The @QVAC_tether release from @Tether_to could be one of the most important levellers of the AI playing field. It opens up huge possibilities for people that don't want to rely on US test time compute hyperscalers or chinese platforms. Moving away from the pytorch + A100s capital intensive approach towards device agnostic, private fine tuning allows anyone to do personalization with llama.cpp, which was previously not possible without cloud depedency. This is the sort of technology that stops us being under the boot of whatever AI hypercompany decides to go bad. Own your money, own your AI. Great to see privacy and sovereignty making real steps forward.