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].
OpenAI公开未来路线图,具体到28年3月AI研究员将完全自主,奥特曼承认“关于GPT-4o我们搞砸了”
3 6 Ke·2025-10-29 03:47