@ethanmclark1: Working in robotics right now is what I imagine working with language models felt like in 2023. Everyone throwing thing…
Summary
A robotics researcher compares current robotics approaches to the language model landscape of 2023, arguing that representation prediction (JEPA) is the most scalable method as it can leverage action-free video data like YouTube, unlike other methods that require action-labeled data.
View Cached Full Text
Cached at: 06/22/26, 11:41 PM
Working in robotics right now is what I imagine working with language models felt like in 2023. Everyone throwing things at the wall to see what sticks
Pixel prediction (Cosmos), action prediction (VLA), reward prediction (TD-MPC), and representation prediction (JEPA). Different paths for the same problem
The recipe that won in language was self-supervised pretraining at internet scale then light finetune on top. Only representation prediction runs that playbook. It learns from action-free video data so you can pretrain on YouTube and egocentric data then add a control layer. Everything else needs action-labeled data that doesn’t scale
As an RL maximalist, I used to hate LeCun’s cake. Turns out he was right all along which is how I ended up a JEPA truther
Similar Articles
@rohanpaul_ai: Language had a strange advantage robotics does not: Text is already a compressed, shared interface for human thought, w…
Discusses the challenges facing embodied AI and robotics, including a 100,000-year data gap and lack of shared benchmarks, and highlights startup opportunities in data loops, eval systems, and deployment.
This startup thinks robotics is about to have its ChatGPT moment
General Intuition, a startup building a foundation model for robotics trained on video game data, argues robotics will follow NLP's pattern with general-purpose models, and has raised $320M at a $2.3B valuation to pursue this vision.
Why Video Agent models are next — Ethan He, xAI Grok Imagine (98 minute read)
Ethan He from xAI discusses why video agent models are the next frontier, arguing that video models derive intelligence from LLMs and that the evolution of video generation will mirror AI coding, shifting from one-shot output to multi-turn planning and execution.
@macrodata_labs: Everyone is betting on Egocentric data to scale robotics But turning that footage into training data requires recoverin…
Macrodata Labs releases a research blog on scaling robotics with egocentric video data by recovering 3D hand motion signals using only open-source models.
@kshenster: Three approaches are racing to build robotics' data flywheel: foundation models (@physical_int @GeneralistAI @SkildAI),…
This article analyzes three competing approaches in robotics development—foundation models, deployment-first, and full-stack—and suggests that each can lead to significant companies in the large robotics market.