Tensor Templar

@Apar_maker @lukas_m_ziegler This is what got me started too

@lukas_m_ziegler You'd better get familiar with Isaac Asimov's robotics books. Old, but never obsolete.

Most robotics books teach you theory. This one teaches you how robots actually work. 📌 If you’re self-learning robotics, this is genuinely one to save for later! It’s one of the few resources that brings mechanics, planning, and control into a single, unified framework. Ladies, and gentlemen: Modern Robotics: Mechanics, planning & control. If you understand basic physics, linear algebra, and a bit of coding, you can get through it. Instead of abstract formulations, it uses screw theory to describe robot motion in a geometric and intuitive way. Exercises at the end of each chapter, dedicated software, and video lectures all reinforce the same concepts from different angles. Which is exactly what you need for something as multi-layered as robotics. Here's the link to the full book: hades.mech.northwestern.edu/images/7/7f/MR… ~~ ♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com

@JessePeltan Its just the harness, but given the spaghetti quality and cognitive resignation of the team, according to their public statements so far, the weights will surely follow at some point

…did we just overtake China for best open source model? USA! USA! 🇺🇸🇺🇸🇺🇸

I get why open source is excited about this. I am excited too, but no hacker can afford the SSD in 2026 to exfil all of that

@jiratickets ESS, heatpump, garage ai servers

oh yeah? name 3 technologies

Ok I’ll happily take shit for this take again, but I struggle to understand how a company that hasn’t found a working business model can be worth $852 *billion*

@JessePeltan They are so cheap now that i routinely impulse buy pylontechs and will start bringing them as gifts when visiting friends if this continues.

It’s a good thing batteries got so cheap because we need SO many batteries.

@chris_j_paxton Can't bots just vibe-navigate?

Anyone who says they do not need SLAM is fundamentally not serious about robotics (unless the robot actually just doesnt move)

📖Robotics World Model Reading Club #01 Summary @BostonDynamics, @Stanford, @AGIBOTofficial, @intbotai, @BytedanceTalk, @Google, @moonlake, @Rivian, @Meta, @Samsung, @UCBerkeley, @Cruise, @encord_team, @ManycoreTech, @OpenGraph_Labs, @neuralmotion, @AMD, @nvidia, @oysterecosystem, @Zoom, @FusionFundVC, @BoostVC, @yzilabs... policy learning→WM VLA: observation→action WAM: latent world→future trajectory→controllable action →Shift=reactive mapping→controllable simulation @nvidia Gr00t (7B, high mem efficiency on Thor)≈DreamDojo-style WAM. Bottleneck is NOT scale, but missing unified interface across perception–geometry–physics–action. 🧠 Representation Pixel space is redundant & non-geometric. Trend→Explicit 3D backbone: point cloud/mesh object+sub-object representations geometry-aware tracking (contact, affordance) Point-flow pipeline: detect→sample keypoint→track→dynamic graph Core tradeoff=which points&density (motion saliency/affordance attn) 🌍 4D Reconstructi→Unified Latent @GoogleDeepMind D4RT encodes video→temporally consistent latent field: geometry+motion+visibility unified Outputs: point clouds, 3D tracks, full reconstruct (300× faster) ❗Gap: no shared latent across: vision/geometry/semantics/action/physics ⚙️ Physics Gap Sim2Real Gap=physics, not vision: discontinuous contact deformable objects (∞ DoF) non-differentiable friction Engineering fails: brittle collision meshes, unstable contact Solutions: learned physics proxy hybrid pipeline convex decomposition (geometry → collision proxy, ~5× speedup) 🎥 Video Pretrain≠Interaction Video=strong prior but no counterfactuals Missing: force, depth, tactile, proprioception →can't answer: what if act differently ⏱️ Control≠Inference Real world=high-freq loop action chunking latent action FastWAM (train with rollout, infer without) KV-cache (AutoGaze) 👉control selects feasible trajectory, not full future modeling Thor is good, but LLM scaling≠robotics scaling 📉 Data No “robotics internet”: sim/video/teleop/factory logs fragmented no unified labeling or metrics Reality: factories use fixed primitives generalization often unnecessary Bitter lesson: data flywheel>pipelines (but robotics lacks one) 🦾 Embodiment Gap manipulation→full-body intelligence loco-manipulation+gaze+coordination Need cross-embodiment align (space, action, kinematics) 🔁 Sim2Real Pipeline human data→semantics→geometry→collision proxy→sim→fine-tuning Unsolved: deformables, contact stability, long horizon 🧩 Paper VQVAE (discrete latent) VL-JEPA (predictive align) token pruning (efficiency) recursive models (depth reuse) multi-path exploration (GRPO) ⚡ Infra→SLM Real-time stack (LLM infra too slow) →WM must compress into SLMs Future=small, domain-specialized, grounded models 🧪Bottlenecks no unified representation no data flywheel inference–control mismatch physics fragmented embodiment Reality can't be scraped like internet. It must be sensed, interacted, simulated. 👉 Goal: jointly optimize representation+simulation+action under physics constraints 💡minimal sufficient representation? can video DiT become WAM? vertical SLM inevitable? robotics ImageNet moment?

