evals, retraining, and reality: silicon valley edition

​evals, retraining, and reality

​post-deployment physical AI

​a policy that worked in the lab meets a customer’s floor. the lighting changes. an object shows up that nobody collected data for. this is where most of the real work in physical AI happens, in the evals, the failure data, and the retraining that turns every deployed hour into a better model.

​after our first edition in bangalore, boundless and qualcomm are bringing this closed technical session to silicon valley, for founders and operators deploying physical AI. how do you verify a robot once it’s live? what do you do with failure data? how do you turn what went wrong into the next model? a few operators open with where their deployments stand, then the rest of the group takes over.

​17 november, 6pm. qualcomm, santa clara