Figure

In a new interview, Figure''s CEO Brett Adcock doubles down on his "autonomy-first" strategy, dismissing competitors'' tele-operated demos as "deceiving" and explaining why his team was "way better" at robotics AI than former partner OpenAI.

Startup Generalist AI has unveiled GEN-0, an embodied foundation model it claims is trained on an unprecedented 270,000 hours of real-world manipulation data. The company reports its new architecture, "Harmonic Reasoning," and massive dataset have unlocked predictable scaling laws for robot intelligence.

A recent LA Times report highlights the massive, low-tech effort to gather real-world data for robots. This data bottleneck is the central challenge in robotics, and top companies are making wildly different bets on how to solve it—from human-video capture and teleoperation to massive simulation.

The $20,000 NEO humanoid robot pre-order has drawn massive online attention, but it has also polarized the tech community, pitting proponents of its transparent, human-in-the-loop strategy against critics—including rival CEOs—who label it an unfinished ''hype reel''.

Speaking at an NVIDIA GTC panel, Figure CEO Brett Adcock dismissed teleoperation as "soy stuff" and claimed the real race is for general-purpose AI, not manufacturing, in the US vs. China competition.