单词学习：![[1df2e958-592c-437c-b512-9a166102ddba.mp3]]

新闻播报：![[218a05f8-9c93-4db0-b395-2af28bc0f1db.mp3]]

普利策奖得主领衔，众作家起诉六大AI巨头盗版侵权

简报：

• 由两届普利策奖得主约翰·卡雷鲁领衔的数十位作家，已向美国法院提起集体诉讼，指控OpenAI、谷歌、Meta等六大AI公司。
• 诉讼指控这些公司系统性地使用从“影子图书馆”（如LibGen、Z-Library）获取的盗版书籍来训练其大语言模型，构成“故意版权侵权”。
• 作家方认为，其作品是AI模型语言能力、知识深度和叙事风格的基石，但未获得任何报酬。
• 若罪名成立，每部侵权作品最高可索赔15万美元，总赔偿金额预计将达到数十亿乃至上百亿美元。
• 该案的判决结果可能为AI训练数据的合法性确立全国性先例，并可能重构AI行业的数据供应链。

相关链接：

• 普利策得主领衔起诉六大AI巨头！集体诉讼直指“盗版书训练模型”，索赔或达百亿级

Pulitzer Prize Winner Leads, Writers Sue Six AI Giants for Copyright Infringement ⚖️

Brief:

• Dozens of writers, led by two-time Pulitzer Prize winner John Carreyrou, have filed a class-action lawsuit in a US court, accusing six major AI companies, including OpenAI, Google, and Meta.
• The lawsuit alleges that these companies systematically used pirated books obtained from 'shadow libraries' (such as LibGen and Z-Library) to train their large language models, constituting 'willful copyright infringement'.
• The writers argue that their works are the cornerstone of AI models' language capabilities, depth of knowledge, and narrative styles, but they have not received any compensation.
• If found guilty, each infringed work could demand up to dJyH6kDkfC5RlM16.answerText50,000 in damages, with total compensation potentially reaching billions or even tens of billions of US dollars. 💰
• The outcome of this case could establish a nationwide precedent for the legality of AI training data and potentially reshape the AI industry's data supply chain. 🚀

Related Links:

• Pulitzer Winner Leads Lawsuit Against Six AI Giants! Class Action Targets 'Pirated Book Training Models', Demands Could Reach Tens of Billions

MiniMax发布M2.1开源模型，编程与办公能力显著提升

简报：

• MiniMax于12月23日为其M2开源模型推出M2.1版本更新，致力于提升模型在真实世界复杂任务中的表现。
• 新版本系统性提升了对Rust、Java、Golang、C++、JavaScript等多种非Python语言的编程能力，并加强了原生Android/iOS开发能力。
• 模型升级了对“复合指令约束”的整合执行能力，以提高在真实办公场景中的可用性。
• 通过优化，模型的回复及思维链更加简洁，显著提升了响应速度并降低了Token消耗。
• 在多语言SWE-bench等基准测试中，MiniMax-M2.1的表现已接近Claude Opus 4.5。

相关链接：

• MiniMax 更新 M2.1 模型，重点聚焦于更多编程语言和办公场景的可用性

MiniMax Unveils M2.1 Open-Source Model: Significant Boost in Programming and Office Capabilities

Brief:

• On December 23, MiniMax launched the M2.1 update for its M2 open-source model, aiming to enhance the model's performance in complex real-world tasks. 🚀
• The new version systematically improves programming capabilities for various non-Python languages such as Rust, Java, Golang, C++, and JavaScript, while also strengthening native Android/iOS development. 💻
• The model has upgraded its integrated execution capabilities for "compound instruction constraints" to improve usability in real-world office scenarios.
• Through optimization, the model's responses and chain-of-thought are more concise, significantly boosting response speed and reducing token consumption. ⚡
• In benchmarks like the multilingual SWE-bench, MiniMax-M2.1's performance is approaching that of Claude Opus 4.5.

