Inside OpenAI’s quest to make AI do anything for you
OpenAI’s push for general-purpose AI agents began with MathGen, a team that taught models to crack high school math contests. By combining large language models, reinforcement learning and ‘chain-of-thought’ planning, OpenAI unveiled its first self-reasoning model, o1, in late 2024. These advances power AI agents that can navigate coding, browsing and other tasks. While current agents struggle with complex, subjective jobs, OpenAI believes new training methods and more compute will soon turn ChatGPT into a truly do-anything assistant.