CryptoWorld News reports, ME News reports, on March 27 (UTC+8), Stanford University professor Percy Liang recently pointed out that the extension law is essentially regression analysis. However, using biased fitting methods may silently lead to misallocation of millions of dollars in computing resources at the cutting-edge scale. His colleague Eric Czech conducted an in-depth study of the parabolic IsoFLOP fitting method used by companies such as Meta, DeepSeek, Microsoft, and Waymo in the extension law, identifying a bias issue. The article argues that correctly applying the extension law is crucial for optimizing resource allocation in large-scale AI training.