揭秘大模型幻觉根源：MoRFI方法用单调稀疏自编码器精准锁定知识特征，为可解释性带来新突破。
arXiv:2604.26866v2 Announce Type: replace-cross Abstract: Large language models (LLMs) acquire most of their factual knowledge during the pre-training…
用稀疏自编码器拆解大模型思考与直接回答的神经差异，揭开推理黑箱的机制奥秘。
arXiv:2608.08168v1 Announce Type: new Abstract: While Large Language Models (LLMs) employing Chain-of-Thought (CoT) exhibit superior reasoning capabil…
稀疏自编码器新应用：高效防御LLM越狱攻击，来自ICML可解释性研讨会的前沿研究。
arXiv:2602.12418v2 Announce Type: replace-cross Abstract: Jailbreak attacks remain a persistent threat to large language model safety. We propose Cont…
稀疏自编码器揭开LLM情感推理黑箱，找出哪些情绪更难被模型识别及其因果机制。
arXiv:2604.25866v2 Announce Type: replace Abstract: Large language models (LLMs) are increasingly used in emotionally sensitive human-AI applications,…
用稀疏自编码器把大模型持续学习从“调权重”升级为“控特征”，让新知识不冲掉旧技能
arXiv:2606.26629v1 Announce Type: new Abstract: Weight-space regularization methods such as Elastic Weight Consolidation (EWC) are the standard approa…
基于论文探索蛋白质语言模型的生成机制，提出电路追踪方法，为理解模型内部原理提供新视角
arXiv:2606.16044v1 Announce Type: new Abstract: Protein language models (pLMs) can generate novel protein sequences with properties beyond those obser…
用激活引导偏好优化解释SAE特征，为稀疏自编码器的可解释性提供新方法。
arXiv:2606.08496v1 Announce Type: cross Abstract: Although Sparse Autoencoders (SAEs) have mitigated the opacity of large language models (LLMs) by de…
用稀疏自编码器实现语音情感的可解释控制，兼顾合成质量与情感表达能力。
arXiv:2606.01479v1 Announce Type: new Abstract: Integrating large language models (LLMs) into text-to-speech (TTS) systems has improved speech express…
提出步骤级稀疏自编码器，精确定位大模型推理中的关键中间表征，提升可解释性
arXiv:2603.03031v2 Announce Type: replace Abstract: Large Language Models (LLMs) have achieved strong complex reasoning capabilities through Chain-of-…
稀疏自编码器成功从Claude 3 Sonnet中提取3400万个可解释特征，验证了字典学习方法在大模型上的可扩展性。
arXiv:2605.29358v1 Announce Type: new Abstract: We demonstrate that sparse autoencoders can extract interpretable features from Claude 3 Sonnet, a pro…
用稀疏自编码器在LLM特征空间合成多样化数据，实现“少即是多”的高效数据扩充新方法。
arXiv:2602.10388v3 Announce Type: replace-cross Abstract: The diversity of post-training data is critical for effective downstream performance in larg…
利用稀疏自编码器挖掘模型内部信号，为LLM强化学习数据工程提供全新无监督指引。
arXiv:2605.27354v1 Announce Type: cross Abstract: Model internals encode rich information about how a large language model (LLM) processes its trainin…
稀疏自编码器首次建立大脑皮层语义拓扑与LLM特征的可解释映射，跨学科新突破
Article URL: https://letsdatascience.com/news/sparse-autoencoders-reveal-cortical-brain-llm-semantic-mappi-bc586635 Comments URL: https://news.ycombin…
揭秘稀疏自编码器如何揭示大脑与LLM的语义对齐模式，探索皮层拓扑结构中的智能映射
arXiv:2605.23035v1 Announce Type: cross Abstract: Intermediate layers of large language models (LLMs) best predict human brain responses to language, …
GPT-2小模型在间接宾语识别任务中的失败与成功，稀疏特征差异揭示模型内部行为，一次小而透明的审计。
arXiv:2605.22719v1 Announce Type: new Abstract: We report a small, reproducible audit of which sparse-autoencoder (SAE) features of GPT-2 small fire d…
揭示稀疏自编码器与主题模型的内在等价关系，为神经网络可解释性研究提供新视角。
arXiv:2511.16309v2 Announce Type: replace-cross Abstract: Sparse autoencoders (SAEs) are used to analyze embeddings, but their role and practical valu…
稀疏自编码器基准测试可信吗？这篇论文深入剖析了现有评估方法的漏洞，挑战技术共识。
arXiv:2605.18229v1 Announce Type: new Abstract: Sparse autoencoders (SAEs) are a core interpretability tool for large language models, and progress on…
OpenAI利用稀疏自编码器技术，从GPT-4内部计算中自动识别出1600万个隐含概念，揭示大模型可解释性新进展
Using new techniques for scaling sparse autoencoders, we automatically identified 16 million patterns in GPT-4's computations.
首个能编辑递归语言模型缓存写入的稀疏自编码器，突破残差SAE局限。
arXiv:2605.12770v2 Announce Type: replace-cross Abstract: We introduce WriteSAE, the first sparse autoencoder that decomposes and edits the matrix cac…