adam_Tehc Couldn't you have used a more recent picture for such big news?
English

@cryptodavidw By the way, we have quite down-to-earth explanations how zkML works, so feel free to read the rest of the sections
English

I really didn't realize the field of zk AI is so big github.com/mimoo/awesome-…
English
David Wong retweetledi











