a vision-language model turning a demo clip into a robot task spec names the task right but puts poses 400 to 480 mm out, past the arm's reach, and adds 5 to 8 filler steps and extra objects.
From the Grok pain-radar digest, Oct 25-26 2026 (Oct 25 daytime to Oct 26 morning slots). Verified live 2026-10-26: issue/PR read on GitHub; title and state confirmed.