Hidden-target projection minimizes regret for online inventory optimization on general convex sets, improving dependence on common-demand probability to inverse square root with matching lower bound, plus polylogarithmic and adaptive dynamic guarantees.
RIG-BENCH evaluates reasoning-driven image generation across four cognitive domains, revealing that state-of-the-art models produce locally plausible but globally illogical outputs.
OpenWebRL enables open online RL for visual web agents, with a 4B model reaching 67% Online-Mind2Web and 64% DeepShop success using minimal initialization data.