ATRIUMsearch → argument graph
Article · 2026-07-20 · 4 moments

Who’s Afraid of Chinese Models?

Who’s Afraid of Chinese Models? Interesting proposal from Ben Thompson that both addresses the hypocrisy of labs outlawing distillation against their models despite training on unlicensed data, and could help US open models compete more effectively with their Chinese counterparts: The U.S. should pass a law that (1) makes explicit that collecting data for training models is fair use, and (2) bars terms of service that forbid distillation, for U.S. companies at a minimum. Stopping distillation  ✦ AI generated

Read original ↗
GUEST Ben ThompsonGUEST Xi JinpingHOST Simon Willison
01
Prediction◆

Alibaba's reversal to release Qwen 3.8 Max as open weights, after declining to release Qwen 3.7 Max in May, may have been influenced by Xi Jinping's recent speech encouraging open source.

Ben Thompson theorizes that Alibaba's about-face on open-sourcing its Qwen model was shaped by a Xi Jinping speech pushing open-source collaboration.

transcript

Ben Thompson: Ben also theorizes that Alibaba's decision to release Qwen 3.8 Max as open weights - a reversal from their decision not to release Qwen 3.7 Max in May - may have been influenced by a recent speech by Xi Jinping.

explains mechanism · 2extends · 1provides context · 1

03
Claim◆

The U.S. should pass a law that explicitly makes data collection for training models fair use, and bars terms of service that forbid distillation, at least for U.S. companies.

Ben Thompson proposes a U.S. law that codifies training-data collection as fair use and outlaws anti-distillation clauses in AI terms of service, at least for domestic companies.

transcript

Ben Thompson: The U.S. should pass a law that (1) makes explicit that collecting data for training models is fair use, and (2) bars terms of service that forbid distillation, for U.S. companies at a minimum.

explains mechanism · 1extends · 1

04
Mechanism

Trying to legally stop distillation is nearly impossible because it is functionally indistinguishable from ordinary API querying, so policy should instead embrace openness via a new copyright framework that indemnifies labs while letting others benefit from what models learn.

Ben Thompson argues distillation can't realistically be blocked since it's just API queries, so the better path is a copyright policy that indemnifies labs and lets their learned knowledge benefit everyone.

transcript

Ben Thompson: Stopping distillation — which is literally just querying the API — is nearly impossible; the U.S. should go the other way and lean into a new copyright policy that both indemnifies the labs and also guarantees that what they learned fuels further innovation for everyone else.

explains mechanism · 1extends · 1supports · 1

Highlight slides
Fair Use for Training Data✦ from: The U.S. should pass a law that explicitly makes data collection for training models fair use, and bars terms of service that forbid distillation, at least for U.S. companies.Alibaba Reverses Course on Qwen Open Weights✦ from: Alibaba's reversal to release Qwen 3.8 Max as open weights, after declining to release Qwen 3.7 Max in May, may have been influenced by Xi Jinping's recent speech encouraging open source.Ban Anti-Distillation TOS Clauses✦ from: The U.S. should pass a law that explicitly makes data collection for training models fair use, and bars terms of service that forbid distillation, at least for U.S. companies.Xi Jinping Speech as Possible Catalyst✦ from: Alibaba's reversal to release Qwen 3.8 Max as open weights, after declining to release Qwen 3.7 Max in May, may have been influenced by Xi Jinping's recent speech encouraging open source.
Related episodes
Audio · Dwarkesh Podcast

Eric Jang – Building AlphaGo from scratch

→
Audio · No Priors

Biohub: The Future of Biology is Open-Source with Co-Founders Mark Zuckerberg, Priscilla Chan, and Head of Science Alex Rives

→
Audio · No Priors

SAP: Bringing the ‘Operating System’ of a Company into the AI Era with CTO Philipp Herzig

→