Core Insights - Jerry Tworek, a key figure in AI model reasoning, has founded a new company named Core Automation, focusing on "continuous learning" in AI models [1][5][7] - The company aims to raise between $500 million to $1 billion to develop a new type of AI model that can learn continuously from new data and experiences [1][8][10] Company Background - Jerry Tworek has a strong theoretical and mathematical background, having completed a master's degree in mathematics and worked in quantitative research before joining OpenAI in 2019 [3][5] - At OpenAI, he played a significant role in developing major models like o1, o3, GPT-4, ChatGPT, and Codex, pushing the boundaries of AI from mere generation to reasoning capabilities [3][5] Industry Context - The current mainstream AI models are primarily trained once and deployed, which limits their ability to adapt to new situations [5][10] - Continuous learning is seen as a solution to reduce costs and improve efficiency, allowing models to learn from real-world experiences rather than relying solely on static data [10][12] - The concept of continuous learning is gaining traction, with other companies and academic institutions, such as Google Research, also exploring this area [15][17] Future Outlook - The industry consensus suggests that achieving Artificial General Intelligence (AGI) will require models to possess continuous learning capabilities, which is a key focus for Tworek's new venture [12][15] - There is a growing belief that 2026 could mark a significant advancement in continuous learning technologies [19]

Core Viewpoint - Jerry Tworek, a key figure in AI model reasoning, has founded a new company called Core Automation, focusing on "continuous learning" in AI models and plans to raise $1 billion (approximately 70 billion RMB) for this venture [1][15][20]. Company Background - Jerry Tworek played a crucial role in the development of OpenAI's reasoning capabilities and has a strong theoretical and mathematical background, having completed a master's degree in mathematics at the University of Warsaw [4][6][9]. - Before joining OpenAI in 2019, he worked in quantitative research, which shaped his interest in reinforcement learning [7][9]. Focus on Continuous Learning - The new company aims to address the challenge of how models can continuously learn from new data and experiences, rather than being static after deployment [12][15]. - Tworek believes that current mainstream models are limited to a "train and deploy" approach, which does not adapt to new situations encountered in real-world applications [12][22]. Implementation Strategy - Core Automation plans to develop a new architecture that does not rely on Transformers and aims to integrate the training process into a continuous system, allowing models to learn while in operation [17][20]. - The goal is to enable AI models to learn from ongoing experiences while retaining previously acquired knowledge [16][22]. Industry Context - The continuous learning approach is gaining traction, with other companies and academic institutions also exploring similar directions, such as Ilya's SSI company and Google Research's new methodologies [24][28]. - The industry consensus suggests that achieving Artificial General Intelligence (AGI) requires models to possess capabilities akin to biological systems, including continuous evolution and self-optimization, making continuous learning a critical aspect [23][24]. Future Outlook - The ambition to raise $1 billion reflects the high expectations for the potential of continuous learning in AI, with industry experts predicting that 2026 could be a pivotal year for this field [31].

Core Insights - The article discusses the launch of AlphaGenome, a new AI tool by DeepMind that predicts the effects of single nucleotide mutations in human DNA sequences, enhancing the understanding of genetic diseases and guiding DNA design [2][3]. Group 1: AlphaGenome Overview - AlphaGenome is a DNA sequence model capable of processing up to 1 million base pairs, accurately predicting a wide range of genomic features and mutation effects [10]. - The model represents a significant advancement in genomic AI, moving from specialized models to a unified approach that can handle multiple tasks simultaneously [11][12]. Group 2: Technical Innovations - AlphaGenome achieves a breakthrough by maintaining single-base resolution while analyzing long sequences, combining the strengths of convolutional neural networks and transformer architectures [11][15]. - It can evaluate the impact of genetic mutations on various molecular characteristics in just one second, facilitating rapid identification of potentially disease-causing genetic variations [13]. Group 3: Performance Metrics - In 24 DNA sequence function prediction tasks, AlphaGenome achieved state-of-the-art performance in 22 tasks, and in 26 genetic variant impact prediction tasks, it excelled in 24 tasks, outperforming many specialized models [19]. Group 4: Practical Applications - AlphaGenome has been utilized to explore the mechanisms of mutations related to cancer, linking non-coding region mutations to the activation of oncogenes [22]. - It also aids in understanding rare genetic diseases caused by RNA splicing errors and can guide the design of synthetic DNA sequences for targeted gene therapy [24]. Group 5: Future Implications - The introduction of AlphaGenome signifies a shift in genomic AI from single-task specialists to comprehensive models, paving the way for predictive science in biology [26]. - It enhances the ability to predict molecular functions and mutation effects from DNA sequences, opening new avenues for biological discoveries and applications in biotechnology [26].

