The open-weight coding crown just changed hands and Meta capped its own employees' AI spend -- but the real story is what AI is quietly automating underneath. We break down SkinTokens, which turns 3D character rigging into a token-prediction problem and roughly doubles skinning quality; Dockerless, which grades AI-written code by reading the repo instead of running its tests; and a pair of on-policy distillation papers that let models learn from their own mistakes and assemble one generalist from a roster of specialists. Three unglamorous grind steps, three automations, one theme: the win was a smarter representation, not a bigger model.

Chapters

0:00 Cold Open
0:52 The Headlines
4:18 Intro
5:23 Teaching AI to rig a 3D character
10:55 Grading AI code without running it
14:48 Models that learn from their own mistakes
19:19 Wrap-up

Links

Teaching AI to rig a 3D character -- https://arxiv.org/abs/2602.04805
Grading AI code without running it -- https://arxiv.org/abs/2606.28436
Models that learn from their own mistakes -- https://arxiv.org/abs/2606.30626