Michael ZHANG·Oct 28Physical RSI: From Self-Improvement to Measurable ValueCan robots keep learning from physical experience — and gradually improve the way they learn?
Michael ZHANG·Oct 13How General-Purpose Foundation Models Could Reshape Physical AIGeneral-purpose foundation models point toward an important shift: Physical AI systems may inherit more cognitive capabilities and use AI…A response icon1A response icon1
Michael ZHANG·Sep 15AI Can Run a Million Experiments. Can It Make a Discovery?AI is becoming remarkably good at doing research.
Michael ZHANG·Sep 3Diversity Is the Key to Scaling Real-World Data for Physical AIThe next data challenge for Physical AI is not only collecting more real-world video, but capturing more diverse ways in which humans…
Michael ZHANG·Aug 19From Texts to Pixels to Physics (Part II)From Capturing Pixels to Capturing Physics
Michael ZHANG·Aug 5From Texts to Pixels to Physics (Part I)How Foundation Models Are Learning the Real World
Michael ZHANG·May 28From Robot Demos to Deployment ReliabilityWhy Physical AI Must Master Ordinary Tasks Before General IntelligenceA response icon1A response icon1
Michael ZHANG·May 26The Hardest Tasks in Physical AI May Look SimpleRethinking robotics through Moravec’s paradox — and why ordinary physical tasks may define the next breakthrough in AI.
Michael ZHANG·Apr 26When AI Becomes Smarter Than Humans, Trust Becomes EverythingLarge models may surpass human knowledge and reasoning — but their long-term success will depend not on model brilliance alone, but on data…
Michael ZHANG·Apr 15The Hardest Part of Physical AI is Entering the Real WorldThe next phase of Physical AI may be won not by the smartest model, but by the ability to capture millions of real-world human workflows…