Core Insights - OpenAI has undergone a significant organizational restructuring and has publicly shared a detailed timeline for its research goals, notably aiming to achieve fully autonomous AI researchers by March 2028 [1][6]. Group 1: Organizational Changes - The restructuring has simplified OpenAI's architecture, reducing it to two main layers: the OpenAI Foundation, a non-profit organization, and the OpenAI Group, a public benefit corporation [11][13]. - The foundation aims to control the public benefit corporation and has committed to investing $25 billion in AI-assisted medical research [16]. Group 2: Research Goals and Timeline - OpenAI's internal roadmap includes milestones such as having AI research interns by September 2026 and fully automated AI researchers by March 2028 [6][18]. - The organization believes that deep learning systems could achieve superintelligence within the next decade, with significant advancements in AI capabilities already being observed [6][10]. Group 3: Infrastructure and Investment - OpenAI has committed over 30 GW of infrastructure development, with total financial obligations amounting to approximately $1.4 trillion [18]. - The company aims to establish a factory capable of generating 1 GW of computing power weekly, with a target cost of $20 billion per GW over a five-year lifecycle [18]. Group 4: AI Safety and Ethics - A new technique called "Chain of Thought Faithfulness" has been introduced to ensure AI models maintain fidelity to their internal reasoning processes [8][10]. - OpenAI emphasizes the importance of aligning AI values with human principles, especially as AI systems become more intelligent and face complex, ambiguous problems [10][11]. Group 5: Public Engagement and Feedback - During a recent live event, OpenAI acknowledged mistakes in its approach to user feedback and expressed a commitment to improving user experience while ensuring safety for vulnerable populations [3][5]. - The company is actively seeking to balance user freedom with the protection of minors and individuals in vulnerable situations [5][19].

首次公开了内部研究目标的具体时间表，其中最引人注目的是 "在2028年3月实现完全自主的AI研究员" ，具体到月份。 这次发布会信息密度非常大，连奥特曼自己都说："鉴于这些内容的重要性，我们将以不同寻常的透明度分享我们的具体研究目标、基础设施 计划和产品战略。" 难道重组后的OpenAI，真的重新Open了？ 梦晨 发自 凹非寺 量子位 | 公众号 QbitAI OpenAI完成史上最重要的一次组织架构调整后，紧接着开了一场直播。 不过也有一些事故，本来OpenAI发帖征集大家的问题，结果抱怨GPT-4o对敏感对话强制路由机制的人太多，两人支支吾吾面面相觑了一 阵。 奥特曼最终还是承认"这次我们搞砸了"。 我们的目标是在保护脆弱用户的同时给成人用户更多自由。我们有义务保护未成年用户，保护那些不在合理心态下的成人用户。 直播一开始，奥特曼就承认自己的错误。 过去，我们把AGI想象成"天上的神谕"，超级智能会自动为人类创造美好的事物。 但现在我们意识到，真正重要的是创造工具，让人们用这些工具创造自己的未来。 这种思维转变并非偶然，人类历史上每一次技术革命都源于更好的工具，从石器到蒸汽机，从计算机到互联网。 Ope ...

Core Insights - OpenAI has outlined a clear roadmap for AI development, aiming to have AI research interns by September 2026 and fully automated AI researchers by March 2028, emphasizing a systematic approach rather than a sudden emergence of AGI [1][2][49] Group 1: AI Development Timeline - The goal is to create a system capable of independently completing research projects, with AI expected to reach intern-level capabilities by September 2026 and fully autonomous research capabilities by March 2028 [2][3] - The evolution of AI is shifting from merely solving problems to becoming a genuine research system, with expectations of achieving superintelligence within a decade [3][6] Group 2: Training and Infrastructure - OpenAI is focusing on two key variables in model training: the duration of "context computation" during training and the length of "inference time" post-training [8] - A significant investment is being made in infrastructure, with plans to build a facility capable of producing 1 gigawatt of computing power weekly, with a total investment commitment exceeding $1.4 trillion [26][29] Group 3: AI as a Platform - OpenAI is transitioning from viewing ChatGPT as a "super assistant" to developing it into an AI cloud platform where users can create their own AI services [12][15] - The platform will allow enterprises to integrate OpenAI technology via APIs and enable developers to create application-specific intelligent agents [15][18] Group 4: Safety and Security Framework - OpenAI has introduced a five-layer safety framework focusing on value alignment, goal alignment, reliability, adversarial robustness, and system security to ensure AI's safe integration into society [20][21] - A new research direction called "Chain of Thought Faithfulness" aims to make AI's reasoning process transparent and understandable [22][24] Group 5: Organizational Structure - OpenAI's structure has been redefined into a dual-layer system: the OpenAI Foundation, a non-profit controlling the mission, and the OpenAI Group PBC, a public benefit corporation executing commercial tasks [33][39] - The foundation has committed $25 billion to support AI healthcare and resilience initiatives, aiming to leverage AI for disease treatment and societal defense mechanisms [41][42] Group 6: Future Vision - OpenAI emphasizes that the competition among large AI enterprises will not only depend on the strength of their models but also on their ability to establish a balanced structure that integrates research, business, and public welfare [46] - The company envisions a future where AI capabilities are fully integrated into society, with a focus on preparing the necessary social structures before deploying advanced AI technologies [45][49]