In favour of local AI: an incentive-based case for open-source models on your hardware

arpitingle · x · 2026-09-18

Developer Kush Bhardwaj published an essay arguing for local/open-source AI.

Premise: frontier LLMs concentrate in few hands due to hardware constraints; optimizations that let capable models run on consumer hardware would democratize intelligence while still pushing the frontier forward.
Core argument is incentives: frontier labs optimize across billions of users balancing profit, regulation, and their own moral compass — you're just a data point, not the alignment target.
Answering "what if it falls into the wrong hands?": a capable model on your own hardware, aligned to you, can protect you — often without you knowing.
Includes a formal model of capability (n units) allocated across population segments with weights pi to illustrate the misalignment.

More from AGI Musings