Nathan Labenz on the final push for AGI, understanding OpenAI's leadership drama, and red-teaming frontier models

Abstract

This article presents a conversation between Nathan Labenz, an entrepreneur and AI scout, and host of The Cognitive Revolution podcast, and Rob Wiblin, head of research at 80,000 Hours. Labenz describes his experiences as part of OpenAI’s red team that probed GPT-4, a powerful language model, before its public release. He initially expressed concerns about OpenAI’s lack of attention to safety, but later found that the company had undertaken extensive efforts to address these concerns. He argues that OpenAI’s leadership is among the most thoughtful and competent in the industry, recognizing the potential risks of advanced AI and advocating for reasonable regulations. Labenz, however, questions the wisdom of OpenAI’s single-minded pursuit of general artificial intelligence (AGI) given the apparent disparity between rapid capability advancements and less developed safety measures. He proposes that focusing on narrow, specialized AI applications could both yield significant benefits and provide more time to develop robust safety protocols for AGI. Labenz concludes by emphasizing the unique responsibility of AI researchers and developers, particularly those at OpenAI, to carefully consider the potential implications of their work and actively contribute to shaping the future of AI. – AI-generated abstract