AI 导读
Benchmark: https://t.co/RgkyrHFZPk DiligenceBench 是一个股票研究智能体评测，衡量事实准确性、分析推理和风险意识，涵盖 150 项任务，横跨能源、银行、生物科技、保险、科技、REITs、餐饮、工业和公用事业。 该评测已经捕捉到一些推理时扩展：H1 → H3 为智能体提供了更好的工具和更丰富的工作流。但我认为我们需要更多评测来理解性能如何随算力、搜索和工具使用而扩展。
正文
Benchmark: https://t.co/RgkyrHFZPk
DiligenceBench is an equity-research agent eval measuring factual accuracy, analytical reasoning, and risk awareness across 150 tasks spanning energy, banking, biotech, insurance, technology, REITs, restaurants, industrials, and utilities.
The eval already captures some inference-time scaling: H1 → H3 gives agents better tools and richer workflows. But I think we need more evals to understand how performance scales with compute, search, and tool use.
来源：@karinanguyen · x.com