Tag
Introduces CICL, a decision-aware context layer that selects and compresses evidence for tool-using LLM agents by treating context as a decision-time intervention, using counterfactual-inspired scoring and typed memory cards under a token budget. Experiments on SWE-bench and RepoBench show concrete gains in retrieval accuracy and action criticality.
Aurora is an agentic video editing framework that pairs a tool-augmented vision-language model agent with a diffusion transformer to automatically resolve textual and visual underspecification in user requests, enabling unified video editing tasks like replacement, removal, style transfer, and reference-driven insertion.