让大模型读文档时自动发现知识，靠问答验证真理解。
Autolearn: Learn by Surprise, Commit by Proof
- 用问答链自验证异常文本，按信念强度调整训练
- 问答格式使理解度指标下降0.106，远超标准微调
- 可自动跳过已学内容，适合长期阅读学习场景
我们提出Autolearn框架，使语言模型在无外部监督下从阅读文档中学习。当段落产生异常高的单标记损失时被标记，通过自生成的问答链验证后，以信念比例调整β₂进行训练。引入扰动差距（重述与原文困惑度比）作为区分记忆与理解的指标。关键在于训练数据格式：问答格式使扰动差距降至预训练基线以下（2.098 vs. 2.204，Δ= -0.106，>10σ），抑制了词元序列记忆；而标准微调最佳表现仅在噪声水平内（Δ= -0.010，<1σ）。在覆盖Qwen3和Phi-4系列的四个模型中，只有Autolearn进入此状态。随机评估显示，段落级知识获取概率从6%升至54%（p < 10⁻⁴），问答格式在生成真正新事实上优于标准微调。系统具有自终止性：已学内容降低意外感，重遇时将被跳过。
原文摘要 · Abstract (English)
We propose Autolearn, a framework that enables language models to learn from documents they read, with no external supervision. Passages that produce anomalously high per-token loss are flagged, verified through a self-generated Q&A chain, and trained on with conviction-proportional $β_2$ adjustment. We introduce the perturbation gap (paraphrase-to-original perplexity ratio) as a metric that distinguishes memorization from understanding. The key mechanism is the training data format: Q&A-format training drives the perturbation gap below the pre-trained baseline (2.098 vs. 2.204, $Δ= -0.106$, $> 10σ$), suppressing token-sequence memorization, while standard fine-tuning's best attempt remains within noise ($Δ= -0.010$, $< 1σ$). Across four models spanning Qwen3 and Phi-4 families, Autolearn is the only method that enters this regime. Stochastic evaluation reveals passage-specific knowledge acquisition: the probability of generating a correct novel fact rises from 6% to 54% after training ($p < 10^{-4}$), and Q&A format outperforms standard fine-tuning on genuinely novel facts. The system is self-extinguishing: learned content reduces surprisal below threshold and is skipped on re-encounter.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。