• The real bottleneck for code agents is the hand-off from "find the data" to "write the code." CODA-BENCH drops 1,009 tasks into a Linux sandbox averaging 980 files each, and the strongest system clears only 61.1% — a blind spot you never see when you test code and data separately.

• Reasoning budget can be allocated like a resource. AVA-VLA lets a robot model reason internally with latent variables, then exits early by confidence — 98.3% success on LIBERO and 6x faster than explicit chain-of-thought.

• Diffusion training was missing a whole-trajectory consistency constraint. This ICML work imports temporal-difference learning from RL as a drop-in objective; it improves FID, and the fewer the sampling steps, the bigger the gain.

• Offline and multi-objective is the real constraint behind a lot of tuning and design search. DOMOO bakes diversity into the optimization itself with nested Pareto set learning, instead of cherry-picking solutions afterward.

Also Notable

• Google Builds a Transparent, Reproducible Benchmark for Dual-Network PINN Optimal Control on a Classic Mass-Spring-Damper System — it pits physics-informed neural nets against traditional methods, so anyone wanting to check PINN reliability should look.
• DDIM Is Fast for Diffusion Inversion but Accumulates Error; This Work Cuts Inversion Error by Reordering Timesteps — relevant if you do image reconstruction or editing.

Read the full edition →