Foundation models for robotics are entering a GPT-3 era where pre-training on massive diverse physical interaction data yields emergent capabilities like ambidexterity and tool generalizati…

Collecting millions of hours of human demonstration data via egocentric sensors (GoPro + simple grippers) creates the critical data flywheel for robotics foundation models, analogous to int…

Pete Florence argues robotics foundation models are at a GPT-3-like inflection — Gen One achieves 99%+ reliability on new tasks with just one hour of data, training time has collapsed from…

Foundation models for robotics (Gen One) now show clear scaling laws — more data/compute yields predictable gains — crossing the threshold from research to commercial viability for simple t…

Unlike LLMs, robotics foundation models cannot download training data from the internet — they require closed-loop systems that generate, filter, and iterate on proprietary physical interac…

Florence emphasizes that robotics data cannot be scraped from the internet; the moat is a closed loop between large-scale data creation, model training, and intelligent data prioritization…

Generalist's Gen One model proves scaling laws work in robotics, achieving mastery-level reliability (99%+) on new tasks with just one hour of training data, pulling commercial viability fo…

Florence draws from his sustainable-energy research background: the bottleneck in physical science has always been PhD-student hands running experiments; dexterous robots turn any lab into…

The primary bottleneck in scientific discovery (solar cells, drug synthesis, etc.) is PhD-student-scale physical experimentation; general-purpose embodied AI replaces specialized high-throu…

Lonsdale and Florence argue general-purpose robots amplify skilled tradespeople, letting one expert oversee a fleet — 10xing output without requiring a full labor pool — making robotics ess…

Robots as general-purpose manipulators remove the physical bottleneck in manufacturing and science labs, letting one skilled operator oversee a fleet — turning every facility into high-thro…

Lonsdale predicts robotics follows software's trajectory — coding agents let anyone create digital products, shifting engineers to review; similarly, robot fleets will let anyone assemble p…