Yep, things are going to get crazy pretty fastDisclose.tv (@disclosetv):NEW - Google DeepMind launches new AI tool AlphaGenome, which can create entirely new "designer" DNA in humans — Telegraph https://t.co/45SKb2TS5R ...

在达沃斯世界经济论坛上，当代表们不在讨论唐纳德·特朗普时，他们正被人工智能的前景所震撼。 在瑞士阿尔卑斯山小镇的主街上，几乎每个店面都临时贴上了科技公司的霓虹标语，或是承诺告诉高管 如何将AI融入业务的咨询公司广告。Cloudflare的木质镶板总部敦促代表们"连接、保护并共同构建"， 而Wipro则高呼："梦想解决证明重复"。 在会议上，科技CEO们阐述了他们对AI物理表现形式在未来几年如何覆盖世界的希望。微软首席执行 官萨提亚·纳德拉向专注的观众讲述了"Token工厂"（他对数据中心的称呼）如何必须分布在世界各地， 以在全球范围内传播AI的好处。 "对我来说，一个长期的、可扩展的解决方案是让所有这些Token工厂成为真实经济的一部分，连接到电 网，连接到电信网络——这就是推动规模化的因素，无论是在全球南方，还是在发达国家，"纳德拉 说。 与此同时，谷歌正在向兴奋的代表们展示其谷歌眼镜的最新迭代产品；在达沃斯会议中心有无数关于这 项技术潜在好处的会议——包括与临时加入议程的埃隆·马斯克进行的激动人心的聊天，不过由于 SpaceXIPO似乎即将到来，他最热衷于谈论前往火星的话题。 然而，在华丽的店面之外，有人 ...

论文第一作者、北京通用人工智能研究院张驰博士介绍，TongGeometry能从浩如烟海的空间组合中， 精准捕捉到具备人类数学家审美标准的高质量题目，在国际上首次实现从"模仿解题"到"自主创造"的范 式转变。 我国科研团队近日开发出全球首个同时具备自主出题和自动解题双重能力的通用人工智能系统——"通 矩模型"（TongGeometry）。相关成果"基于引导树搜索的奥数几何问题提出与解答系统"1月26日发表于 《自然·机器智能》上。 奥林匹克数学竞赛被视为人工智能逻辑推理能力的"试金石"。2024年初，DeepMind公司开发的 AlphaGeometry人工智能系统展示了AI在解题方面的巨大潜力，但其本质上是一个"被动解题者"，训练 极度依赖大规模的合成数据和昂贵的计算资源。与之相比，我国自研的TongGeometry则展现出更高维 度的智能：不仅是一个能满分交卷的"优等生"，更是一位能创造优美、新颖题目的"出题名师"。其自主 生成的3道几何新题，已正式入选2024年全国中学生数学联赛（北京赛区）及美国精英奥赛。 论文共同通讯作者、北京大学心理与认知科学学院助理教授朱毅鑫表示，这意味着中国科研团队在自动 化推理 ...

