Generalist AI

Generalist published a blog post describing GEN-1.5’s one-shot learning via in-context prompting, including one-shot/few-shot learning from demonstration and zero-shot physical generalization, and describing its multimodal inputs and 100 Hz trajectory output.

Generalist described GEN-1 being trained to support many robot end effectors (“hands”), including a dataset spanning more than half a million hours of real interaction data and thousands of end-effector/tool variations; it also describes on-the-fly adaptation tests when the end effector changes mid-task.

Generalist announced $400 million in new funding (bringing total raised to more than half a billion dollars) and named Radical Ventures as lead, along with several new major investors and participation from existing backers.

Generalist published GEN-1 details, describing crossing a performance threshold for “mastery” of simple physical tasks and discussing reliability improvements, task execution speed, and improvisational intelligence claims.

Generalist described a concept it calls “physical commonsense” as reactive, closed-loop intelligence behind interacting in the physical world, positioned as emerging through scaling.

Generalist introduced GEN-0 as a class of embodied foundation models trained on multimodal training data from high-fidelity raw physical interaction.

Generalist published a blog post discussing its perspective on why its approach goes beyond world models and VLA framing (the post appears in the company’s blog, though the excerpted search view did not provide detailed technical claims).