This AI shortcut could destroy the industry's profits (www.businessinsider.com)

A recent exploration of AI distillation has revealed a growing concern among industry leaders, as this practice evolves into a shadow economy that threatens the profitability of AI giants. Distillation involves training one AI model using the outputs of another, and its use has prompted significant debate over ethics and legality. Prominent companies like OpenAI and Anthropic warn that this technique allows rivals, particularly in China, to replicate sophisticated models quickly and affordably, undermining the billions invested in developing cutting-edge AI systems. The emergence of models like GLM-5.2, which allegedly draws knowledge from U.S. models, has exacerbated investor fears, leading to declines in AI stock values. The implications of this trend challenge traditional economic models in AI development. While distillation initially served as a method for improving existing models, many researchers assert that its current application leans more toward appropriation than innovation. Current efforts by U.S. companies to restrict access, such as blocking certain users and tightening model usability, have inadvertently led to the rise of "transfer stations"—services allowing access to these models through proxies. This not only fuels a cycle of cheaper, distilled alternatives but also complicates the industry's regulatory landscape, as the line between legitimate research and exploitation blurs.