Can someone more knowledgeable than me help me understand a few points about this article ?
It claims to be diffusion-based, but the main 2 differences from an approach like Stable-Diffusion is that (1) they only consider a single step, instead of a traditional 1000 and (2) they directly predict the value z^y instead of a noise direction. According to their analyses, both of these differences help in the studied tasks. However, isn't that how supervised learning has always worked ? Aside from having a larger model, this isn't very different from "traditional" depth estimation that don't claim anything to do with diffusion.
It also claims to have zero-shot abilities, but they fine-tune the denoising model f_theta on a concatenation of the latent image and apply a loss using the latent label. So their evaluation dataset may be out-of-distribution, but I don't understand how that's zero-shot. Asking ChatGPT to output a depth estimation of a given image would be zero-shot because it hasn't been trained to do that (to my knowledge).
I've recently had to implement a few kernels to lower the memory footprint and runtime of some pytorch function : it's been really nice because numba kernels have type hints support (as opposed to raw cupy kernels).
I use it as my main YouTube app on my phone, it's working nice.
One drawback for me is the failure if you have too many YouTube subscriptions (>200-300). I removed some creators and it's working fine. You see Grayjay's likes and comments on videos but there are very few of them (you also see YouTube comments and likes). I wish the community around it grew, it has potential.
Ins't it still possible to construct the zero value for Color ? Like how bytes.Buffer does it https://pkg.go.dev/bytes#Buffer : the implementation is not exported but you the usage is to construct the zero value and use it.
So unless Color{} is a valid enum value, this solution would not work.
Not the OP making the claim, but I also looked for recent data and couldn't find any (especially for M1 macs because I just switched to one).
So I just compared them myself for my typical work and media playback (Youtube & Twitch), just anecdotal and totally unscientific though, ranked by MacOs's Energy Impact with the best being 100%.
All with the same extensions (if available, for Safari I used adguard instead of ublock origin).
Safari Technlogy Preview: 100%
Brave Beta: 110%
Orion: 300%
Firefox: 1200% :(
While Safari also used slightly less memory, Brave is the best compromise for me for now. (e2ee sync, chrome extensions and with their crypto shenanigans disabled.)
But again, not scientific at all and of course might differ for people depending on the extensions they use and their browsing behavior.
The number for Orion looks wrong (Orion dev here). It should be same or better than Safari as both are using WebKit with Orion being more lightweight. Would be happy to look into this if you can post methodology used at https://orionfeedback.org
I'd expect battery life to be similar (or maybe even better):
> By using your computer’s built-in website rendering engine (the same one Safari uses), and by blocking trackers before they load (unlike all the major browsers), you’ll get really fast browsing.
While I don't doubt the goal is cloud, it's not the only use.
As a student, I have a desktop at one of my parents' house that I can control over ssh, this kind of features make remote development much easier and is often needed when I run an intensive task for hours. The experience with VSCode over ssh is really great. Some have pointed out local VMs, which is another use for this.
> But please, give me a way to pay for it. I don't want to be the product, one day.
From what I understand, this is already planned.
> Brave Search is currently not displaying ads, but the free version of Brave Search will soon be ad-supported. Brave Search will also offer an ad-free Premium version in the near future.
It claims to be diffusion-based, but the main 2 differences from an approach like Stable-Diffusion is that (1) they only consider a single step, instead of a traditional 1000 and (2) they directly predict the value z^y instead of a noise direction. According to their analyses, both of these differences help in the studied tasks. However, isn't that how supervised learning has always worked ? Aside from having a larger model, this isn't very different from "traditional" depth estimation that don't claim anything to do with diffusion.
It also claims to have zero-shot abilities, but they fine-tune the denoising model f_theta on a concatenation of the latent image and apply a loss using the latent label. So their evaluation dataset may be out-of-distribution, but I don't understand how that's zero-shot. Asking ChatGPT to output a depth estimation of a given image would be zero-shot because it hasn't been trained to do that (to my knowledge).