# OpenAI just said it Source: https://www.youtube.com/watch?v=WrEVCsK4XOQ Recap page: https://rapidrecap.app/video/WrEVCsK4XOQ Generated: 2025-10-29T22:32:25.789+00:00 --- ## Quick Overview OpenAI's internal roadmap, discussed by Sam Altman and others, projects Automated AI Research Intern by September 2026 and Automated AI Research by March 2028, suggesting an intelligence explosion where AI development accelerates past human capability, fundamentally altering scientific progress and research timelines. **Key Points:** - OpenAI's internal timeline projects achieving an 'Automated AI research intern' capability by September 2026. - The roadmap further projects 'Automated AI research' capability by March 2028, signifying AI conducting research autonomously. - The progression suggests an intelligence explosion where AI research accelerates rapidly, potentially surpassing human scientific discovery rates. - The current state of AI development is estimated to be around the '5-hour tasks' level on the automated research scale, with models still requiring significant human input. - The rapid scaling of deep learning models means that within a decade, AI systems could become smarter than all humans combined, necessitating careful planning for this future. - The video highlights the importance of this potential shift, suggesting that the move toward self-improving AI is the most critical news, not just incremental AI updates. ![Screenshot at 00:00: The timeline graphic illustrating OpenAI's internal projections for 'Automated AI research intern' by September 2026 and 'Automated AI research' by March 2028, which forms the central subject of the discussion regarding future AI capabilities.](https://ss.rapidrecap.app/screens/WrEVCsK4XOQ/00-00-00.png) **Context:** The video features a discussion, likely involving AI leaders like Sam Altman and potentially others such as Jakob Pachocki and Yann LeCun/Jürgen Schmidhuber (mentioned in passing regarding past papers), analyzing a projected timeline for Artificial General Intelligence (AGI) and superintelligence development, specifically focusing on OpenAI's internal roadmap for automating research. ## Detailed Analysis The discussion centers on an internal OpenAI roadmap detailing aggressive timelines for achieving automated AI research capabilities. The presenter points out that the roadmap suggests an 'Automated AI research intern' will be achieved by October 2025 or September 2026, and full 'Automated AI research' by March 2028. This acceleration implies an intelligence explosion where AI research progresses exponentially, eventually leading to AI systems that are 'smarter than all of us' and capable of autonomously driving scientific breakthroughs. The current state is likened to solving problems that take humans about 5 hours, but the trend line shows rapid compression of task completion times (e.g., 5-second tasks to 5-year tasks). The speakers emphasize that this potential shift, far beyond incremental product updates, is the most crucial development to monitor, requiring organizations to plan for a future where AI development outpaces human efforts significantly, potentially leading to superintelligence much sooner than commonly anticipated. ### OpenAI Internal Roadmap - Automated AI research intern by September 2026 - Automated AI research by March 2028 - Implication of rapid intelligence explosion ### Automated Research Progress Scale - 6-second tasks completed currently - 5-hour tasks recently achieved - 5-year tasks projected far out ### AI Development Trajectory - Scaling deep learning is fueling acceleration - AI systems becoming smarter than all humans within a decade - Need for alignment and safety planning ### Key Individuals Mentioned - Sam Altman, Jakob Pachocki, Yann LeCun, Jürgen Schmidhuber (in context of past papers) ![Screenshot at 00:00: The timeline graphic illustrating OpenAI's internal projections for 'Automated AI research intern' by September 2026 and 'Automated AI research' by March 2028, which forms the central subject of the discussion regarding future AI capabilities.](https://ss.rapidrecap.app/screens/WrEVCsK4XOQ/00-00-00.png) ![Screenshot at 00:04: An insert image showing a person holding a fan of US currency, used metaphorically during a discussion about capitalization or value related to AI development.](https://ss.rapidrecap.app/screens/WrEVCsK4XOQ/00-00-04.png) ![Screenshot at 00:15: Portraits of Sam Altman and Jakob Pachocki, who are mentioned as discussing these internal goals and timelines.](https://ss.rapidrecap.app/screens/WrEVCsK4XOQ/00-00-15.png) ![Screenshot at 00:30: A dramatic, dystopian-style visual overlay emphasizing the seriousness of the topic, accompanied by a 'please subscribe' banner.](https://ss.rapidrecap.app/screens/WrEVCsK4XOQ/00-00-30.png) ![Screenshot at 02:36: The 'Scenario Intelligence Explosion' chart, showing effective compute normalized to GPT-4, with projections for automated research leading to superintelligence.](https://ss.rapidrecap.app/screens/WrEVCsK4XOQ/00-02-36.png) ![Screenshot at 02:42: Zoomed-in view on the chart highlighting key milestones like 'Automated AI Research' and 'Automated Alex Rad\[ford?\]', referencing specific points on the compute curve.](https://ss.rapidrecap.app/screens/WrEVCsK4XOQ/00-02-42.png) ![Screenshot at 05:14: Slide titled 'AI impact on science' with three categories: Research, Product, Infrastructure, setting the context for internal goal alignment.](https://ss.rapidrecap.app/screens/WrEVCsK4XOQ/00-05-14.png) ![Screenshot at 06:14: A slide titled 'Automated research' showing a diagonal line representing progress from 6-second tasks up to 5-year tasks, illustrating the speedup in research cycles.](https://ss.rapidrecap.app/screens/WrEVCsK4XOQ/00-06-14.png) ![Screenshot at 06:57: Yakob Pachocki \(in the inset window\) discussing how the organization structures its research program around these timelines.](https://ss.rapidrecap.app/screens/WrEVCsK4XOQ/00-06-57.png)