1
Towards Understanding Steering Strength
ICML 2026论文深入解析模型引导技术中的强度量化问题，为AI可解释性提供新视角。
arXiv:2602.02712v2 Announce Type: replace Abstract: A popular approach to post-training control of large language models (LLMs) is the steering of int…
2
Understanding How MLLMs Describe Artworks Using Token Activation Maps
揭示多模态大模型描述艺术品时的视觉推理机制，用token激活映射解读模型如何“看”画。
arXiv:2606.27947v1 Announce Type: new Abstract: Multimodal Large Language Models (MLLMs) describe artworks with remarkable fluency, yet the visual rea…
3
Vis-CoT: A Human-in-the-Loop Framework for Interactive Visualization and Intervention in LLM Chain-of-Thought Reasoning
人机协同干预LLM推理过程，可视化思维链让AI思考透明可控，复杂推理任务更可靠。
arXiv:2509.01412v3 Announce Type: replace Abstract: Large language models (LLMs) show strong reasoning via chain-of-thought (CoT) prompting, but the p…
4
GEOPHYS: The Geometry of Physical Plausibility
最新arXiv论文提出GEOPHYS框架，用几何方法量化物理合理性，为AI生成内容提供可解释性评估。
arXiv:2606.20707v1 Announce Type: cross Abstract: While humans can identify physically implausible events within milliseconds, machine learning approa…
5
Beyond Memorization: Distinguishing Between Pattern-Based and Epistemic Reasoning in LLMs Using Epistemic Puzzles
用认知谜题测试LLM推理能力，揭示模型是真正理解还是仅靠模式匹配
arXiv:2603.21350v2 Announce Type: replace Abstract: Epistemic reasoning requires agents to infer the state of the world from partial observations and …
6
XtrAIn: Training-Guided Occlusion for Feature Attribution
基于训练引导的遮挡归因方法，解决特征重要性评估中基线选择的可靠性难题
arXiv:2606.10877v1 Announce Type: new Abstract: Occlusion-based attribution methods provide an intuitive way to estimate feature importance by perturb…
7
Enhancing AI Interpretability and Safety through Localised Architectures
提出通过本地化架构的设计思路，提升AI系统的可解释性与安全防护能力。
arXiv:2606.07998v1 Announce Type: new Abstract: Recent advances in generative AI, especially powerful Large Language Models (LLMs) and Large Reasoning…
8
Thinking Through Signs: PEEL as a Semiotic Scaffolding for Epistemically Accountable AI-Enabled Research
用符号学框架PEEL为AI研究注入认知问责，让机器思考更透明可信。
arXiv:2606.04152v1 Announce Type: new Abstract: Large language models are reshaping research practice while quietly eroding researchers epistemic acco…
9
Encoded but Not Routed: Explaining the Table-Chart Gap in Scientific Claim Verification
揭秘表格与图表在科学声明验证中的表现差异，AI模型为何"编码但不路由"？附论文全文。
arXiv:2606.01679v1 Announce Type: new Abstract: Multimodal LLMs are increasingly used to assist scientific peer review, where a core requirement is ve…
10
From Features to Actions: Explainability in Traditional and Agentic AI Systems
从特征到行动，揭示传统AI与智能体系统可解释性的前沿对比与演进。
arXiv:2602.06841v4 Announce Type: replace Abstract: Over the last decade, Explainable AI has primarily focused on interpreting individual model predic…
11
Fundamental Limitation in Explaining AI
最新论文揭示：现有AI解释方法存在根本性局限，无法完全忠实说明大模型行为，挑战可解释性研究基石。
arXiv:2605.24727v1 Announce Type: new Abstract: While large-scale models such as LLMs and diffusion models have achieved practical success, public ins…
12
AI Interpretability Is a Revolutionary Skill
深入探索AI语言模型的非语言内部世界，揭秘可解释性这一革命性技能。
Article URL: https://www.outcryai.com/research/the-dark-between-the-stars Comments URL: https://news.ycombinator.com/item?id=48263216 Points: 1 # Comm…
13
Stabilising Explainability Fragility in Cybersecurity AI: The Impact and Mitigation of Multicollinearity in Public Benchmark Datasets
研究发现多重共线性会严重削弱AI入侵检测的可解释性稳定性，并提出缓解策略，值得关注。
arXiv:2605.22529v1 Announce Type: cross Abstract: This paper investigates a unexplored yet impactful vulnerability in AI explainability used in intrus…
14
A Mechanistic Explanatory Strategy for XAI
提出一种机制性解释策略，为可解释人工智能提供全新理论框架。
arXiv:2411.01332v5 Announce Type: replace-cross Abstract: Despite significant advancements in XAI, scholars note a persistent lack of solid conceptual…