新的SFT方法可减少语言模型的幻觉

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-01 04:00

研究人员开发了一种称为知识对齐SFT的方法，以减少语言模型中的事实幻觉。该方法将训练目标限制在模型现有的参数知识范围内，这与可能使用外部知识的标准SFT形成对比。引入了新的变体Evidence Rewrite和Recall Rewrite，并在Qwen 3 4B和OLMo 3 7B等模型上进行了测试。结果表明，知识对齐SFT可以提高事实准确性并保留通用能力，其中Recall Rewrite在事实准确性和拒绝行为方面显示出最显著的提升。 AI

影响这项研究通过改进微调过程，有可能带来更具事实准确性的语言模型。

排序理由该集群包含一篇研究论文，详细介绍了语言模型监督微调的新方法。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 25 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇研究论文，详细介绍了语言模型监督微调的新方法。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, model release; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.AI TIER_1 English(EN) · Arthur Becker, Jakob Kemmler, David Thulke, Christine Sch\"afer, Christian Dugast, Hermann Ney · 2026-09-01 04:00

坚守所知：一项关于知识对齐监督微调的研究

arXiv:2608.30987v1 Announce Type: cross Abstract: Supervised fine-tuning (SFT) trains a base language model to imitate target responses, and these targets may require knowledge the base model has not robustly internalized. We study this as a source of hallucinations and frame a g…

本文如何被排名

报道来源 [1]

坚守所知：一项关于知识对齐监督微调的研究

相关实体

相关话题