NousResearch Questions Bias Layers in Chinese-Aligned AI Models

NousResearch said bias layers in Chinese-aligned AI models are shallow and often applied late in training, raising concerns about whether the systems can be trusted. The critique focuses on how developers address bias during model development.

The organization argued that applying the layers late may weaken confidence in the models. Its comments come amid broader scrutiny of AI transparency and reliability across the technology industry.

Former SEC Chair Gary Gensler has warned that lower-cost Chinese AI models could spread faster than U.S. systems and pressure American companies’ pricing power and data-center returns. Read more about Gensler’s warning on the competitive effects of Chinese AI models.