Yesterday, the most insane A.I. update of 2024 was released... and it wasn't by OpenAI. In fact, it's entirely open source. Let me explain. You've heard of text-to-image generators (e.g. Midjourney), maybe even text-to-video generators (e.g. Runway, Sora), but Genesis has released text-to-4D. 4D dynamical worlds—physics generators—will make the next few years feel like centuries in terms of human progress. Here are a couple of ways this will transform our reality in the coming years: 1️⃣ As an advertiser, I'd be remiss if I didn't talk about how this will completely transform video production. A long-standing issue with text-to-video generators is consistency. No more; now, you aren't just generating a video, but an entire world in which you can shoot consistent, never-before-possible shots from any desired angle and scale. You can see this in the video. 2️⃣ As a nerd, I'd be remiss if I didn't talk about how this will completely change the pace at which robotics is able to progress. It delivers a simulation speed ~430,000 faster than in real-time, and takes only 26 seconds to train a robotic locomotion policy transferrable to the real world on a single RTX4090. In layman's terms, embodied AI (intelligent robots) will be able to train in virtual space at an extremely accelerated rate that translates to real-world physics. This technology all but guarantees we'll be living in a future with intelligent robots before this decade is out. It may just be commonplace by 2030.
I had the same initial reaction... but in the critiques I've seen of this announcement/video, some folks rightfully called out that this was (likely? obviously? Dunno which) a heavily edited video and they've open sourced only a bit (i.e. "everything you see is on the roadmap") The thing that was/is most critical is whether you can reliably get the output in real time; that's the piece that remains to be seen (i.e. it seems more likely that they took the best examples after running multiple simulations and put them in the video). Seemingly game changing for video creation (and video _game_ creation), but unclear that it will be actually useful for realtime robotics until/unless they can prove realtime reliability. Good teaser though. I'm going to guess that we'll see actual real-world results from this much later than this video would have you believe (i.e. next decade not this one), but that it will play a big role.
No more trust at all in anything than face to face! Or?
While the pace of technological advancement is remarkable, predicting the arrival of truly intelligent robots requires caution. Progress often encounters unexpected hurdles, and intelligence itself is a complex, nuanced concept. That said, the trajectory certainly suggests significant breakthroughs by #2030 , if not sooner.
this is the umpteenth time I've seen it posted in a few days, but I can't stop being amazed.
yeah this was a big one
I wish they provide a feature where we could manipulate each of the forces minutely. That would be huge
I think LinkedIn should add the 🤯 emoji as a standard reaction to the stream of AI announcements these days.
Love this
High-ROI Facilitator • Geeky about the future of collaboration • Crusher of Soulcrushers
20hMy head is already exploding trying to keep up with ask the AI announcements this month, and this one took me over the edge. Elijah Kleinsmith.