Yash

This is a great watch! Scaling data (what type of data?) and building the infrastructure required for it, is one of the biggest bottlenecks in robotics. Extra cool that egoscale was one of the first papers that validated our thesis. youtu.be/3Y8aq_ofEVs?si…

@thejesonlee would love to join!! Building general purpose robotics (current YC batch)

May 12th in SF, Where the Opportunity Sits in Physical AI by @join_savant For our next Choose Good Quests session, we are hosting @bznotes ,managing partner at @redglassvc and @KanuGulati, partner at @khoslaventures to share about opportunities in physical AI. Physical AI is moving fast — but for founders, the map is still being drawn. Bilal and Kanu have backed some of the most ambitious companies across AI, robotics and autonomy. They’ll share what they’re seeing from the investor seat: where the biggest opportunities are, what makes a strong founding team, how founders should think about wedge, market timing, defensibility, and what kinds of Physical AI companies they’d be excited to fund. Come join us. Link in the thread 🧵

Awesome thesis! Super excited to launch Intelligence Factory soon

@chris_j_paxton @gs_ai_ This makes sense for pre-trained models, but post-training is still unsolved. I imagine when they begin deploying, they’d see multiple edge-cases and the data flywheel approach takes months to hit metrics that a human would do

Pipetting, cracking an egg, using tape -- we have seen these from other teams now but they're all very bleeding edge capabilities, exciting stuff; huge congrats to the @gs_ai_ team Another entry into the 'data and good controls is all you need' camp of robotics learning, interesting that they seem to have moved away from the simulation angle and go for this glove instead

@sir_aymansaleh Appreciate it😂😂

@coldifl Now it’s 3

Excited to annouce the launch of my company Chronicle Labs (YC P26); the staging environment for enterprise AI agents. Just like trading teams backtest algorithms before deploying capital, companies should be able to backtest agents before deploying them into real workflows. We turn operational history into replayable sandboxes, so teams can test, debug, and safely ship better agents.

@chris_j_paxton Egocentric data is not enough. That is one of the modalities, you cannot solve generalized robotics without action and tactile information

Its not the only thing holding general purpose robots back! But its one of the big things. Its also one that, to be honest, seems like it will be solved in short order given how much egocentric data is being collected

The way we learn any task - by seeing, doing, and feeling - is how we train robots. We're building general intelligence that works on any robot and fits seamlessly within our world. In the coming weeks, I will share more about our approach as we deploy our across verticals.(2/2)

Intelligence Factory is backed by @ycombinator and @CommaCapital . With hardware converging, Intelligence is the biggest bottleneck. Our world was built for humans and we're giving robots the ability to reason and act like us. (1/2)

@pliang279 Looks sick! We are also developing hardware for tactile data collection. Although, inclusion of tactile feedback into current model architecture is still an open question. Would love to chat

Most of today's AI can see the world, but it doesn’t **feel** it. Capturing the sense of touch is crucial for dexterous robotic manipulation, user modeling, and understanding physical interactions. Introducing OpenTouch: bringing full-hand tactile sensing into real-world AI🖐️ OpenTouch is collected in-the-wild using tactile sensing gloves, hand pose tracking gloves, and egocentric glasses. It includes: • 5 hours of real-world data, • 3 hours densely annotated contact-rich interactions, • 2,900 curated interaction clips, • across 800 objects, 14 environments, and 29 grasp types. all open at: opentouch-tactile.github.io

@oyhsu @chris_j_paxton The only certainty is that high quality data is a defensible moat (even though everyone has a subjective opinion of what is high quality). After that model architecture and hardware are open questions

Robotics is in an interesting place right now where there is space for a variety of methods and approaches, even within the scaling paradigm. The ecosystem is evolving in a Darwinian fashion with regular breakthroughs, pivots, and emergent behaviors. This is a dynamic that rewards adaptability, comprehensiveness, and an exploration-first approach. The thing to do is to do everything. The age of research indeed.

Collecting and scaling the right data is the unlock for robotics. Everyone has a different interpretation of this.

@xiao_ted @peteflorence Couldn’t agree with this more. I think people underestimate how big of an infra challenge robotics is, along with technical

Robotics pre-training *from scratch* has been a heretical idea for the last two years. That “there’s no internet of robotics data” has led to two prevailing conclusions: 1) we need to use pretrained model backbones and 2) we need to scale robotics data. The first conclusion in particular is painful. We want the generality and scaling properties that have transformed the rest of AI; but we aren’t sure if the designs and models that were appropriate for text, vision, and audio will also be appropriate for robot actions as a first-class modality. So is this trade-off worth it? All leading robotics labs have believed so for the last 2 years. Every serious player has relied on rich pretrained priors resulting from large-scale internet training (via VLMs, LLMs, or video models). The correctness of the second conclusion seems more obvious, but the execution is not. The devil is in the details, and *what* robot data you scale and *how* you scale it matters immensely. I really appreciate that from the start, @GeneralistAI has taken an unusually considered and unique approach to both conclusions above. Eschewing pretrained representations and co-designing novel hardware/data collection/model from scratch is not an obvious path to take. This requires immense conviction and necessitates long-term pain tolerance. The team at Generalist has displayed both with clear-eyed thoughtfulness. Like all things in Physical AI today, there are still so many open research questions. I deeply respect and enjoy the rapid progress that @peteflorence @andyzengineer @coolboi95 @ColinearDevin are making on a very differentiated direction. The GEN-* reports have been some of my favorite recent robotics works, and I eagerly await the next updates! 👏