PhysOPSD: Code as Privilege for Physically Plausible Video Generation
PhysOPSD uses executable simulations as privileged context for video-model training, improving physical plausibility with 3,000 samples while preserving general video quality.
ls ./papers
Trending AI research papers from Hugging Face, summarized in plain language: what each paper does, why it matters and where to read more.
140 papers
PhysOPSD uses executable simulations as privileged context for video-model training, improving physical plausibility with 3,000 samples while preserving general video quality.
dNaViT tests whether one discrete visual vocabulary can support image understanding, reconstruction, generation, and editing without sacrificing semantic or fine-grained detail.
APEX uses condition shifting to share one network between streaming video generation and fake-distribution estimation, reducing training cost while improving instruction following.
World In-Context Learning lets frozen robot agents rehearse tasks in simulation, store semantic lessons in context, and transfer skills across scenes and embodiments.
JoVox improves lightweight voxel generation by jointly modeling coarse and fine hierarchy levels, revisiting early decisions, and reducing compute compared with TRELLIS.2.
AutoVend tests whether LLM agents can turn web-based vending decisions into sustained profit while managing shared inventory, delays, paid visits, and uncertainty.
Flide rethinks LLM decision serving as a shared state with isolated branches, reporting major speedups while preserving standalone probabilities and reproducibility.
ReproBridge turns hidden troubleshooting work into reusable ML artifacts, helping coding agents reproduce experiments more reliably, consistently, and at lower API cost.
WorldFoundry proposes shared infrastructure and a stricter definition of world models: persistent state, intervention-driven dynamics, and feedback-backed evaluation.
Sketchpad turns a small vision-language model into a certifiable arithmetic engine by limiting each step to a finite worksheet view, then checking every possible view.
HEELO turns tools, code, proof methods, and workflows into reusable agent-native primitives that apply resources in context and revise failures.
AgentBaton explores one-time handoffs from capable to efficient models, reporting major API cost reductions while preserving accuracy on long-horizon agent tasks.