LifeSearch: Benchmarking Deep Research Agents for Real-World Life and Productivity

Key points

• Deep research (DR) agents address open-ended information needs through multi-step reasoning, long-horizon web search, and evidence synthesis.1
• We conduct extensive experiments on 12 advanced DR agents under a unified evaluation harness, systematically analyzing their performance patterns and failure modes.1
• We select the grading configuration by comparing candidate LLM judges against human rubric labels, and a human ranking study shows strong alignment between aggregate benchmark scores and human judgments (Spearman's $ρ= 0.89$, Kendall's $τ= 0.81$).1
• LifeSearch provides a rigorous evaluation framework and foundation for advancing DR agents toward real-world life and productivity.1

Sources (2)

• [1]LifeSearch: Benchmarking Deep Research Agents for Real-World Life and Productivity
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 13, 10:12 AM

  To bridge this gap, we introduce LifeSearch, a challenging benchmark comprising 203 Chinese and English tasks derived from real-user query seeds across 7 concrete life and productivity domains.

  Deep research (DR) agents address open-ended information needs through multi-step reasoning, long-horizon web search, and evidence synthesis.

• [2]LifeSearch: Benchmarking Deep Research Agents for Real-World Life and Productivity
  Hugging Face Daily Papers · Oct 13, 12:00 AM · same content

Extractive summary: sentences quoted from the sources.