1
特斯拉 FSD 研发逻辑曝光：Cybercab 先用完整版，量产车运行精简版
特斯拉先为Cybercab研发完整FSD再“蒸馏”给HW4量产车，揭秘自动驾驶软件策略差异。
IT之家 7 月 15 日消息，据 Notateslaapp 报道，特斯拉正在推动不同代际车辆硬件实现“无需人工监督的自动驾驶”，这一过程也揭示了其背后的一些软件秘密。 目前，搭载新一代 Hardware 4（HW4/AI4）计算机的消费者车辆承担了大部分公开道路测试和实际驾驶任务。但据悉，驱动这些…
2
Knowledge Distillation for Automated AI Tutor Evaluation
用知识蒸馏技术自动化评估AI导师的教学质量，FATE方法为教育大模型评价提供新思路
arXiv:2607.10647v1 Announce Type: new Abstract: The rapid integration of Large Language Models (LLMs) into K-12 and higher education has outpaced the …
3
Reference-Based Distillation Detection in LLMs
揭秘LLM模型蒸馏检测新方法，从学生模型中溯源教师模型，保障AI合规与公平竞争。
arXiv:2607.09692v1 Announce Type: new Abstract: Model distillation -- training on outputs from stronger third-party models -- is widely used to boost …
4
微软 CEO 纳德拉警告企业：你花钱用的 AI，最终可能会成为你的竞争对手
微软CEO纳德拉警示：企业付费使用的AI可能正在“掏空”自身知识，模型蒸馏风险让巨头稳坐钓鱼台。
IT之家 7 月 14 日消息，据外媒 TechCrunch 今天（14 日）报道，硅谷围绕人工智能风险的争论持续无休，最让 AI 拥护者坐立不安的，却是大型 AI 实验室可能借专有模型充当“特洛伊木马”。 初创企业和大型企业把敏感资料喂给模型后，模型提供商便能 逐渐摸清客户的业务 ，还可能 利用积…
5
微软 CEO 纳德拉批评 AI 模型公司“双标”：一边主张“合理使用”数据，一边限制他人蒸馏
微软CEO纳德拉怒批AI公司“双标”：一边白嫖公开数据，一边严防模型蒸馏
IT之家 7 月 13 日消息，微软 CEO 萨提亚 · 纳德拉（Satya Nadella）日前在 X 平台发文，对包括 Anthropic 在内的 AI 大模型公司训练模型的做法进行了含蓄批评。 纳德拉表示，一些模型厂商一边主张拥有利用公开数据训练 AI 模型的“合理使用（Fair Use）”权…
6
Inside the secret AI war between Silicon Valley and China
揭露硅谷指控中国公司“蒸馏”AI知识的内幕，一场悄无声息的技术冷战正在开打。
Article URL: https://www.washingtonpost.com/national-security/2026/07/06/why-anthropic-alleges-chinese-firms-are-distilling-knowledge-claude/ Comments…
7
Compete Then Collaborate: Frontier AI Teachers Build a Verifiable Curriculum to Improve a Coding Student Beyond Imitation
让前沿AI老师互相竞技，再合作打造可验证课程，帮助学生超越模仿式学习。
arXiv:2607.08255v1 Announce Type: new Abstract: Large language models increasingly serve as teachers generating training data for smaller students. Pr…
8
Toward Efficient Uncertainty in LLMs through Evidential Knowledge Distillation
新方法通过证据知识蒸馏让大语言模型高效量化不确定性，无需额外推理开销
arXiv:2507.18366v2 Announce Type: replace Abstract: Accurate uncertainty quantification remains a key challenge for standard LLMs, prompting the adopt…
9
Token-level Response-visual Attention Guidance for Multimodal LLMs Knowledge Distillation
针对多模态大语言模型压缩难题，提出Token级响应-视觉注意力引导，提升蒸馏效果
arXiv:2607.02593v1 Announce Type: cross Abstract: While knowledge distillation (KD) is widely adopted for training lightweight models by leveraging su…
10
Auto: The AGI Compiler
让LLM agent告别昂贵低效：Auto编译器将实时行为中确定性部分提取为验证程序，实现智能体行为的精准编译与加速。
arXiv:2607.04542v1 Announce Type: cross Abstract: Every LLM agent run re-derives its behavior token by token on a frontier model: brilliant, expensive…
11
Teacher Supervision over Representation Equivalence Classes
颠覆知识蒸馏常规认知：预训练表征只有等价类意义，匹配坐标是伪命题
arXiv:2607.03572v1 Announce Type: cross Abstract: Knowledge distillation is usually framed as a choice of what to match in the teacher - its logits, h…
12
Effective Distillation to Hybrid xLSTM Architectures
将知识蒸馏引入混合xLSTM架构，探索高效模型压缩新方向
arXiv:2603.15590v2 Announce Type: replace Abstract: There have been numerous attempts to distill quadratic attention-based large language models (LLMs…
13
Trust Region Policy Distillation
将信任区域优化引入策略蒸馏，解决模型压缩中的策略漂移问题。
arXiv:2607.04751v1 Announce Type: cross Abstract: Big goals are hard to achieve all at once; breaking them into small steps is wiser. We present Trust…
14
Alibaba reportedly bans employees from using Claude Code
阿里巴巴禁止员工用Claude Code，防账户滥用和模型蒸馏，科技巨头对AI工具严加管控。
Alibaba has reportedly classified Claude Code as high-risk software.
15
DemoPSD: Disagreement-Modulated Policy Self-Distillation
提出一种基于分歧调度的策略自蒸馏方法，有效提升大模型推理训练效果。
arXiv:2607.02502v1 Announce Type: cross Abstract: On-policy self-distillation (OPSD) has emerged as a practical method for training large language mod…
16
Neuron-Aware Data Selection for Annotation-Free LLM Self-Distillation
无需人工标注，通过神经元激活模式筛选数据，实现LLM高效自蒸馏训练。
arXiv:2607.02460v1 Announce Type: cross Abstract: Post-training large language models (LLMs) without real-world interaction feedback or human-labeled …
17
消息称阿里巴巴全面禁用 Anthropic 旗下 Claude 产品：7 月 10 日生效，全员卸载
阿里全面禁用Claude，7月10日生效全员卸载，背后涉及模型蒸馏攻击防范与商业博弈。
IT之家 7 月 3 日消息，据智东西今日消息，阿里巴巴内部宣布全面禁用 Claude，全体员工被要求卸载 Anthropic 旗下产品， 涵盖 Sonnet、Opus、Fable 等多个模型，以及 Claude Code 在内的 Agent 产品 ，7 月 10 日正式生效。 据消息人士透露，自今…
18
TallyTrain: Communication-Efficient Federated Distillation
提出一种压缩模型大小与类别数双重带宽瓶颈的联邦蒸馏方法，大幅提升通信效率。
arXiv:2607.00173v1 Announce Type: new Abstract: Federated learning is bandwidth-bound on two orthogonal axes: model size, which limits how often param…
19
Building Multi-Task Agentic LLMs via Two-Phase Distillation
两阶段蒸馏法让LLM成为多任务智能代理，训练效率与推理能力双提升
arXiv:2606.30044v1 Announce Type: new Abstract: A key step toward artificial general intelligence is to train models that can perform multiple tasks. …
20
MOPD: Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Training
提出多教师在线策略蒸馏框架，高效整合多个大模型能力，优于传统方法
arXiv:2606.30406v1 Announce Type: cross Abstract: Modern large language models (LLMs) rely on reinforcement learning during post-training to push spec…