Experts Doubt Kimi K3's Capabilities Stem Solely From Anthropic Fable Exploitation
Experts are questioning claims that Chinese firm Moonshot's Kimi K3 AI model's advancements are purely derived from distilling Anthropic's Fable model. The model's rapid two-week development cycle raises skepticism.
Michael Kratsios, the White House science advisor, has accused Chinese firm Moonshot of copying Anthropic's Fable AI model in its development of the Kimi K3 model, reportedly using chips restricted from export to China.
"Large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable," Kratsios stated on social media, though he did not detail the sources of his allegations. The administration has expressed concerns about Chinese open-weight AI models, and discussions around potential bans have roiled the industry.
However, industry experts are skeptical that distillation alone—the process of querying one LLM to mimic another's capabilities—explains the advanced performance of Kimi K3. Given that Anthropic's Fable model was only publicly released two weeks prior, many find it improbable that sufficient data could be distilled and a new model trained in such a short timeframe.
Researchers like Braden Hancock of the Laude Institute and Nathan Lambert of the Allen Institute for AI suggest that the speed of Kimi K3's development indicates more sophisticated training methods were likely employed. They propose that techniques such as reinforcement learning were likely utilized, alongside any distillation, to achieve a competitive performance against other leading large language models.