Lila Ibrahim is the COO of Google DeepMind. James Manyika is the senior Vice President for Research, Technology, and Society at Google. The two join Big Technology Podcast to discuss how Google's AI effort operates and runs experiments. In this conversation, we discuss the fundamental operating structure of DeepMind, how Google proper has become more experimental with the revival of Labs and other programs, and how the company is thinking about AI and education. We also cover weather and flood prediction at ...

在性能表现上，TongGeometry展现了极高的国产原创技术优越性。相比AlphaGeometry需要庞大的算力 集群，TongGeometry仅需单张消费级显卡（如RTX 4090）即可在最多38分钟内，解决近25年所有的奥 数几何难题，其推理效率和准确率均达到世界顶尖水平。此外，该系统通过创新的"规范化表示"技术， 将搜索空间压缩了几个数量级，有效解决了传统方法中的路径爆炸问题。 论文共同通讯作者、北京大学心理与认知科学学院助理教授朱毅鑫表示，TongGeometry的意义不仅在 于解题速度的提升，更在于它通过模拟人类数学家的直觉和审美，实现了"小数据、大任务"的范式转 化。这种不依赖海量标注数据、通过内部逻辑自我演化的路径，正是通用人工智能（AGI）发展的关 键。"我们的系统不仅能与国际最先进的AI系统对标，更在理解逻辑底层美学和自主发现科学规律方面 走在了前列。"他说。 来源：科技日报 科技日报记者 杨雪 我国科研团队开发出全球首个同时具备自主出题（Proposing）和自动解题（Solving）双重能力的通用 人工智能系统——"通矩模型"（TongGeometry）。相关成果"基于引导树搜索的奥数 ...

Core Insights - The article discusses the upcoming AI+ Renaissance Summit 2026, highlighting its significance in the AI era and the involvement of Z Potentials as a partner [1][3]. Group 1: Event Details - The summit will feature 40 prominent speakers from various sectors, including AI entrepreneurship, cutting-edge research, and industry applications [3]. - The event is expected to gather 2000 founders, builders, and investors for in-depth discussions on key directions for the next generation of AI [4]. Group 2: Notable Participants - Founders from AI unicorn companies such as Replit, Cognition, Parallel Web System, and Tavus will be present [7]. - Key contributors from major AI models and frameworks, including OpenAI, xAI, Anthropic, and DeepMind, will also participate, alongside technology and business leaders from top tech companies like Salesforce, NVIDIA, Cisco, and Microsoft [7].

