Tag
The paper introduces KNOWS, a benchmark for evaluating web agents on complex, long-horizon tasks that involve synthesizing and organizing knowledge into artifacts, revealing that current agents struggle with visual steps and long-horizon reasoning.
This article introduces the Stanford STORM method, which uses four prompts to achieve multi-perspective AI-assisted research, producing high-quality research briefs and significantly improving research depth and efficiency.
This paper introduces S1-DeepResearch-32B, an open-source model and 15K trajectory dataset for deep research agents, achieving state-of-the-art performance across 20 benchmarks by jointly modeling information acquisition, knowledge synthesis, and planning.
Google quietly enabled Gemini to ingest and reason over NotebookLM notebooks, letting users fuse private, cited documents with live web search for faster, grounded research and creative workflows.