LesenForschungRadarAnlageframework
Anmelden / Registrieren
← Zurück zur Lektüre
NewsThema · 思维链延续证据遵循研究Wesentlich1 Min. LesezeitDisconfirmAI Redaktion

Preprint: Separate Answer Call Improves Evidence Following Over Reasoning Continuation

A new arXiv preprint finds that using a separate call for answering, rather than continuing the model's own reasoning trace, significantly increases adherence to current evidence.

2026-10-16 12:00 GMT+8

A preprint published on arXiv titled "When Reasoning Traces Conflict with Evidence: How Answer Configuration Shapes Evidence Following" investigates how large language models handle conflicts between their internal reasoning traces and external evidence. The study was conducted by Shaoang Li, Yanhang Shi, and Jian Li.

The researchers found that retaining a reasoning trace at the time of answering plays two distinct roles: it can improve accuracy, but it can also compete with current evidence when the two disagree. By testing three model families (gpt-oss, Gemma4, Qwen3) on HotpotQA and MuSiQue datasets, they discovered that using a separate answer call—rather than continuing the model's own thinking process—raises evidence-following by 29.9 to 60.6 percentage points.

Although removing the trace lowers accuracy in both configurations, the data shows that continuation often repeats stale conclusions in most conflict scenarios, whereas the separate call follows current evidence far more frequently. This suggests that benefiting from a trace and deferring to it are distinct properties.

The study recommends that because trace deletion alone cannot reveal which source an answer follows, evaluations should report both properties together with the answer configuration. This finding provides important reference for optimizing agent decision logic in Retrieval-Augmented Generation (RAG) systems.

Quellen

Anmelden / Registrieren
LesenForschungRadarAnlageframework
arxiv.org
↗
Thema

思维链延续证据遵循研究

【背景】大型语言模型(LLM)在生成答案时,其内部推理轨迹(思维链)可能与最新的外部证据发生冲突。一项由美国石溪大学(Stony Brook University)研究人员 Shaoang Li、Yanhang Shi 和 Jian Li …

2 Schritte0 BeiträgeAktualisiert 2026-10-23 21:06

Wie es dazu kam

  1. 2026-10-14Shaoang Li 等人提交预印本《When Reasoning Traces Conflict with Evidence》,提出通过交叉验证证据版本与思维链来分离准确率提升与证据遵循度。arxiv.org ↗
  2. 2026-10-16论文更新至 v2 版本,确认在 GPT-OSS、Gemma4 和 Qwen3 模型上,独立回答调用比延续思维链更能让 AI 遵循最新证据,差距达 29.9-60.6 分。arxiv.org ↗

Was noch ungeklärt ist

  • 在实际部署中,如何平衡‘独立回答调用’带来的证据遵循优势与‘思维链延续’带来的准确率损失?
  • 其他主流大模型(如 Claude, Gemini)是否表现出相同的‘过时结论重复’倾向?
  • 是否有更轻量级的干预方法能在不切断思维链的情况下提高模型对新证据的敏感度?
Vollständiges Thema lesen →
Recherche fortsetzen →