Large Language Models (LLMs) like GPT and Llama have sparked a revolution in the field of artificial intelligence, but efficiently training these massive models while aligning them with human values remains a challenge. Reinforcement Learning with Human Feedback (RLHF) has become an important training method for LLMs in recent years, but traditional RLHF frameworks have limitations in flexibility, efficiency, and scalability. To address these issues, the ByteDance Doubao Large Model Team has open-sourced an RLHF framework called HybridFlow.

With the development of AI technology, many difficult problems in learning scenarios are being solved efficiently. On November 1, Quark fully upgraded its 'AI Question Search' product, making the process of searching and solving problems faster and more capable. Users can find various new and difficult questions through their smartphones, tablets, and computers, and the AI will provide step-by-step explanations, allowing users to ask questions at any time. After launching the 'One-Stop AI Services' and 'System-Level All-Scenario AI Capabilities', Quark's AI capabilities are being applied to learning scenarios, helping users enhance their learning abilities and efficiency.

On October 30, Mianbi Intelligence announced a strategic partnership with Baidu Intelligent Cloud, aiming to jointly develop collaborative solutions for large models. According to the partnership agreement, the two parties will leverage Baidu Intelligent Cloud's Qianfan Large Model Platform, Wenxin series large models, as well as Mianbi's MiniCPM series edge large models and cutting-edge technology accumulation across multiple terminals and different scenarios, to collaboratively develop and promote edge-cloud collaborative technology solutions that further enhance the performance of large models in practical applications such as embodied intelligence, smart terminals, and edge computing, while reducing the inference costs of large models.

Is AI actually starting to ‘get smarter’?! The latest research from MIT indicates that the internal structure of large language models (LLMs) astonishingly resembles that of the human brain! This study utilized sparse autoencoder techniques to conduct an in-depth analysis of the activation space of LLMs, revealing three hierarchical structural features that are truly remarkable: first, at the microscopic level, researchers discovered the existence of structures akin to ‘crystals’. The surfaces of these ‘crystals’ are composed of parallelograms or trapezoids, similar to familiar lexical analogies such as ‘man:’

Beijing Aishi Technology Co., Ltd. has announced the global release of its latest version of the video generation large model, PixVerse V3. This version has made significant improvements in its underlying model capabilities, offering multimodal generation abilities and a wealth of AI features, including creative templates, lip synchronization, story continuation, and style transformation.

The Beijing ZhiYuan Artificial Intelligence Research Institute, in collaboration with Shanghai Jiao Tong University, Renmin University of China, Peking University, and Beijing University of Posts and Telecommunications, has launched an ultra-long video understanding model named Video-XL. This model is an important demonstration of the core capabilities of multimodal large models and a key step towards General Artificial Intelligence (AGI). Compared to existing multimodal large models, Video-XL shows superior performance and efficiency in processing long videos exceeding 10 minutes.

On October 24, Xpeng Motors held the 'Xpeng P7+ AI Driving Technology Sharing Conference and the Launch of AI Tianji 5.4.0' in Guangzhou, announcing that the P7+ and subsequent models will come standard with AI advanced smart driving without the need for optional purchases, subscriptions, or additional fees. Dr. Li Liyun, Vice President and Head of Autonomous Driving at Xpeng Motors, emphasized that building a cloud-based large model is the key to success in the smart driving competition. Xpeng Motors chose the same path as OpenAI from the research phase, which is to first build a powerful 'cloud-based large model'.

On October 22, Baidu released the top ten frontier technology inventions of 2024 in Beijing, highlighting multiple innovative native applications of artificial intelligence and high-value patent outcomes. Among these frontier technological inventions, 'large models' emerged as a prominent keyword, showcasing Baidu's in-depth research and application in this field.

Tencent has recently launched a smart workbench product called ima.copilot (shortened to ima), which is powered by Tencent's Hunyuan large model technology. This product aims to provide users with a new experience in searching, reading, and writing through its thinking knowledge base. The core features of ima.copilot include knowledge acquisition, the creation of a personalized knowledge base, and intelligent writing assistance.

According to media reports, Tan Xu, former Chief Research Manager at Microsoft Research Asia, joined the startup Dark Side of the Moon in August, primarily responsible for developing end-to-end speech models. One important goal after Tan's joining is to create a speech experience similar to GPT-4o for the company. Dark Side of the Moon has not yet responded, but insiders have confirmed Tan's employment, and specific position details have not been finalized. Public information indicates that Tan Xu's research areas include deep learning, language/speech/music processing, and AI content generation, with publications in AI academic conferences.

Dow Jones, the parent company of The Wall Street Journal, and the New York Post are suing AI-driven search startup Perplexity for using their news content to train its large language model. The publications from these two news groups accuse Perplexity of copyright infringement by generating answers to people's queries using their articles, thereby diverting traffic from the publications' websites. 'This lawsuit is brought by news publishers seeking compensation as Perplexity openly plans to provide services for free.'

Recently, ByteDance officially addressed rumors regarding an 'intern disrupting large model training'. The company confirmed that an intern did maliciously interfere with the model training tasks of the commercialization technology team's research project. However, this action did not affect the company's official commercial projects and online business, nor did it involve other large model businesses of ByteDance. Moreover, ByteDance pointed out that claims circulating online about 'involving over 8,000 nodes and causing losses of over ten million dollars' are seriously exaggerated. According to public reports, the intern was part of ByteDance's commercialization technology team.