OpenAI disbands robotics team, blames data gap for AGI push

OpenAI has shut down its robotics research team. Cofounder Wojciech Zaremba disclosed the decision on a Weights & Biases podcast, as reported by VentureBeat. "Here's a reveal ... as of recently we changed the focus at OpenAI, and I actually disbanded the robotics team," Zaremba said.

The reason he gave was data. "From the perspective of what we want to achieve, which is to build AGI, I think there was actually some components missing," he said on the podcast, per The Register. OpenAI found pretraining in data-rich domains delivered gains faster than reinforcement learning on physical robots; by one line from the podcast, pretraining buys models "100 times cheaper 'IQ points.'" Zaremba added: "The decision was quite hard for me. But I got the realization some time ago that...that's for the best."

The group's best-known demo, an October 2019 five-fingered hand that solved a Rubik's Cube with a 20-60% success rate, is now orphaned. Staff are being folded into pretraining projects such as GPT-3, DALL-E, and Codex.