OpenAI builds automated research intern for AI R&D

The author argues that the automation of AI research could fundamentally change the economics of frontier model development by making specialized applications cheaper to build relative to general foundation models. While general foundation models benefit from economies of scope through large training runs, automated research labor could disproportionately reduce the costs of targeted, labor-intensive customization. This shift could weaken the economic case for scaling general intelligence and complicate current AI governance frameworks that rely on training run size as a proxy for capability. Open article →