# New AI Makes Everybody Dance!

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=AnCsmHrMPy0
- **Дата:** 07.01.2024
- **Длительность:** 7:42
- **Просмотры:** 231,803

## Описание

❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambdalabs.com/papers

📝 The papers are available here:
https://dreamoving.github.io/dreamoving/
https://www.science.org/doi/10.1126/science.ade4401
https://www.biorxiv.org/content/10.1101/2022.09.01.504602v4

Super res: https://blogs.nvidia.com/blog/rtx-video-super-resolution-ai-obs-broadcast/

📝 My latest paper on simulations that look almost like reality is available for free here:
https://rdcu.be/cWPfD 

Or this is the orig. Nature Physics link with clickable citations:
https://www.nature.com/articles/s41567-022-01788-5

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Alex Balfanz, Alex Haro, B Shang, Benji Rabhan, Bret Brizzee, Gaston Ingaramo, Gordon Child, Jace O'Brien, Jie Yu, John Le, Kyle Davis, Lukas Biewald, Martin, Michael Albrecht, Michael Tedder, Owen Skarpness, Richard Putra Iskandar, Richard Sundvall, Taras Bobrovytsky, Ted Johnson, Thomas Krcmar, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: https://www.patreon.com/TwoMinutePapers

Thumbnail background design: Felícia Zsolnai-Fehér - http://felicia.hu
Károly Zsolnai-Fehér's research works: https://cg.tuwien.ac.at/~zsolnai/
Twitter: https://twitter.com/twominutepapers

## Содержание

### [0:00](https://www.youtube.com/watch?v=AnCsmHrMPy0) Intro

today we are going to create imaginary characters and make them dance and we'll use an AI to greatly improve the details in lowquality streamed gaming videos and we will see if it is possible for a robot to learn to smell spoiler alert there is a new AI that has no nose and yet it can still smell kind of now this

### [0:25](https://www.youtube.com/watch?v=AnCsmHrMPy0&t=25s) New AI

is an AI paper from 2016 this is what was possible 8 years ago you could take a source video of an amazing dancer and take a video of yourself moving about and then it would transfer these amazing dance moves to you but today not only text to image but text to video is also getting more and more available this means that you can enter a piece of text and out comes a video of exactly that what's more today you can do this with free and open- Source techniques but if that is possible does that mean that yes it does so now hold on to your papers fellow Scholars because today we have come so far you don't even need a video of yourself just a little stick figure moving about and one text prompt to synthesize the video and then a character of you will start dancing wow look at how far we have come in the last few years these new results are still not perfect but they are Leaps and Bounds beyond the Everybody Dance Now paper and this is not only better but it also solves a significantly more difficult problem you see here with your text prompt you can choose your background and your attire As You Wish no video of you is needed but it gets better if you can use a text prompt you can write whatever you want there even the Moa Lisa or any other character you can think about a couple years ago we were happy if we got a text to image result with this and now we can just make them up and make them move exactly the way we want it to absolutely incredible dear fellow Scholars this is 2minute papers with Dr carool now I will quickly note that this paper is a tech report and as of the making of this video it was not peer-reviewed yet it just came out a few days ago and supposedly the source code and the data sets for creating all this will also be available now we talked

### [2:42](https://www.youtube.com/watch?v=AnCsmHrMPy0&t=162s) Video Compression

about a brilliant little paper earlier where scientists at Nvidia wanted to create high quality video conferencing videos by compressing them down far more than previously thought possible I would tell you what it does but I will now not it does not compress the entirety of the video information but it only takes the first image from the video and then throws away the entire video afterwards what why well it is able to do it because it learns about your head movement and gestures and is able to synthesize that in a way that is super close to reality that is fantastic however look compression arst effects still remain the lower we go with the bandwidth the worse it gets and now you can use their newer technique to eliminate those look when streaming games or video unless you have a really fast connection you get some of these artifacts along these lines and something perhaps even worse blocky artifacts but now with these new super resolution AI techniques ha no more look at that flag if you wanted to get this kind of quality you would either need a significantly more powerful internet connection to stream it or an expensive computer to render it yourself once again no more love it now not even this technique is perfect its earlier version was widely tested and testers typically found it to be working really well on static and slow moving content however when used on fast moving footage it might introduce its own artifacts now to the best of my knowledge it is already available on the RTX 2000 series graphics cards and app which is fantastic news these are from an age when these cards were a little more affordable more longevity for us I like it note that we do not have any business ties with Nvidia and now the ultimate question fellow Scholars we know that these modern AI techniques learn to listen see and even talk but

### [5:00](https://www.youtube.com/watch?v=AnCsmHrMPy0&t=300s) Smelling AI

here is a crazy question is it possible for an AI to learn to smell well this crazy paper attempts to give this a try first they give it a bunch of molecule structures as an input and they get information on what humans think this smells like for instance this molecule smells like a musty research paper with a hint of caramel and then the goal is that we give it new compound sounds that it hasn't seen yet excuse me hasn't smelled yet so what did it learn well it learned how we label these smells and now hold on to your papers fellow Scholars because when given these new molecule structures it can tell them apart at least as well as a human can that is mind-blowing now just to avoid any misunderstanding here this is a piece of software and it does not have a nose actually ual smelling in a way that humans smell something does not happen here it is given a molecule structure and it makes a prediction based on that H an AI that has no nose and yet it can still smell kind of what a time to be alive when I first heard this I was a little skeptical but I had a look at the incredibly thorough evaluation in the paper and I found that it really knows which smells typically appear together and can also put these smells on a map I think this is an absolutely amazing concept and with a wonderful execution as always the link to the paper is available in the video description if you're looking for inexpensive Cloud

### [6:44](https://www.youtube.com/watch?v=AnCsmHrMPy0&t=404s) Lambda GPU Cloud

gpus for AI Lambda now offers the best prices in the world for GPU Cloud compute no commitments or negotiation required just sign up and launch an instance and hold on to your papers because with the Lambda GPU Cloud you can now get on demand h100 instances for just1 199 per hour yes $19 and they are one of the first Cloud providers to offer publicly available on demand h100 accs did I mention they also offer persistent storage so join researchers at organizations like apple MIT and ctech in using Lambda Cloud instances workstations or servers make sure to go to lamb. com papers to sign up for one of their amazing GPU instances today

---
*Источник: https://ekstraktznaniy.ru/video/12810*