Running local models is actually good now

Vicky Boykis put together a long-form post about running AI models locally — and the TLDR crew picked it up as one of the week's best reads. The piece walks through what's changed: you no longer need a $5,000 GPU to get useful results, and the software stack has caught up.

What matters is the shift. Cloud models are great when you have a steady stream of traffic and someone to manage the bills. Local models are for the people who want to own their data, who don't want to pay per call, who want to run things on their own machines. The hardware is cheaper. The tools are easier. You can train your own models now without becoming a full-time engineer.

Boykis also mentions emailing complete strangers — a small thing that says something bigger. When models run locally, they stop being a service and start being a tool. You stop talking to a company and start talking to your own machine. La gente who work in side businesses, who run their own sites, who don't want to pay $0.01 every time they ask something — this is for them.

Why this matters for us: the people building things outside of Silicon Valley just got a lot more options, and the cost of doing it on our own terms keeps dropping.