Tag
A practitioner shares the architecture for trustworthy quoting agents: the LLM only extracts specs from messy customer messages at temperature 0, nulls become follow-up questions, pricing is done by a deterministic code engine with regression tests, and a human approves every quote before it goes out.
该研究评估了 Llama-3 8B、Qwen-2.5 14B 和 Llama-3.3 70B 等 LLM 在从 SEC 10-K 文件中提取财务数据时的表现，发现将模型规模与文档结构复杂度对齐可在零样本设置下获得高准确率，从而缓解量化金融数据集中超过 70% 公司面临的缺失数据偏差问题。
SCTab-Diff is a semantics-consistent tabular diffusion framework that uses weak semantic priors to generate high-fidelity synthetic tabular data, improving distributional fidelity and semantic consistency over existing methods.
The author conducted a small experiment on LlamaIndex's extractbench benchmark, comparing Claude Opus 5 and Qwen models, finding similar performance with cost advantages and a correlation between extraction difficulty and document length.
The author argues that agent memory layers should skip LLM-based extraction for deciding what to remember, instead using simple storage, embeddings, and retrieval, exemplified by their open-source memU tool.