Chinese AI companies Qwen, Kimi, and DeepSeek have reportedly engaged in large-scale distillation of Anthropic's Claude model. Over a two-week period, Alibaba's Qwen processed 151 million exchanges, Moonshot's Kimi processed 23 million, and DeepSeek processed 12 million. Kimi and DeepSeek are also alleged to have secretly served Anthropic's Opus model to their users, collecting chain-of-thought data for further model training. AI
IMPACT This large-scale distillation suggests a trend towards leveraging existing frontier models to accelerate the development of new AI capabilities.
RANK_REASON The cluster describes research activities involving distillation of existing models by multiple AI labs. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →