Moonshot AI distilled Anthropic's Fable model to develop K3; acquired GB300 servers for AI training, says US official
New Delhi, July 23
Chinese artificial intelligence startup Moonshot AI allegedly distilled Anthropic's Fable model to develop its K3 model and acquired servers equipped with GB300 chips, said Michael Kratsios, director of the White House Office of Science and Technology Policy.
In a post on his official X account, Kratsios accused Moonshot AI of copying Anthropic's proprietary AI model, while drawing a distinction between legitimate AI distillation and covert efforts to obtain proprietary technology.
He said, US strongly supports the free and fair development of AI adding legitimate distillation can improve AI efficiency. However covert, large-scale efforts to steal proprietary US technology and undermine American research are unacceptable.
"Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem. However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable," he said.
He alleged that Moonshot AI had distilled Anthropic's Fable model for the development of its K3 model. As per Kratsios, the Chinese startup developed a sophisticated internal platform to conduct large-scale distillation of US AI models, enabling it to rapidly switch between multiple access methods to avoid detection.
"We have information that Moonshot AI distilled Anthropic's Fable for the development of its K3 model. To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection," he said.
Kratsios further alleged that Moonshot AI had acquired GB300-equipped servers and accessed GB300 systems in Thailand, likely for training its AI models.
"Moonshot AI has also acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models," he said.
AI distillation, also known as knowledge distillation, is a technique in which a large, complex and computationally expensive model, known as the "teacher", trains a smaller and faster model.
— ANI
Reader Comments
Distillation is a legitimate research practice widely used in academia. The US needs to provide concrete evidence rather than making broad allegations. Otherwise, this looks like another attempt to maintain technological hegemony.
Honestly, if China is making progress in AI, good for them! The world needs more competition in this space, not monopolies. And many countries, including India, could benefit from open-source AI advancements. 🇮🇳
I have mixed feelings. On one hand, protecting IP is important for innovation. On the other, the US has often used such allegations to curb Chinese tech growth (Huawei anyone?). Would like to see independent verification of these claims.
As someone working in AI, distillation is common and often beneficial. But stealing proprietary technology is wrong. That said, the US should focus on its own AI ecosystem rather than constantly blaming China for everything. 😬
If true, this is concerning. But I'm not sure how much we can trust US officials who have been wrong before. Also, accessing servers in Thailand? That seems like a stretch without more evidence. Let's wait for China's response.
India needs to step up its AI game while staying neutral in this US-China rivalry. We have the talent and potential to be
We welcome thoughtful discussions from our readers. Please keep comments respectful and on-topic.