@MalmSanta yeah haha this was a hopeful / "hyperstition it into existence" post lol.... it'd be cool to have a framework that wraps open models to fine tune w desired / likely use cases -> distill & push smol model to execute locally & dispatch to big model when the desired use cases occur
More notes on nearby ideas
Tweet · Oct 16, 2023I keep wanting a local-inference-sized multimodal dispatcher model that can consume streams of input and decide which parts should be sent to bigger…Reply · May 27, 2024@RokoMijic i think we can go further w efficiency @ inference time, getting better model performance from smaller models that run locally ... and that…Quote tweet · Feb 12, 2024yep. models dispatching to other models, spawning narrow models, learning from other models, competing with other modelsReply · Jan 21, 2026@beffjezos hopefully she's diversifying across multiple models n providers including at least some local inference!