Viewing profile — grungegun

In a sense, I spend most of my time on material after quals because it's so much harder to understand.

So there's no performance gain for quantization enabled by the transformer architecture? It seems very strange that quantization works so well since in most of my experiments, the …

Does anyone know if this works on vanilla deep networks? These quantization articles always seem to target LLM's which leads me to wonder if there's something special about the LLM…

I'd like to allow customers to submit orders for custom playing cards with both sides printed on. Are there any services that provide an API for this since I don't have printers?

I taught one of my younger brothers a simple sign language when I was younger. His speech was delayed by like 8 months compared to the rest of us. Can't say for sure, but there's a…

This is somewhat overblown, and only really an issue for some particular theorems and obscure corners of mathematics. And this isn't a recent problem, Hilbert's proof of Nullenscht…

the infinite density is actually after a finite amount of time, since black hole collapse has already happened, which is why it matters that they are predicted to have infinitely d…

would be interesting if you could used deepfloyd first for image composition, then apply stable diffusion after for purely stylistic modifications

What does this mean? Isn't the quality of a model determined by how easy it is to get a good picture?

Loved this book. When I first started programing, I was convinced that text applications weren't real programs, and the visual aspect of the book pulled me in to be a life-time pro…

For diversity, Dalle 2 has a random chance of injecting "women" or "black" after a prompt. When this happens, at least for me, it generally destroyed the quality of the images. Pro…

I'd argue that the incredibly fast rise in quality over the last year alone is what most people are interested in. DALLE and GPT-2 were always able to make heaps of trash. The traj…

yep, that sounds right to me. also I see you're the author of the Noether article from Ars Technica. I enjoyed the read.

Yes, the quantumness of a differential equation is not a property of the differential equation, but a statement about one possible taxonomy of differential equation. Then, whether …

> Differential equations don’t give rise to physical systems. Sure, but differential equations describe physical systems, and there is a canonical way to derive a quantum different…

The differential equations which give rise to physical chaotic systems operate on quantum states interacting with each other at the atomic level, so there has to be a link of some …

It looks like in the results section eqs (8)-(13), they give the main thrust of their idea. To demonstrate that not all machines are reversible they have to take the limit of the n…

Actually, it doesn't increase much faster than linear. It's quasilinear, and so in O(n^(1+\epsilon) for all \epsilon > 0. There are a variety of such results, so most objections do…

False, except for a naive implementation, see: https://en.wikipedia.org/wiki/Quantum_threshold_theorem If you look at the theorem, it is definitely sub-exponential in the number of…

For reference, see the book High Dimensional Probability by Vershynin. It's free online. See Theorem 3.1.1. It proves that a sub-gaussian random vector is in some sense close in no…

>It didn't have very many interesting properties I could find Usually notation is defined for higher order functionals when the need arises - when a new result is found to have int…

The doctors don't believe the story. The woman does. The doctors believe that the woman's belief caused her early death, but that's independent of whether they accept that story as…

Is there a reference for this to verify that this isn't just some historical reenactment? I'm trying to be more careful after I was once pranked by a friend who showed me a black a…

For those interested, there is a fairly basic WYSIWYG editor for this: http://www.tikzedt.org/ It's useful to throw together basic 2d diagrams, which can then be refined later

I'm using cryptomator for file encryption right now. I'd like to use a small, portable, C-based (for familiarity and guaranteed future support), open source, command-line/terminal …