Related Link:

• MiniMax updates M2.1 model, focusing on usability in more programming languages and office scenarios

HBM3E内存罕见涨价，三星与SK海力士涨幅近两成

简报：

• 据韩媒报道，三星电子、SK海力士等存储供应商已决定上调其下一年度的HBM3E内存价格。
• 此次价格涨幅接近20%，此举在业内被认为是罕见的，因为供应商通常会在新一代产品推出前降低前代产品的价格。

相关链接：

• 三星、SK海力士上调明年HBM3E价格，涨幅近20%

HBM3E Memory Sees Rare Price Hike, Samsung and SK Hynix Increase Prices by Nearly 20% 📈

Brief:

• According to Korean media reports, storage suppliers such as Samsung Electronics and SK Hynix have decided to increase the price of their HBM3E memory for the upcoming year.
• The price increase is nearly 20%, a move considered rare in the industry 😮, as suppliers typically lower the prices of previous-generation products before new ones are launched.

Related Links:

• Samsung, SK Hynix Increase HBM3E Prices by Nearly 20% for Next Year

英伟达计划农历新年前开始对华出口H200芯片

简报：

• 据路透社援引三位知情人士消息，在美国政府批准后，英伟达已通知中国客户，计划于农历新年前（约2月中旬）开始出货其H200 AI芯片。
• 此举是美国政府的一项重大政策转向。本月，特朗普政府宣布同意以征收25%费用的方式，开放H200对华销售。
• 消息称，英伟达计划先以现有库存满足初步订单，预计出货5000至1万个芯片模组，约合4万至8万颗H200 AI芯片。
• 该计划仍存在不确定性，因北京当局尚未正式批准任何H200采购案，最终时程可能因政府决策而变动。

相关链接：

• 川普允「H200晶片」銷中國！《路透》：輝達擬農曆年前開始出貨

Nvidia Plans to Begin Exporting H200 Chips to China Before Lunar New Year

Brief:

• Nvidia has informed its Chinese customers that it plans to begin shipping its H200 AI chips before the Lunar New Year (around mid-February), following approval from the U.S. government, according to three sources cited by Reuters. 🚀
• This move marks a significant policy shift by the U.S. government. This month, the Trump administration announced its agreement to allow H200 sales to China, albeit with a 25% tariff. 💸
• Sources indicate that Nvidia plans to fulfill initial orders using existing inventory, with an estimated shipment of 5,000 to 10,000 chip modules, equivalent to approximately 40,000 to 80,000 H200 AI chips.
• The plan still faces uncertainties, as Beijing authorities have not yet formally approved any H200 purchase orders, and the final timeline could change due to government decisions. ⏳

Related Link:

• Trump allows 'H200 chips' to be sold to China! Reuters: Nvidia plans to start shipments before Lunar New Year

阿里通义开源语音交互模型，实现低延迟与情感感知

简报：

• 阿里巴巴通义实验室正式开源新一代端到端语音交互大模型Fun-Audio-Chat-8B。
• 该模型采用Speech-to-Speech（S2S）架构，旨在解决传统多模块拼接带来的延迟问题，实现超低延迟的实时交互。
• 模型具备情感感知能力，能通过分析用户语音的语气、语速等细节来理解情绪，并支持通过自然语音指令执行任务的语音函数调用（Voice Function Calling）功能。
• 在OpenAudioBench、MMAU等多项国际基准测试中，Fun-Audio-Chat-8B在同尺寸模型中排名第一。
• 此次开源内容包括完整的8B模型权重、推理代码和Function Call示例，已在GitHub、Hugging Face和ModelScope平台发布。

相关链接：

• 阿里通义开源语音交互大模型Fun-Audio-Chat-8B！超低延迟，能读懂情绪
• Fun-Audio-Chat项目地址

Alibaba Tongyi Open-Sources Speech Interaction Model, Achieving Low Latency and Emotional Perception

Briefing:

• Alibaba Tongyi Lab has officially open-sourced its new generation end-to-end speech interaction large model, Fun-Audio-Chat-8B.
• The model adopts a Speech-to-Speech (S2S) architecture, aiming to solve the latency issues caused by traditional multi-module concatenation and achieve ultra-low latency real-time interaction. 🚀
• The model possesses emotional perception capabilities, understanding emotions by analyzing details like user voice tone and speech rate. It also supports Voice Function Calling, which allows tasks to be executed via natural speech commands. 💬
• Fun-Audio-Chat-8B ranked first among models of the same size in multiple international benchmarks, including OpenAudioBench and MMAU. 🥇
• The open-source release includes the complete 8B model weights, inference code, and Function Call examples, published on GitHub, Hugging Face, and ModelScope platforms.

