The Download: OpenAI is building a fully automated researcher, and a psychedelic trial blind spot

OpenAI is pivoting its long-term strategy toward a new grand challenge: developing a fully autonomous AI researcher. In an exclusive interview, Chief Scientist Jakub Pachocki outlined the plan, describing the creation of an agent-based system capable of independently tackling large, complex scientific and technical problems as the company's new 'north star' for the coming years. This marks a significant shift from building general-purpose chatbots like ChatGPT to developing specialized AI that can reason, plan, and execute research workflows.

The development will be phased, beginning with a more limited 'autonomous AI research intern' slated for completion by September 2024. This initial agent will be designed to take on a small set of specific research problems. It will serve as the precursor and testing ground for a more advanced, fully automated multi-agent system, which OpenAI aims to debut in 2028. This long-term project underscores the company's ambition to move beyond tools that assist human researchers to creating AI that can conduct end-to-end research autonomously.

Concurrently, OpenAI is consolidating its consumer-facing products by building a 'super app' that merges ChatGPT, a web browser, and a coding tool into a single interface. This move, alongside the acquisition of coding startup Astral to enhance its Codex model, suggests a strategic focus on core, integrated products while scaling back side projects. The dual focus on a powerful research agent and a unified user app highlights OpenAI's efforts to solidify its position in both advanced AI development and the competitive enterprise market.