the etiology of this hack is pretty obvious with a simple google search:

https://docs.nvidia.com/cutlass/index.html

it presumably makes various assumptions and speedups for NVIDIA's matrix multiplication library... called cutlass

Not many know this but Goldman indeed has between 2,000 and 3,000 MD physicians in their medical staff. Other point Re: "VPs" : tech director = finance VP tech manager = finance director tech IC lead = finance manager tech bathroom janitor = finance IT

MD is managing director, not medical doctor.

of note AndrewFS (think cousin of NFS).. with features pertaining to systems being "offline" and "online" alternately, but still accessing the same global file / directory namespace, still reading and writing, so some type of git or cvs style merging also exists in that filesystem

kim dotcom ran basically a pirated game/book/music/movie site. Telegram (what I have seen) is mostly hacking leaks although rumored to have CSAM (to those not familiar with the acronym, it means cheese pizza).

Of course you can find both somewhere in the walled planetary garden of googlotron in facebooks sure. But they clamp down on it hard as they can. They clamp down on anything marginally offensive much less illegal. Have you tried the facebook "report post" interface? there are 3,000 various types of offensiveness you may report. That's their bar, their standard, that's 1,000 miles away from definitely illegal content. If their censorship apparatik is so bold as to be wiping out vast swaths of totally valid free speech, anything illegal has no chance.

If the question is "what's a great place to go for piracy" - megaupload, pirate bay, etc, then any common answer to that question is a target... "where to I go for data breaches" - breachforums, telegram, etc. Don't get worked up, all those places were destroyed by the feds and no longer exist.

There are a set of words you can't utter on these sites or your content gets removed. That's not moderation, that's censorship.

And they love that it's not actual moderation.

could the generator and challenger be cross trained against each other, so as to actually both get better? like a generative-challenger network (GCN) or something like this?

I think this is how RLHF actually works. https://huyenchip.com/2023/05/02/rlhf.html#3_1_reward_model

All right now, here's the big question: how do you know that the evidence your sensory apparatus reveals to you is correct? What I'm getting at is this: the only experience that is directly available to you is your sensory data. And this sensory data is merely a stream of electrical impulses which stimulate your computing center. In other words, all that I really know about the outside universe is relayed to me through my electrical connections.

          Why, that would mean that...I really don't
          know what the outside universe is like at
          all, for certain.

I see it more simplistically. If there's two guys love neutrinos, neutrinos is they passion, one happens to be born in Russia, one in US (neither one chose to be born there), and they want to work together to understand life, the universe, everything, then so what? What matters more, life, the universe, everything, or some years-long dramatic squabble, years long on the scale of a 5 billion year old planet and millions year old species and perhaps tens of thousands of years old civilization.

The "ideal" of science is that there should be this purity, this honesty, etc. Deviations do occur, politics are common in science itself, data falsification, etc, but we must not let that deny the purity ideal. In the same way, I believe the idea here is to acknowledge that purity over political differences as something transcendent to politics. But yes, such collaborations do have political value in terms of being a "relationship". And yes of course I don't want to poo-poo politics. Whereas it's true that politics is as inescapable to the human condition as say language, perhaps we shouldn't let political constraints guide and dictate scientific work just as we shouldn't let linguistic barriers limit science, both part of the human condition, both miniscule in the face of the work being done.

truly the darkest side of these is the opportunity to exploit if someone shares a document with the internet with read/write access, can I upload smut (or anything very naughty or illegal) onto their google doc from some anon connection/account? Can I get them banned thusly? Can I log into their wifi and post threats on whatever renowned toxic boards and again, get a person in trouble?

The problem with the right-think industrial complex is the heavy hand itself can be fooled easily and thus weaponized.

>Very glad to see GANs (or GAN-likes) coming back! Adobe has had GigaGAN out a while now though?

your point is valid but the paper explains it clearly and obviously. they are NOT dimensionally reduced hyperparameters, no. The hyperparameters are learning rates, that's it. X axis, learning rate for input (1 hidden layer). Y axis, learning rate for output layer.

So what this is saying, for certain ill-chosen learning weights, model convergence is for lack of a better word, chaotic and unstable.

Just to add to this, only the two learning rates are changed, everything else including initialization and data is fixed. From the paper:

Training consists of 500 (sometimes 1000) iterations of full batch steepest gradient descent. Training is performed for a 2d grid of η0 and η1 hyperparameter values, with all other hyperparameters held fixed (including network initialization and training data).