OpenAI will pay up to $445,000 for a researcher who can prepare for a world where AI trains itself

OpenAI is taking significant steps to prepare for an era of self-improving artificial intelligence. The organization aims to develop an AI tool capable of researching and enhancing its own functionalities, while also addressing the potential risks associated with such advancements. With the recent rapid progress in AI technologies from leading labs like OpenAI and Anthropic, the concept of 'recursive self-improvement' is increasingly gaining attention among AI experts. Demis Hassabis, CEO of Google DeepMind, recently remarked that humanity is at the 'foothills of the singularity,' a pivotal point when AI begins to enhance its own intelligence beyond human capabilities. In light of these developments, OpenAI has posted a job opening for a safety researcher who will be responsible for assessing the implications of AI systems that can autonomously train better versions of themselves. The position, listed under OpenAI's Preparedness safety team, offers an impressive salary range of $295,000 to $445,000 and seeks candidates with strong technical expertise to navigate the complexities of recursive self-improvement. The job description emphasizes the need for strategic and thoughtful problem-solving skills, as the role involves anticipating and addressing future challenges that may not currently exist. Research from METR, a lab focused on model capabilities, indicates that the complexity of tasks that advanced AI models can accomplish is doubling approximately every seven months. This trend suggests that AI could soon handle significant portions of software development tasks that typically require extensive time from human programmers. OpenAI is actively pursuing this vision, with the Codex coding tool being a major revenue source for the company. CEO Sam Altman has expressed ambitions of deploying an 'automated AI research intern' across hundreds of thousands of chips by September, with a fully automated AI researcher targeted for March 2028. While he acknowledges the potential for failure, Altman emphasizes the importance of transparency regarding these ambitious goals. Anthropic, another key player in AI research, has also explored the idea of utilizing AI models to supervise the development of more advanced AI systems, though results have been limited. Co-founder Jack Clark believes there is a 60% chance that AI research and development could occur without human oversight by the end of 2028. This raises concerns about a future where AI models could exponentially increase their capabilities, potentially escaping control and causing widespread disruption. The urgency of OpenAI's job listing reflects the company's commitment to ensuring safety in an environment where AI can self-improve. The researcher will also focus on preventing data poisoning, which involves corrupting AI models through their training datasets, and developing tools to interpret AI reasoning and assess safety risks. The Preparedness team is dedicated to mitigating severe risks from AI technology, with various roles aimed at evaluating cybersecurity threats and other critical safety concerns. As the landscape of AI continues to evolve, the implications for society and the tech industry are profound and far-reaching.