OpenAI is throwing everything into building a fully automated researcher
OpenAI's new 'North Star' is an autonomous AI that can tackle complex research problems for days without human help.
OpenAI is making a massive strategic pivot, declaring the development of a fully automated AI researcher as its new 'North Star' for the coming years. Led by Chief Scientist Jakub Pachocki, who helped develop GPT-4 and reasoning models, the company plans to build an 'autonomous AI research intern' by September 2024. This precursor will evolve into a sophisticated multi-agent system by 2028, designed to tackle complex, long-running scientific and business problems formulated in text or code without constant human oversight.
Pachocki believes current tools like OpenAI's Codex agent, which many staff already use for coding and analysis, are early versions of this vision. The key breakthrough needed is extending an AI's 'coherence time'—its ability to work autonomously on a single problem for days, similar to a human researcher. This push is partly driven by the success of coding agents, raising the question of whether similar delegation can work in broader scientific domains. OpenAI sees this as the logical next step, leveraging advances in reasoning models and interpretability to create what Pachocki describes as 'a whole research lab in a data center.'
- OpenAI's new 'North Star' is an autonomous, multi-agent AI researcher system with a 2028 launch target, preceded by an 'AI research intern' in September 2024.
- The system aims for extended 'coherence time,' allowing it to work on complex problems (like proofs or policy dilemmas) for days without human guidance, a leap from current agent capabilities.
- The project is led by Chief Scientist Jakub Pachocki and builds directly on existing agent tech like Codex, which is already used internally for coding and analysis tasks.
Why It Matters
This could automate high-level R&D, accelerating discoveries in science and business, but also centralizes immense problem-solving power with one company.