OpenAI aims to build AI systems that can work on problems for hours or even days

OpenAI is working on AI systems that can tackle problems for hours or even days at a time. In the company's official podcast, Chief Scientist Jakub Pachocki and researcher Szymon Sidor share inside stories about building these long-term thinking models, which are designed to plan, reason, and experiment over extended periods. OpenAI's math and code models, which recently won Olympic gold in their fields, offer an early glimpse of this approach.

The goal is to automate parts of the research process, such as AI-driven discovery of new ideas in medicine or AI safety. According to the researchers, making this possible will require significantly more computing power than most users have today, which explains Sam Altman's willingness to invest "trillions of dollars" in data centers over the coming years.