aruss

It seems like the more honest comparison would be to OCR the screen and send that as input to the LLM?

Tame provocation is an oxymoron in my book :)

I can’t say that I expect any profit from this. (I’d be shocked if anyone spent money on it, and I lose money every time someone uses the free credits).

It’s an art project - a provocation. Beyond a PoC is flattery it truly doesn’t deserve imo

This is one class to deeply understand one of the most politically impactful texts in history. What about that is religious brainwashing? Do you even know the professor’s religious beliefs? Because they’re not obvious to me simply from the course description.

I took a religious studies course in college amid a liberal arts math/CS major. I can say it was one of my favorite classes in college, primarily for exposing me to ways of thinking that differ from my own, and yet I left with my (lack of) religious beliefs intact.

Yes, agreed that it had a lot of limitations as a proof of concept! It’s only a 70B model after all. I could barely give it a personality.

The next iteration will use frontier-level intelligence and have more capabilities and context.

Thanks! Could a mod edit this for me? Otherwise I’ll resubmit.

Honestly the seinfeld reference just amused me at the beginning and then kinda stuck, nothing more!

This changed in 2020! Indistinguishability Obfuscation from Well-Founded Assumptions: https://eprint.iacr.org/2020/1003.pdf

You're right that this isn't about data collection/privacy, though it would be a bit scary to have our nation's cars remotely shut off at the whims of a foreign government.

Instead, we want/need to be protectionist of our manufacturing industries so that if we were to go to war, we keep our ability to make more missiles, planes, tanks, etc.

Yes, there are ways to speed up your time. When I took the SAT (2011), the main skill I developed was to quickly categorize the kind of question. Since there are only a few different kinds of questions, you could train yourself to do this formulaically/algorithmically. The type of question tells you how to eliminate a couple of the answer choices, and then you apply another question-type-specific algorithm to figure out the answer from the remaining options.

Then I would say your theoretical model is wrong or incomplete or makes for a circular argument (it's an assumption and not proven that finite matter evolving through time reduces to a lookup table).

Where are you going to get all that time and space to build a lookup table? Are you sure you're able to measure all state at enough precision to make an accurate table?

Why should we lower the bar on AI models to your superficial understanding of Indian culture?

The problem with this specific instance is that the images generated mix and match characteristics that are unique to those ethnic groups. So the models are reducing real ethnic differences to simple stereotypes. In other words, the models are wrong, and wrong in ways that eliminate diversity.

Dunning-Kruger begs to differ...

The title is a Kierkegaard quote :)

I like that their pieces dig into the backgrounds of the researchers, which humanizes and popularizes the science itself. It has to get down from the ivory tower somehow, and storytelling is a great medium. Most Quanta articles are not "void of any science;" rather, I find them to be great lay explanations of high-level concepts that most scientists (including myself) would struggle to articulate to non-academic family and friends. If you want "just the science, ma'am," you can go read the papers themselves.

Also, not to nitpick, but I thought the crypto article was well-written and covered the subject properly. It seems you are the one misunderstanding the science if you believe that iO is just a form of FHE, which makes me doubt your whole enterprise of discrediting Quanta's work ;) Speaking as a (former) cryptographer here.

I'm a former cryptographer and the headline and news story aren't sensationalist by any means. iO has been considered the "Crown Jewel" of cryptography (by cryptographers) ever since the 2013 iO candidate by Sahai, Waters, Gentry, et al, and Sahai/Waters' followup "How to Use Indistinguishability Obfuscation." From a cryptographer's perspective, being able to use iO is like having a superpower.

If you're interested, check out the paper called "How to Use Indistinguishability Obfuscation" by Sahai and Waters. They were able to prove some very surprising things -- including how to hide a private key by way of iO.

iO isn't going to be used in video games for a long time, if ever. For one, even if it had no overhead, it'd completely ruin any optimizations that the devs needed for the game to run in the real world. But likely iO will add a huge overhead, degrading performance by at least a quadratic factor or more. iO also doesn't consider things like syscalls or hardware instructions, which might break the security model anyway.

Video/music DRM wouldn't (couldn't?) change from what it is now: the plaintext bits have to be surfaced to the user at some level (so that it can be displayed/played), so they can simply be ripped at that point.

