The Case for the Proposal
The author's argument comes in three layers. Privacy is the first: every conversation sent to a cloud model passes through someone else's servers, and in sensitive scenarios that's a matter of principle, not preference. The second layer is control: cloud services raise prices, change terms, and cut models, and your workflow is built on a foundation the other party can move at any time. The third layer is feasibility: the miniaturization of open source models and the progress of consumer hardware have already turned running models locally from geek behavior into something ordinary people can reach with a little stretch.
Reality's Discount
The corrective opinions in the comments are worth reading alongside. The capability gap between local models and frontier models is still real, and forcing local on complex tasks is making trouble for yourself; the hardware threshold isn't that low for most people either, and a good-enough GPU is still an investment of a few thousand yuan. The more widely accepted compromise is tiered usage: everyday summarizing, rewriting, and simple Q&A go local, and heavy work goes to the cloud. On this direction, vendors are actually running ahead of public opinion—on-device models in phones and computers are landing faster and faster, and "local first" may not win by ideals but be carried into reality straight by product defaults.
via: Hacker News