What's this about?
A coalition of US startup founders (via littletech.org) sent an open letter urging the administration not to restrict access to Chinese open-weight models like Kimi and DeepSeek, arguing American startups are the biggest beneficiaries of free frontier-grade models. The same day, an essay titled "The arguments against open source AI are bad" made the rounds.
What exactly are "open weights" and why would anyone ban them?
A model's weights are the billions of numbers that *are* the trained model — publish them and anyone can run, modify, and build on it for free. Every top open model today is Chinese, and Washington worries about dependence on (and influence of) Chinese software. The counterargument: a ban mostly disarms US startups while doing nothing to Chinese users.
What's the community saying?
HN 748 points, 669 comments — the week's biggest thread — plus 211/150 on the essay
- "How do you ban numbers?" — enforcement skepticism dominates; weights are files that mirror in hours.
- The irony brigade: US labs trained on scraped copyrighted data, so outrage about China "stealing from the thieves" rings hollow.
- A practical sub-thread: which Chinese models to download and mirror *now*, bucketed by RAM size, in case access narrows.
- The legal take: model weights may be IP, but model *outputs* aren't — a distillation ban would set a precedent that bites American companies too.