The Ability of Frontier Models to Control Robots is Rapidly Improving — LessWrong

A LessWrong post demonstrates that frontier multimodal language models are increasingly capable of directly controlling real robotic arms, with rapid performance improvements.