Group 1 - OpenAI CEO Altman announced the release of significant Codex-related content starting next week, with a technical blog revealing the core architecture of Codex CLI, specifically the intelligent agent loop [1] - The intelligent agent loop coordinates user instructions, model inference, and local tool execution through the Responses API, employing a "consistent prompt prefix" strategy to trigger cache optimization [1] - Codex supports zero data retention configurations to ensure privacy and utilizes automatic compression technology to manage context windows, with further details on tool invocation and sandbox models to be introduced later [1] Group 2 - Google DeepMind released D4RT, which unifies 3D reconstruction, camera tracking, and dynamic object capture into a single "query" action, achieving speeds 18 to 300 times faster than existing state-of-the-art methods [2] - The core innovation is a unified spatiotemporal query interface, where AI first globally "reads" videos to generate scene representations and then searches for 3D trajectories, depth, and poses of any pixel on demand [2] - This technology is significant for embodied intelligence, autonomous driving, and AR, although training still requires a 1 billion parameter model and 64 TPUs [2] Group 3 - Claude Code upgraded its internal "Todos" to "Tasks," enabling multi-session or sub-agent collaboration on long-term complex projects across multiple context windows [3] - Tasks are stored in a file system for easy collaboration among multiple sessions, with updates in one session broadcasting to all sessions handling the same task list [3] - The new feature is compatible with Opus 4.5, enhancing autonomous operation capabilities, allowing users to enable multiple sessions to collaborate on the same task list through environment variables [3] Group 4 - Baidu's Wenxin 5.0 officially launched with a parameter count of 2.4 trillion, utilizing native multimodal unified modeling technology to support understanding and generation of text, images, audio, and video [4] - It has topped the LMArena text and visual understanding leaderboard five times, entering the global first tier, with language and multimodal understanding capabilities leading internationally [4] - Practical tests show the model excels in complex emotional understanding, subtext analysis, and creative writing tasks, earning the title of "strongest liberal arts student" [4] Group 5 - The open-source project Clawdbot has gained popularity in Silicon Valley, capable of running on Mac mini, serving as both a local AI agent and chat gateway, allowing conversations via WhatsApp, iMessage, etc. [5] - Clawdbot addresses the memory limitations of large models, capable of recalling conversations from two weeks ago, proactively sending emails, reminders, and executing tasks on the computer [5] - The project has received 9.2k stars on GitHub, with a minimum monthly cost of approximately $25, though it requires some technical knowledge for deployment, and users report it can automate business management and code writing, replacing paid services like Zapier [5] Group 6 - Turing Award winner LeCun announced that AMI Labs' core direction is "world models," aiming to build intelligent systems that understand the real world, possess persistent memory, and have reasoning and planning capabilities [6] - This approach argues that merely predicting the next token does not lead to true understanding of reality, necessitating predictions and reasoning at a higher representational level to filter out unpredictable noise [6] - AMI Labs is reportedly seeking financing at a valuation of $3.5 billion, targeting applications in industrial control, robotics, and healthcare, where reliability is crucial [6] Group 7 - Anthropic launched the Claude in Excel plugin, available for Pro, Max, Team, and Enterprise users, based on the Opus 4.5 model, which can be installed and activated via Microsoft Marketplace [7] - The plugin can search the internet and automatically fill in spreadsheets, supporting formula reading, debugging errors, zero-based modeling, and pivot table creation, compatible with .xlsx and .xlsm formats [7] - Currently, it does not support conditional formatting, macros, or VBA, and the company warns of prompt injection risks, advising users to only use files from trusted sources, with high-risk functions triggering confirmation prompts [7] Group 8 - Claude Code's creator Boris Cherny provided a detailed tutorial on using Cowork, emphasizing its role as an "executor" rather than a chat tool, capable of directly manipulating documents, browsers, and various tools [8] - He reiterated that the core workflow involves running multiple tasks in parallel while overseeing Claude instances, starting with "planning mode" for communication until satisfaction is achieved, then switching to "auto-accept edits" mode for execution [8] - Cherny highlighted the importance of Claude.md as a team compounding knowledge base, where any mistakes made by Claude should be documented, and methods for validating Claude's outputs can significantly enhance quality [8] Group 9 - Google Cloud AI Director Addy Osmani warned that programmers who only write prompts will be eliminated by 2026, stating that AI can handle 70% of preliminary work, but the remaining 30% requires experienced engineers [9] - A Stack Overflow survey indicated that developer trust in AI accuracy dropped from 40% to 29%, with 73% of respondents encountering issues with code comprehension due to "ambient coding" [9] - By 2026, the true core competency will be transforming vague problems into clear execution intentions, designing appropriate contextual structures, and distinguishing what is truly important [9] Group 10 - At the Davos Forum, tech giants shared notable insights, with Musk predicting that AI will surpass human intelligence by the end of 2026 and be smarter than the collective intelligence of humanity by 2030, with Tesla set to launch the humanoid robot Optimus next year [10] - Microsoft CEO Nadella warned that if AI only consumes resources without improving outcomes, society will lose tolerance, while Huang Renxun stated that embodied intelligence represents a "once-in-a-generation opportunity" [10] - DeepMind CEO Hassabis believes AGI will still require 5-10 years, while Anthropic CEO Dario claimed that models are just 6-12 months away from being able to complete software development end-to-end [10]