WATCH: @StanfordHAI Director James @Landay at the AI+Science conference on why openness isn't optional: It's our responsibility at this critical inflection point.
diversity is all you need - “10x increase in parameters only yields ~ 8% bump in success rate. Meanwhile, a 10x increase in data volume yields a ~40% jump, and a 10x increase in task diversity yields a ~55% jump in out-of-distribution generalization”
That's sick! 🤯
Genesis AI simulates robots playing yo-yo! 🪀
@gs_ai_ just open-sourced Genesis World 1.0, and it might be one of the most important infrastructure releases in robotics this year.
Robotics is still bottlenecked by the 1× speed of the physical world. Every model needs to be tested on real hardware, slowly, expensively, with limited coverage.
Genesis World 1.0 from Genesis AI flips that equation:
One hour in reality becomes 100 days in simulation.
That turns a wall-clock bottleneck into a compute problem. And compute problems are solvable.
The technical stack they rebuilt from scratch is serious:
→ GPU-accelerated cross-platform compiler via Quadrants, 10x faster launch time and up to 4.6x runtime vs the initial Genesis release
→ Penetration-free multi-physics contact solvers, the thing that makes simulation actually trustworthy
→ Unified rigid AND deformable physics in a single engine
→ Nyx, a high-performance path-traced rendering engine purpose-built for physical AI
The sim-to-real gap has historically been the graveyard of robotics research.
Policies that work beautifully in simulation fall apart on real hardware.
Genesis World 1.0 is a direct attack on that problem. And it's fully open-source.
The companies that master simulation infrastructure will train better robots faster than anyone else.
Find it here:
Genesis World 1.0: https://t.co/VPqoOgH9oG
Quadrants: https://t.co/ljleUrpktg
Nyx: https://t.co/zWNNP73Lvl
@theo_gervet, @zhou_xian_ congrats! 👏🏼
~~
♻️ Join the weekly robotics newsletter, and never miss any news → https://t.co/GoA3ZuwWF9
@hamandcheese Thanks for making a good case @hamandcheese
- I agree that "anyone who claims to know with certainty is deceiving themselves". But I think there are good reasons to be skeptical that substrate-independent computation is sufficient for consciousness https://t.co/jeawYtBXKN
Atlas hebt einen Kühlschrank. Klingt nach Demo, ist aber ein Durchbruch.
Denn der humanoide Roboter von @BostonDynamics weiß vorher nicht, wie schwer das Objekt ist oder wo sein Schwerpunkt liegt. Er passt sich in Echtzeit an, nutzt seinen gesamten Körper und balanciert das Gewicht dynamisch aus. Das wirklich Entscheidende: Dahinter steckt ein generelles KI-Modell. Atlas hat genau diesen Kühlschrank vorher noch nie “gesehen” oder trainiert. Und genau das ist der Unterschied: Keine Einzelfähigkeit, sondern echte Generalisierung. Ein weiterer durch für Humanoide Roboter 🤖
@frank_thelen@BostonDynamics Es ist kein Durchbruch. Zeigt aber wie BD das Thema völlig unaufgeregt und kontinuierlich nach vorne trägt. Weiterhin der ganzheitliche Technologieführer was vor allem auch Qualität und Robustheit betrifft.
How can VLAs achieve 95+% reliability?
Using RL post-training with EXPO-FT:
- π0.5 improves to 30/30 success on all 8 tasks tested
- uses only 19 min of RL data on average
Paper & videos: https://t.co/54nO9tFU0Z
looks promising .. "Interaction-Centric Tokens (ICT) encode spatial relationships among all entities"; "Even with half the data, HumanEgo outperforms robot teleoperation."; aux loss: "lightweight world model of hand–object interaction"
I believe that if the long-term goal is robots that operate effectively in the human world — helping people in homes, kitchens, labs, and workshops — then the most direct and natural source of supervision is people themselves interacting with that exact world!
@TX_Leo_Wang Looking at full humanoids this makes sense. But this comes with a limiting factor. Machines can do things different and better in sense of enabling and simplification. The future is diverse and we have to find ways of understanding the core of interacting and behaviors.
Instead of watching an hour of Netflix, watch this 2 hour hour Stanford lecture will teach you more about how LLMs like ChatGPT and Claude are built than most people working at top AI companies learn in their entire careers.
@StuartHameroff That’s just guessing not more. You know this. All we do is to apply what we know to the unknown. If we’re not transparent here we limit ourselves to arrogance.
Daughter: „looks nice, but a bit slow“ .. ok Claude - please auto-research fast inference on Jetson Thor, my cuda graphs baseline is 691ms e2e …5h 130 experiments later auto-research plateaued at 285 ms. Integrating into real robot repo easy >2x speedup 🤯
🪶Our hand can take a hammer hit but also detect a feather. The finger moves until it meets a tiny resistance and stops, with no tactile sensors.
Back drivability and torque transparency let it feel the world through its own drive currents, enabling simple, reliable interaction.
🎊📜 NEW PAPER 📜🎊
Can we seriously build synthetic consciousness?
And if so, where do we start?
I’m super excited to present recent publication in Neuroscience & Biobehavioral Reviews where @jaaanaru and I confront this challenge head on.
https://t.co/6Jk9Ce6eE5
1/n
Thinking Machines chief scientist John Schulman on what’s changed in AI since 2015–2017:
“Previously the people were a bit weirder.”
“Engineering skill matters more now… as opposed to research, taste and the ability to do exploratory research.”
“There’s so much low hanging fruit just from scaling the simple ideas and executing on them well.”
“People who have more of a software engineering background have more of an advantage now.”
@johnschulman2@thinkymachines@mntruell@cursor_ai
Top story of 2025: AI models are 280x cheaper, adoption hit 78%, China caught up, and incidents spiked 56% to record highs. See the full picture in the #AIIndex2025: https://t.co/mmq0qsKrfs