The Download: gig workers training humanoids, and better AI benchmarks

Editor's take

Gig economy workers are being enlisted to remotely train humanoid robots by performing tasks in simulated environments, a novel approach to developing sophisticated AI for physical interaction.

This distributed training model, reminiscent of early crowdsourcing efforts for data annotation, could significantly accelerate the development of robots capable of performing complex, real-world tasks, potentially impacting industries from healthcare to logistics. The reliance on a global gig workforce raises questions about labor practices and the scalability of such a decentralized training paradigm.

Future developments will hinge on the efficiency gains achieved through this method compared to traditional robotics development and the ability of these trained models to translate effectively to physical hardware like Agility Robotics' Digit or Tesla's Optimus. Observing the cost per training hour and the real-world performance metrics of these humanoids will be crucial indicators.