Related Links:

• Alibaba Tongyi Open-Sources Speech Interaction Large Model Fun-Audio-Chat-8B! Ultra-low Latency, Understands Emotions
• Fun-Audio-Chat Project Page

字节跳动新数学模型，IMO模拟赛达金牌线

简报：

• 字节跳动Seed团队发布了新一代形式化数学推理专用模型Seed Prover 1.5，通过大规模训练显著提升了推理能力与效率。
• 在针对IMO 2025的模拟测试中，该模型取得了35/42的成绩，达到了国际数学奥林匹克竞赛（IMO）评分标准的金牌分数线。
• 在北美本科级别数学竞赛Putnam 2025的模拟中，模型在9小时内解决了12道题中的11道。
• 该模型在Putnam历史评估集、Fate-H（硕士难度）和Fate-X（博士生难度）等多个数学推理评测集上刷新了最好成绩（SOTA）。

相关链接：

• 字节跳动发布新一代形式化数学推理专用模型 Seed Prover 1.5
• 阿里上线“千问智学”，字节推“AnyGen”
  应用正式接管你的书包与办公桌

ByteDance's New Math Model Reaches IMO Gold Medal Threshold in Mock Competition 🥇

Briefing:

• ByteDance's Seed team has released Seed Prover 1.5, a new generation model dedicated to formal mathematical reasoning, significantly enhancing its reasoning capabilities and efficiency through large-scale training.
• In a mock test for IMO 2025, the model scored 35/42, reaching the gold medal threshold of the International Mathematical Olympiad (IMO) scoring standards. 🏆
• In a simulation of the North American undergraduate-level math competition Putnam 2025, the model solved 11 out of 12 problems within 9 hours.
• The model has set new state-of-the-art (SOTA) records on multiple mathematical reasoning evaluation sets, including Putnam historical evaluation sets, Fate-H (Master's difficulty), and Fate-X (PhD candidate difficulty). 📈

Related Links:

• ByteDance Releases New Generation Formal Mathematical Reasoning Model Seed Prover 1.5
• Alibaba Launches "Qianwen Zhixue," ByteDance Promotes "AnyGen": AI Applications Officially Take Over Your School Bag and Office Desk

面壁智能获数亿元新融资，端侧大模型已落地多家车企

简报：

• 端侧AI公司面壁智能宣布完成数亿元新一轮融资，由京国瑞、国科投资、中金保时捷基金、米聚资本与和基投资联合参投。
• 本轮融资将用于推进端侧高效大模型的研发与生态建设。
• 其核心产品MiniCPM“面壁小钢炮”系列端侧模型，已在智能手机、PC、智能家居及智能汽车场景实现商用。
• 在汽车领域，公司已与长安马自达、吉利、大众汽车达成合作，搭载其模型的MAZDA EZ-60和吉利银河M9已量产上市。
• 公司技术路线为“模型小型化、推理本地化”，使AI模型可在终端设备上离线运行，实现毫秒级响应并保障数据隐私。

相关链接：

• 面壁智能完成数亿元融资！端侧大模型加速落地，吉利、长安、大众已上车

Minibox AI Secures Hundreds of Millions in New Funding, Edge AI Models Deployed in Multiple Automakers

Brief:

• Edge AI company Minibox AI announced the completion of a new round of funding totaling hundreds of millions of RMB, jointly invested by Jingguorui, Guoke Investment, CICC Porsche Fund, Mijuli Capital, and Heji Investment. 💰
• This round of financing will be used to advance the research and development of efficient edge large models and ecosystem construction. 🚀
• Its core product, the MiniCPM "Minibox AI Pocket Rocket" series of edge models, has achieved commercial application in smartphones, PCs, smart homes, and smart car scenarios.
• In the automotive sector, the company has partnered with Changan Mazda, Geely, and Volkswagen, with the MAZDA EZ-60 and Geely Galaxy M9, featuring its models, already mass-produced and launched.
• The company's technical approach focuses on "model miniaturization and localized inference," enabling AI models to run offline on terminal devices, achieving millisecond-level responses and ensuring data privacy. 🔒

Related Link:

• Minibox AI Completes Hundreds of Millions in Funding! Edge Large Models Accelerate Deployment, Geely, Changan, Volkswagen Already Onboard