Over the past few months, a series of departures from OpenAI have raised concerns about the organization's approach to safety, with some employees citing reservations about its commitment to this aspect. Among these departures was Leopold Aschenbrenner, a researcher who was dismissed in April and later shared his insights on the evolution of AI in a 165-page treatise. Aschenbrenner, who was part of OpenAI’s super alignment team focusing on mitigating AI risks, claimed that his firing was due to leaking information related to the company's preparedness for artificial general intelligence.

In his essay, Aschenbrenner forecasts a significant progression in AI technology, particularly in the transition from current models like GPT-4 to AGI, which he believes could happen faster than anticipated. He highlights predictions that by 2027, AI models may match the capabilities of human researchers, potentially leading to an intelligence explosion surpassing human intelligence levels. Aschenbrenner also emphasizes the substantial economic investments being poured into developing the necessary infrastructure for advanced AI systems, underscoring the importance of securing these technologies to prevent misuse, especially by state actors such as the CCP.

Furthermore, Aschenbrenner delves into technical and ethical challenges associated with controlling AI systems smarter than humans, considering it a crucial aspect to prevent potential catastrophic outcomes. He suggests that the impact of AI on industries, national security, and ethical considerations is often underestimated and discusses the need for a better understanding of how AI will transform various sectors in the future. Aschenbrenner also raises the issue of 'superalignment,' emphasizing the necessity to ensure superintelligent AI remains aligned with human values and interests.