Two Minute Papers (Dr. Zsolnai) explains three recent NVIDIA offerings: Picasso (generative AI for text-to-image, text-to-video, and 3D model generation, with examples like video inpainting and person-boundary editing and mentions of copyright-safe training), Omniverse (high-fidelity virtual-world, factory and robotics simulation including light-transport and physics rendering used to prototype layouts and train autonomous systems), and Bionemo (an AI-assisted drug-discovery platform integrating tools like AlphaFold and ProtGPT2 to speed parts of the discovery pipeline). The episode includes a short advertisement for Lambda GPU Cloud pricing and closes with a call for viewer comments.
Questions this video answers
April 15, 2023
$1.10 per hour
AWS: $4.10 per hour; stated savings: 73%
Chapters
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today we will talk about Nvidia's new products that will likely revolutionize art, medicine and even virtual worlds. These are the folks who make these amazing graphics cards that ChatGPT and GPT4 are trained on, but they also do so much more. So here are three new things that really caught my eye and I hope they will catch yours too. First was their new generative AI solution they call Picasso. What does it generate? It takes a piece of text and generates an image. Ok, we already know about this, but it can also generate a video. Normally I would say this is going to be huge, but it already is. You see, Runway is already making use of Nvidia's tools for amazing video editing magic. It provides a holistic solution where we can find the boundaries of a person in a video and move him, but only him, change the background, use video inpainting to get rid of unwanted details, or even perform video to video transformations and give us a complete makeover with just one text prompt. This is going to revolutionize filmmaking. Loving it. Also, Picaso can make sure that these systems are only trained safely on data that we have copyright to. But wait, hold on to your papers, fellow scholars, because from a piece of text it can also render something else and that is creating 3D models. Oh yes, now we're talking. Don't forget, Nvidia also sells the graphics cards that run today's games and anything that includes virtual worlds. For instance, here with Omniverse, we can breathe life into these virtual worlds by only using our imagination and of course some text prompts to make them a reality. Second, their Omniverse system is now so powerful that using it, we can build a virtual factory and simulate how it will work before building the real factory. Technologies like this will only become more important as the USA is looking to build a ton of cheap fabrication facilities domestically in the near future. What's more, Amazon is already using these autonomous self driving robots in a warehouse so that they can move these boxes around safely and efficiently. But before they build these, they are using these tools to simulate what such a warehouse should look like and how it should work. And now you see something at work that is very close to my heart and that is light transport simulations, or in other words, the rendering of these videos. These are now so accurate that we can see in advance what our little world and our little robots will look like in reality before we build them. In such a system, we can place the sensors and design their geometry according to our wishes. And of course you fellow scholars know that I cannot resist the temptation that they can also simulate not only what things look like, but what they should behave like. Yes, that means fluid simulations, a fire simulation, or even other kinds of physics simulations as well, many of which can run in real time. These systems are now good enough to test self driving cars in a virtual world and train them there before deploying them into the real world. Technological advancement is accelerating at a rapid pace and just imagine what good this could do. And in general, just think about it. We are simulating reality and steer the simulation into one particular reality that we are looking for. How cool is that? What a time to be alive. Third, they also launched Bionemo, a system that is aimed to save human lives. Yes, this is a new AI based drug discovery service. It includes the amazing AlphaFold from DeepMind and a lot of other systems. Protgpt2, a language model for designing proteins and systems for molecule generation. Remember that whenever we talked about medical papers I always say that this is an important step towards drug discovery. But that is a long, long journey. And this is finally a system that runs super fast and helps scientists with several elements of this journey at the same time. And once again here it is not the AI that discovers the drugs. The AI is there to help humans do their work a lot better. So what do you think? What would you use these tools for? Let me know in the comments below.
If you're looking for inexpensive 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 Lambda GPU Cloud you can get on demand a 100 instances for $1.10 per hour versus $4.10 per hour with AWS. That's 73% savings. Did I mention they also offer persistent storage? So join researchers at organizations like Apple, MIT and Caltech in using Lambda Cloud instances, workstations or servers. Make sure to go to lambdalabs.com papers to sign up for one of their amazing GPU instances today.
Thanks for watching and for your generous support and I'll see you next time.