@blind_via I want a 2-4m wingspan light cargo version of this to carry other robots.

That is a slick drone design. Literally a quad that transforms into a fixed wing, and its huge. Shut up and take my money!!!

More egocentric data is urgently needed for high voltage work. Volunteers?

@AndrewCurran_ Really really suspicious about such early eval results, given the silent collapse and jagged frontiers, the first explanation is always a stroke of lucky numbers. If the architecture is new though, they cant distill into old ones, meaning we wont get anything as early as april

Three weeks ago there were rumors that one of the labs had completed its largest ever successful training run, and that the model that emerged from it performed far above both internal expectations and what people assumed the scaling laws would predict. At the time these were only rumors, and no lab was attached to them. But in light of what we now know about Mythos, they look more credible, and the lab was probably Anthropic. Around the same time there were also rumors that one of the frontier labs had made an architectural breakthrough. If you are in enough group chats, you hear claims like this constantly, and most turn out to be nothing. But if Anthropic found that training above a certain scale, or in a certain way at that scale, produces capabilities that sit far above the prior trendline, then that is an architectural breakthrough. I think the leaked blog post was real, but still a draft. Mythos and Capybara were both candidate names for the new tier, though Mythos may now have enough mindshare that they end up keeping it. The specific rumor in early March was that the run produced a model roughly twice as performant as expected. That remains unconfirmed. What is confirmed is that Anthropic told Fortune the new model is a 'step change,' a sudden 2x would certainly fit the definition. We will find out in April how much of this is true. My own view is that the broad shape of this is correct even if some of the numbers are wrong. And if it is substantially accurate, then it also casts OpenAI's recent restructuring in a new light. If very large training runs are about to become essential to staying in the game, then a lot of their recent decisions, like dropping Sora, make even more sense strategically. For the public, this would mean the best models in the world are about to become much more expensive to serve, and therefore much more expensive to use. That will put pressure on rate limits, pricing, and subscription plans that are already subsidized to some unknown degree. Instead of becoming too cheap to meter, frontier intelligence may be about to become too expensive for most of humanity to afford. Second-order effects; compute, memory, and energy are about to become much more important than they already are. In the blog they describe the new model as not just an improvement, but having 'dramatically higher scores' than Opus 4.6 in coding and reasoning, and as being 'far ahead' of any other current models. If this is the new reality, then scale is about to become king in a whole new way. It would also mean, as usual, that Jensen wins again.

@basedjensen @mustafasuleyman It would have been sufficient to have asked how that belief is working out for mustafa so far. The data flywheel is increasingly shifting into sim, rl envs in post at the frontier - something that takes actual talent to train vs. the throw more compute and data at it approach.

No latency does not matter what matters is intelligence per $. You can serve slop at the speed of light and no one wants that crap. people will gladly wait for correct answers even pay more for it. But then again I can understand why you will be pushing this narrative given inflection bros you brought in m$ is still yet to train a decent model or one that comes close to frontier. All you guys have managed to do is drive people who can train intelligent models like wizard boys out

For the next couple years at least, the entire AI industry is going to be defined by this fact: demand is going to wildly outstrip supply, and so what matters is which companies / products have margin to pay for tokens. Those products will then rapidly improve because latency drives retention, and retention creates data to spin flywheels that improve the product and drive more adoption.

@redtachyon I think people looked up short term node rentals too much and think those are the prices the labs have to pay. Everything one can rent on demand is 2-5x overpriced at the moment.

Why do people keep saying that, AI is not subsidized. In the frontier labs, there's no way that each token is a loss (claude code discount might - MIGHT - be an exception) You might make an argument that R&D makes it more expensive, but - no shit? That's how investing works?

@cloneofsimo I always print the ones i read to get the best of both worlds

heavily underestimates the utility of toilet paper: they have strictly positive utility, unlike research papers which often have negative utility

@Truthful_ast If we needed consensus from such people we would, ironically, still be living in humid and dim rock

"Well I don't like it so why would humanity do it!!!!!"

@adcock_brett Nice, didn't know it had companion features.

So proud to see F.03 make history as the first humanoid robot in the White House 🤖 🇺🇸

What wail fail miserably is this take trying to age well. A classical category error, where we take a limited imagination and apply it to only tasks done today. Instead imagine what new jobs, what new capabilities a general platform could unlock, if a workforce could be raised on demand where none exists - to leverage resources that aren't profitable to utilize at scale, have no trained workforce for it, but are very profitable and relevant for local communities that exist nearby. If one looks at it with the single platform takes whole market lens of the pre-AI era - i can see how it remains a blind spot for investors.

I wonder if my build order from @TheCrustGame is going to carry over. Its not exactly the KSP of lunar mining, but the modular robot rovers drive into workshops to get their specialized setup, based on the tasks most needed, which is a reasonable way of doing it.