NVIDIA’s New AI: Insanely Good!

❤️ Check out Lambda here and sign up for their GPU Cloud:

Guide for using DeepSeek on Lambda:

📝 The paper "GR00T N1: An Open Foundation Model for Generalist Humanoid Robots" is available here:

📝 My paper on simulations that look almost like reality is available for free here:

Or this is the orig. Nature Physics link with clickable citations:

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Benji Rabhan, B Shang, Christian Ahlin, Gordon Child, John Le, Juan Benet, Kyle Davis, Loyal Alchemist, Lukas Biewald, Michael Tedder, Owen Skarpness, Richard Sundvall, Steef, Taras Bobrovytsky, Thomas Krcmar, Tybie Fitzhugh, Ueli GallizziIf you wish to appear here or pick up other perks, click here:

My research:
X/Twitter:
Thumbnail design: Felícia Zsolnai-Fehér –

#nvidia

Joe Lilli
 

  • @nisbahmumtaz909 says:

    sadly, robotics aren’t exactly cheap. a good humanoid arm alone can cost around 5k USD, let alone for an entire body
    so for now it’s still just a rich folks’ game, and i don’t see an arduino-like equivalent being cheap enough to enter the market any time in the current decade

    • @jyjjy7 says:

      These things should cost about as much as a car, after the usual early adopter phase. Cars are not “just a rich folks game”, these won’t be either

    • @leeroyjenkns5182 says:

      Imo it’s gonna become VR headset situation 2.0 when people assumed tech will become cheaper as it advances but it was not the case

    • @diophantine1598 says:

      After they start mass producing robots, they should be much cheaper.

    • @jyjjy7 says:

      @@leeroyjenkns5182 VR headsets are still not a particularly popular technology preventing economy of scale commodification, yet even so psvr and quests are quite reasonably priced. On the other hand soon after robotics reach reasonable functionality they will be able to build more of themselves which differentiates it from any tech production curve you could try to compare it to

    • @OverbiteGames says:

      @@leeroyjenkns5182 they are cheaper?

  • @maythesciencebewithyou says:

    All they would need to do would be to build factories that are not designed for humans, then they wouldn’t even need to build humanoid robots. In many factories, people just carry things from one machine to another.

  • @mihirvd01 says:

    WHAT A TIME TO BE ALIVE !!!!!

  • @JanCRefsgaard says:

    The discrepancy between video length and channel name is proportional to how long AI has come, I predict the day he makes an 1 hour video is the day the singularity comes.

    • @jyjjy7 says:

      @@JanCRefsgaard N ah, the singularity is when any video anyone would want to make on the subject of AI is obsolete before being released

    • @creatvsdd99 says:

      I desperately hope the Singularity comes before 2045:) Very cool video though. Keep em coming.

  • @xewi60 says:

    cat waifu android in our lifetime?

  • @Bublark says:

    I wonder if the robots with guns will be super cute. The future is going to be nuts

  • @CharlotteLopez-n3i says:

    GR00T-N1’s 76% success rate is wild. Ready to see devs and researchers push this open model further!

  • @slzzzzzzzz says:

    This is why I pour all my money into the tech sector. I did not pay much attention to AI until ChatGPT came out in late 2022. But during this 2.5 years, I saw with my own eyes the exponential progress in AI. When I first saw Will Smith’s spaghetti video in 2023, I said, “nah, AI video generation is a LONG road ahead, at least 5 years”. And now, merely 2 years later, look at the quality of AI videos now! There are countless examples like this. If you compare it with tech innovation in the past 50 years, if you are familiar with the history, you’ll immediately realize how unthinkably fast this AI race goes. I have seen much skepticism about robotics and other AI things. I’d say, “let’s wait another 2–3 years”, and if you look back then, you might be shocked how we achieve the future so fast.

    • @sahillodhi7269 says:

      @@slzzzzzzzz you are absolutely right brother 👍

    • @ErikLiberty says:

      @@slzzzzzzzz I watched a 60 Minutes episode about how AI video needs to be trained on labeled videos in order to get better, where every individual thing in the video has a label (car, human, house, tree etc). The AI companies are paying people in Africa to label these videos manually. This is why AI videos have gotten better, but it is time consuming. Now that this paper enables AI to do the labeling, I bet there will be a huge leap in video creation quality!

  • @mindovermatter82 says:

    What a time to be a robot

  • @FuZZbaLLbee says:

    As a Dutch person seeing a cute robot and the word “Groot” at the same time, does not make sense 😛

  • @joel16961 says:

    As much as I love the info and insights – my favorite part of your videos is being called “Fellow Scholar” 😃

  • @greevar says:

    While AI is doing impressive things, NVIDIA has been seriously neglecting its gaming customers. Melting power cables, driver issues, price inflation, losing AIB partners. It’s not a good look on NVIDIA.

    • @lordvadher4 says:

      Yeah… the real money is in AI now…. games are thing of the past… until AI games are a thing at least

    • @zyang056 says:

      Explosives can be made into fireworks for fun, or TNT to reshape earth for large scale civil engineering projects. Nvidia is transitioning from a fireworks company to a TNT company.

    • @kevinforget549 says:

      @@lordvadher4 its a literal power connector though. all they had to do was make it load balanced or just reduce the number of wires and make them and the connector larger. weve had xt90 connectors for a long time that can handle 600 watts at 12 volts no problem.

  • @worskaas says:

    I’m still waiting for them to create a virtual reality that simulates everything, for example the jitter and flex of the limbs and the rest of the unit. How it can make use of gravity to its advantage and seamlessly move around. Simulation always appears so smooth and then when we see the actual real world operation it seems very far away from how a human can do the same task, not only do things efficiently with the least amount of effort (albeit over time of course).

    I would like to hear more about feedback sensors and how that will update the virtual world so that the simulation training can include that and try to achieve actual real world smoothness and efficiency.

    Maybe I’m asking for too much.

  • @pasqualz says:

    2 more robots down the line and we will all hail our robot overlords. 😂

  • @elzurotsyry says:

    I want this guy to make a video when Skynet becomes self-aware: “Skynet just sent two Terminators to my house—WOW. Their shooting accuracy was 89.5%. AMAZING“

  • @abderrahimbenmoussa4359 says:

    What I like is that in few years we moved from basic stuff to “Oh robots, ok no problem”, “Genius AI that can do whatever you want ? Sure”. It’s strange that all of it feels just normal and how science fiction stuff just enters real-life without the chock we would expect. It’s just incrementally getting in our lives.

    • @IceMetalPunk says:

      Which is why every time someone says “AI will never be able to [X]”, I’m always like, “You mean it can’t do [X] *now* — never say never.” Because so much of what AI was supposed to “never be able to do” just years ago, it’s already doing today.

  • @Tiogar60 says:

    LLM’s are great and all, but we all want AI for robots!

  • @Asonc1 says:

    if China didnt open source anything, this would be behind a pay wall but now opensourcing itself has become a bit competitive

  • @GimR says:

    I’m gonna guess it’s open source because it encourages people to use NVIDIA chips / framework

  • @tvtvtvtv98 says:

    Did we learn nothing from the Terminator movies?

  • >