Opensource

An open source PR is now a donation in two currencies: code, and the LLM tokens that produced it. I figured this out the slow way, needing to patch a Go CLI in an hour, in a language I don’t write. In a previous era, this is where the story ends: file a feature request, hope someone with the right neurons gets bored on a weekend, or shrug and ship a workaround. Instead, in one hour with Claude Code, agents inventoried the upstream, surfaced three reference implementations that had already solved the underlying problem, located the right hook in 1,200 lines of Go I’d never seen, and wrote the patch. Along the way I (or they, really) noticed 22 other forks of the same project, only 4 of which had ever shipped a contribution back. I almost became the 19th wheel-reinventer. Instead I filed the patch back upstream with a working PoC. ...

What a great 4th meeting of the [Gen]AI Italy meetup yesterday in Milan, where I joined an A-Team of speakers by FairMind who delved deep into open-source models and architectures for GenAI, moving from theory to real-world implementation. 🔹 Alexio Cassani kicked it off with a survey on the state of the art of open models. Starting from a thought-provoking exploration of “Open… What?”, he challenged our assumptions about open-source LLMs while pressing an important point: open model training, evaluation, and deployment architectures are crucial for LLM research to move forward outside walled gardens. This sustains a vital open model ecosystem that enables data protection, IP retention, and ultimately, business sustainability of GenAI solutions. ...