Chinese company Moonshot AI don release new version of dem Kimi model dis week, and e don cause wahala for tech world.
Sol, di new open source model, show say e get front-level performance for all di tests wey dem run am, and e dey beat oda tested models.
Vals AI also tok say Kimi dey compete with flagship frontier models.
Di announcement, wey happen same time as speech from Chinese president Xi Jinping for World AI Conference for Shanghai, don make Wall Street fear, as Nasdaq drop about 1% on Friday because investors dey sell stocks for chip companies like Nvidia.
Many posts from tech industry figures wey come out go sound familiar to those wey remember di debate after anoda Chinese company, DeepSeek, release dem open source R1 model for January 2025.
But now, everything don dey worse because of Trump administration tariff war with China, repeated fights over national security threat wey Anthropic supposedly pose, and as major AI companies dey prepare to finally go public.
For example, David Sacks — wey be Trump administration former AI czar and now co-chair of President’s Council of Advisors on Science and Technology — contrast Kimi progress with United States wey dey tie itself for knots: politicians and bureaucrats dey ban new data centers, pile on state regulations, and push for new federal agencies to pre-approve frontier models.
He say dis na how you go lose di AI race.
Di news also give am excuse to take dig at Anthropic, calling Claude example of “woke lobotomized models” wey be “di enemy American competitiveness.”
American AI models.
If distillation no dey enforced, den everybody suppose fit distill from everybody else.
American models back, Kalanick write.
Of course, American models also don build on top of Chinese ones, specifically Kimi.
Meanwhile, OpenAI head of strategic futures Dean Ball say Kimi na “a very good model” wey performance probably no fit explain away by distillation or anyting like dat, and he add say he dey personally surprised di Chinese state continue to allow di open sourcing of models dis good, given potential risks.
For fact, Ball suggest say probable outcome of open-weight-model-dominant world na full AI communism, where AI dey treat as public good wey di state go provide as kind of digital public infrastructure.
Dis future strike am as dystopian hellscape, but he never meet open-weight models advocate wey no eventually concede say dis na where things dey end.
He even suggest say Trump administration (wey he used to work for) go eventually realize say e need to create large amount of regulatory risk around di use of open-weight Chinese models.
You no need to ban open source (one of di dumber motifs of AI policy discussion), Ball say.
You just need to direct every agency to issue soft law wey go create FUD (fear, uncertainty, and doubt).
Federal Reserve Advisory Bulletin find say there may be backdoors for Chinese AI models.
E needn’t be dat well justified.
You just create enough regulatory risk so dat every regulated enterprise go back off.
However, Shakeel Hashim, editor of AI-focused publication Transformer, argue say much of di worry dey overblown, both because Kimi likely no get dangerous cyber capabilities, and because Chinese government go face extremely similar incentives to restrict open Chinese models once dem develop those capabilities.
