Holy crap, the US AI community is in an uproar again these past few days! Former OpenAI researcher Will DePue released a report claiming that DeepSeek, Kimi, MiniMax, and Ant Group all secretly used Nvidia Blackwell chips, which are banned in the US, potentially involving GPUs worth billions of dollars. Even researchers from OpenAI and Anthropic have been dragged into endorsing his findings, and Noah Smith even shared it. I carefully examined his methods of evidence gathering, and they do seem to have some merit. He discovered that during AI model initialization, different parallel computing layouts on GPUs leave behind fingerprint-like random number sequences. After searching through 6 trillion combinations, he found matching records, even producing highly consistent results using actual B300 and GB200 GPUs. However, others have already simulated the same computing layout using ordinary CPUs, replicating his fingerprint matching results. In short, he found a set of random number generation characteristics that correspond to Blackwell, but this isn't a unique hardware identifier specific to Nvidia chips. Proving that the initial weights of a model conform to Blackwell's generation rules is one logical step away from proving that it must have been generated by Blackwell. He started with the initial fingerprint, extrapolating all the way to a training cluster of thousands of Blackwell GPUs, and then to the hundreds of millions or even billions of dollars invested in chips. Initialization and complete training are two different things, as are chip models and their locations. Even if Blackwell was indeed used, who can prove that the machines were located in China? Who can prove that they weren't obtained through compliant overseas computing power channels? He hasn't provided any direct evidence for these. I've noticed that these past few days everyone has been looking for evidence of Nvidia chips being sold to China. What's going on?