FHE can be made IND-CCA secure in practice (i.e. for specific systems), please stop spreading misinformation. It's impossible in the general case for obvious reasons. Also, TFA isn't about FHE, it's about iO.

I know and have worked with the organizers of RWC and they've done a lot of the kind of work that you're dismissing here (which would be like telling Diffie and Hellman to piss off in the 70s because public-key crypto isn't IND-CCA2 secure without additional assumptions).

You can encrypt a function with FHE, but the applications of that are limited because the output is encrypted as well. We've had provably secure FHE since Gentry's 2009 paper.

Obfuscation roughly means "encrypting a function." But here be dragons: since the adversary gets non-black-box access to the implementation of the function (a circuit), and can also learn as many (input, output = f(input)) pairs as they want, it becomes impossible to say that the adversary cannot learn anything else about f in the general case (intuitively, consider a higher-order function f: you could learn a lot about f by computing f(f)).

iO doesn't guarantee that the adversary can't figure out what "key" is. It turns out that this is possible, but it requires some very careful applications of a special kind of pseudorandom generator.

Former crypto PhD student here. Here's the formal definition. FHE means that, given an encryption Enc(x) of plaintext x, you can compute Enc(f(x)). Note that without the decryption key, Enc(f(x)) is just as "useless" as Enc(x) (i.e., it looks totally random). FHE would be useful if you wanted Google to filter all of your emails for spam, but didn't want them reading your emails. Google would receive Enc(email) and compute Enc(isSpam(email)) and then forward Enc(email) and Enc(isSpam(email)) to your email client, which would then decrypt both and either send `email` to spam or not. Note that in this scenario, Google learns nothing about `email`, not even if it is spam or not.

Here's what iO means: given two different circuits C_0 and C_1 that both compute a function f, the obfuscations iO(C_0) and iO(C_1) are indistinguishable (to a polytime adversary). In the email example, you could have a circuit C that computes isSpam over encrypted emails, but with a decrypted result. That is, C(Enc(email)) = isSpam(email). But note that this leaks information to Google, and the security definition of iO does NOT guarantee that Google does not learn anything about your secret key (the obfuscation might not hide the secret key at all).

Definitionally, they're very, very different things and have very different security guarantees. The weird thing about iO is that the security definition doesn't immediately appear to actually secure much, so it doesn't seem very useful. However, it turns out that iO is an incredibly powerful primitive because it can be combined with other things like pseudorandom generators to build up a lot of other primitives. For example, you can combine secret-key encryption and iO to get public-key encryption in a very elegant way.

Category theory isn't a type theory (in the sense that you can have the category of categories, which correspond to different type theories).

It's true that you can get away from Russell's paradox in category theory, but that's because categories are not sets and have different properties (there's less you can say about them because they're more general).

Type theories, particularly homotopy type theory, are becoming popular among some mathematicians, but not because they escape Russell's paradox (I'm not sure that they can say anything more about sets than set theory), but because they are more useful when constructing automated proof systems.

Edit: also, category theory is used a lot in very cutting edge math, like algebraic topology.

GCD(0,0) exists in any commutative ring by definition (and is not unique) :)

d is a common divisor of a,b if there exists x,y such that dx = ay, and d is a GCD of a,b if all divisors c divide d. So there exist many such x where GCD(0,0) = x (including x = 0).

The number of post-hoc internal memos and studies clearly shows that there are many consequences that FB engineers did not anticipate (though FB is not unique in this regard; consider how YouTube's recommendation algorithm feeds misinformation & outrage).

The whole point of TFA is that pure intelligence (particularly "computational thinking" type of intelligence) is no substitute for the knowledge and methodology already explored by various humanities and social science fields.

Isn't this just being pedantic? How is TFA's point changed by substituting "social sciences and other humanities" for "social science"?

Definitely, social sciences have gotten a lot of things wrong. And the pipeline from science (social or otherwise) to policy is pretty broken, to put it mildly. Isn't it all that much more important that computer science absorbs that history, so as not to make the same missteps?

Historically that may have been true; I won't argue that point.

The difference between the atom bomb and CS is that the latter has a ubiquity of access and reproducibility that subject it to completely different ethical and philosophical concerns. The atom bomb doesn't change or influence our understanding of how political discourse operates in modern democracies. Things like social media and GPT-3 do.