
The podcast explores the challenges and potential slowdowns in AI development, particularly concerning compute growth and its impact on AI capabilities. It addresses the question of whether a causal relationship exists between compute and time horizon, suggesting that a halving of compute growth could proportionally reduce the time horizon. The conversation covers the limitations of current AI models, especially in complex tasks requiring tacit knowledge and the difficulties in automating chip production, with opinions diverging on the timeline for achieving full automation. The discussion also examines the effectiveness of AI in various fields, such as data science, law, and robotics, highlighting the gap between theoretical potential and practical application due to issues like data quality and the need for human oversight.
Sign in to continue reading, translating and more.
Continue