OpenAI seeks researchers who can stop AI from improving itself
What's the story
OpenAI is looking for a researcher to tackle one of the biggest challenges in artificial intelligence (AI): what happens when an AI system can improve itself. The role, which falls under company's Preparedness safety team, offers a salary range of $295,000 to $445,000 (around ₹2.5 crore to ₹3.7 crore), according to Business Insider. The job listing highlights a concept called recursive self-improvement, the ability of an AI system to research and design better versions of itself without much human involvement.
Industry focus
Recursive self-improvement concerns
The concept of recursive self-improvement has gone from a theoretical worry to an industry-wide concern in the last six months. This is mainly due to the rapid development of coding tools by OpenAI and Anthropic, which have even surprised their own researchers. Google DeepMind CEO Demis Hassabis recently said humanity is now at the "foothills of the singularity," where AI starts improving itself and surpassing human intelligence.
Rapid progress
AI's capabilities advancing rapidly
Researchers at METR, a lab that studies AI model capabilities, found that the length of tasks frontier AI models can complete doubles roughly every seven months. This means these advanced AI agents will soon be able to do a "large fraction" of software work that human coders take days or weeks to finish. OpenAI CEO Sam Altman has been vocal about the company's plans in this area.
Future goals
Plans for an automated AI research intern
In October, Altman revealed that OpenAI plans to run an "automated AI research intern" on hundreds of thousands of chips by September 2026. The company also hopes to have a "true automated AI researcher by March of 2028." Despite acknowledging the ambitious nature of these targets, Altman believes it's important to be transparent about them due to their potential impact on the public interest.
Business strategy
OpenAI's next frontier and the researcher role
OpenAI, which plans to go public this year, is already monetizing its AI coding tools through its Codex product. The company sees automating its internal research work as the next frontier in this journey. The job listing for the researcher role highlights various tasks they could handle within OpenAI's Preparedness team, such as defending against data poisoning attempts, building tools to interpret models' reasoning, and running experiments on self-improving systems' safety implications.