Jensen Huang: Distillation is Fundamental to Intelligence, Bullish on Chinese Open-Source Models
In an interview with Axios, NVIDIA CEO Jensen Huang shared his core views on "distillation," synthetic data, and Chinese open-source models. He stated that distillation is a fundamental component of intelligence, emphasizing that as AI-generated content dominates the internet, AI systems learning from one another is inevitable. He also noted that the market is currently underestimating Chinese open-source models like DeepSeek and Kimi.
Confirmed
- The Legitimacy and Inevitability of Distillation: Huang emphasized that distillation (using the output of one model to train or improve another) is a common practice in model development and should not be broadly deemed "abuse" in policy discussions. He described it as a natural extension of human learning, noting that both open-source and closed-source models are fundamentally built on existing knowledge. As AI increasingly generates web content, AI learning from other AI systems is inevitable.
- Optimism for Chinese Open-Source Models: Huang argued that Chinese open-source models are not the antithesis of closed-source models; instead, they expand the overall market by making high-quality AI more accessible. He specifically pointed out that the market initially underestimated DeepSeek and is currently underestimating Kimi.
Why it matters
The AI industry is currently rife with copyright and ethical disputes over "distillation" (particularly using closed-source model outputs to train open-source ones). As the helmsman of the computing power giant, Huang's public endorsement—defining it as a necessary stage in the evolution of intelligence—provides a strong defense for the technology's legitimacy. Meanwhile, his positive assessment of Chinese open-source models like DeepSeek and Kimi sends a clear signal to the market that the open-source ecosystem will continue to thrive and grow the overall market.
2026-07-24 ~ 2026-07-26 · 5 related posts
- Episode 1: AI Model Distillation Debate: Normal Tech Evolution or IP Theft?(2026-07-18, 5 posts)
- Episode 2: Irony and Ethics in AI Training Data and Distillation(2026-07-19, 2 posts)
- Episode 3: AI Industry Debates the Legitimacy of Model Distillation(2026-07-23, 9 posts)
- Episode 4: Industry Debates Open-Weight Model Regulation and Distillation(2026-07-24, 2 posts)
- Episode 5: Jensen Huang: Distillation is Fundamental to Intelligence, Bullish on Chinese Open-Source Models(2026-07-24, 5 posts)
- Episode 6: AI Model Distillation Sparks Debate from Silicon Valley to Washington(2026-07-25, 4 posts)
- Episode 7: Jensen Huang Defends AI Distillation as Learning, Not Stealing(2026-07-26, 5 posts)
- Episode 8: Jensen Huang Defends Model Distillation as Learning, Not Plagiarism(2026-08-06, 2 posts)
Primary sources
- [source] Jensen Huang says distillation is fundamental as AI learns from AI — rohanpaul_ai · 2026-07-24
- Full video: Jensen Huang on why AI systems will keep learning from AI — rohanpaul_ai · 2026-07-24
- [source] Jensen Huang says distillation is a standard AI technique, not a reason for broad restrictions — AccBalanced · 2026-07-25
- [source] Jensen Huang says Chinese open models grow the whole AI market, not fight it — rohanpaul_ai · 2026-07-25
- Jensen Huang says AI distillation is fundamental as web content turns synthetic — Aiden_Tech_Ai · 2026-07-26