CFG_Labs_Intern - e/acc

Who invented deep learning? people.idsia.ch/~juergen/who-i… Timeline: 1965: first deep learning (Ivakhnenko & Lapa, 8 layers by 1971) 1967-68: end-to-end deep learning by stochastic gradient descent (Amari, 5 layers) 1970: backpropagation (Linnainmaa, 1970) for NNs (Werbos, 1982):…

Google AI Futures Fund + Accel Atoms are fueling the next wave of innovation. 📷 Get early access to Google DeepMind’s most advanced models (Gemini, Imagen, Veo), capital, expert support, and Cloud credits. Applications are open now—apply here today. 📷 atoms.accel.com…

This week, I am very excited to share that @chris_j_paxton and Heidi Schubert are joining Parallax as advisors! Chris is an amazing roboticist with a career spanning NVIDIA, Meta, Agility Robotics & more at the forefront of robotics research. Heidi previously spent decades…

Also, I received too many emails from students who claim to work on VLA, but don't even know PID/MPC control, 3D reconstruction, or classic point navigation. It is fine to play with LLMs and VLMs on benchmarks, but in the real world, physical experimentation can have catastrophic…

"Infinite use of finite means" has been the most important inspiration for our lab's research. Our works have explored how to compose modular generative models to generate long-form {plans, images, videos}. Excited to introduce the newest incarnation of the idea: CompDiffuser…

cross entropy & KL divergence are very fundamental concept but quite hard to visualize. great video by artem kirsanov where he explains cross entropy from first principles. these kind of videos for some reason helps a lot like whether its of 3b1b or welch labs, most of them are…

Flexion Robotics just raised $50M for building brain for robots! 🧠 Zurich-based @FlexionRobotics closed a $50M Series A to build what they call the intelligence layer for humanoid robots. The team, led by @rdn_nikita (ex-ETH Zurich, NVIDIA, Meta, Google, Tesla), is building a…

don’t overthink robotics. learn the primitives that actually make machines intelligent: learn kinematics (forward + inverse) learn dynamics (forces, torques, energy flow) learn control theory (PID → LQR → MPC) learn estimation (Kalman filters, sensor fusion) learn SLAM…

Bringing robots into the real world needs something we rarely talk about: human-scale data that captures how tasks actually unfold. Spatial AI just released SEA, a new egocentric dataset designed for training robot foundation models. It is focused on long sequences, natural…

My AI investment thesis is that every AI application startup is likely to be crushed by rapid expansion of the foundational model providers. App functionality will be added to the foundational models' offerings, because the big players aren't slow incumbents (it is wrong to…

DeepSeek [1] uses elements of the 2015 reinforcement learning prompt engineer [2] and its 2018 refinement [3] which collapses the RL machine and world model of [2] into a single net through the neural net distillation procedure of 1991 [4]: a distilled chain of thought system.…

Memo for near-term home robots vs Figure is utility-first over human mimicry philosophy: - Wheeled base + lift column = passive stability, 5-10x longer battery life, no falling, far cheaper/safer than bipedal balancing act - Simple dual parallel grippers + elite software =…

Sunday Robotics unveiled its home robot, Memo, a wheeled robot with two arms and pincerlike hands. Training method: Sunday pays remote workers to perform household tasks wearing gloves that resemble Memo's hands. The Mountain View-based company is focused on a full-stack…

🚀 Excited to share that our team at Meta just launched SAM 3 + SAM 3D! These models set a new standard for segmenting & reconstructing media, unlocking new possibilities for developers, researchers, & creators. SAM 3: go.meta.me/d74ec5 SAM 3D: go.meta.me/f019f3

So many exciting breakthroughs from robotics startups recently (@physical_int, @sundayrobotics, @GeneralistAI, @DynaRobotics, etc.). I genuinely believe this won’t be a winner-take-all sector — multiple strong players will coexist, compete, and ultimately push the field forward…

You can just train ViT from scratch to solve ARC.

π*0.6: a VLA that Learns from Experience Blog: pi.website/blog/pistar06 Paper: pi.website/download/pista… New research blog and paper released by Physical Intelligence today: π*0.6 improves previous generation of π model by incorporating reward feedback for RL in all stages of the…

看了段永平和雪球创始人的访谈 我发现他真的有在 非常小心地呵护自己的主体性 他只要感觉到环境不舒服 自己不适合就会头也不回 立刻选择离开 段永平大学毕业后 被分配到北京工作 在北京工作呆着无聊不舒服 然后赶紧通过考研破局 去了广东 广东第一份工作 做了三个月又离开了 感觉自己选对了路…

RoboParty , a fully open-source bipedal humanoid robotics firm, has completed a seed round financing in the tens of millions of US Dollars, marking the largest seed round in China's open-source humanoid sector. The round was co-led by Matrix Partners China, Xiaomi Ventures,…

Sutskever's List In this book, you go through 30 key deep-learning papers and, in plain English, see what each did and the practical ideas you can apply. You’ll discover: - The technical breakthroughs that have shaped AI strategy, safety, and cultural shifts - How to decode…