Travis Kalanick Questions Chinese Distillation of US AI Models and IP Policy
On X, Kalanick argued that uneven rules around AI model distillation could leave American companies at a legal disadvantage.
Entities: travis kalanick
In a post on X, Travis Kalanick argued that Chinese AI models are distilling American ones and asked what U.S. policy should be if American companies cannot do the same without legal risk. He framed it as an asymmetry: either distillation rules get enforced across the board, or everyone should be allowed to do it. In a follow-up post on X, he argued that Chinese business norms can be less sympathetic to firms that fail to protect their own technology.
Combined views
273K
5 posts, first seen 4h ago
Reactions from ranked influencers
5 posts
.@DavidSacks @elonmusk The Chinese models are distilling off American models… my guess is American models can’t legally do this without being sued by Chinese AI companies in American courts .. 🤔.. and distilling a Chinese model is really theft from an American company… What should American AI and IP policy be here? If distillation isn’t enforced against, then everyone should be able to distill from everyone else.. otherwise one arm tied behind American models’ backs. Maybe systems need to get better at preventing distillation techniques, and problem solves itself?
There’s a large grain of salt!🧂 Kimi is still calling itself Claude.
In my experience, China rolls different on these topics… if you’re in China and you tell a business colleague “dude, they frauded me, they took my stuff!!”… the natural response is.. “why’d you let them?!.. get your shit together”… <loose translation of convo in mandarin>
JUST IN: China claims allegations that its AI firms illicitly distill U.S. models are “misguided and counterproductive.”
lol they deliberately never challenged the allegations on fact or merit
JUST IN: China claims allegations that its AI firms illicitly distill U.S. models are “misguided and counterproductive.”
@travisk @DavidSacks @elonmusk Elon admitted to distilling Claude, and Thinly recently also said they used Kimi to bootstrap. This issue is quite overblown.
@travisk @DavidSacks @elonmusk
These innovations weren’t distilled: * KDA hybrid linear attention (for efficient long context scaling) * Attention Residuals (a more efficient memory retrieval) * Stable LatentMoE (only 1.8% experts will be activated each time) * Quantile-balanced routing (infra optimization)