Egregoros

cc @jonny who has been digging into their AI fiasco lately...

@liaizon
Sure yeah the open weight models are free for everyone who can afford $22,000 worth of computer

@jonny Open Weight (as in you are open to weigh out all the gold bars you have to drop on silicon)

@liaizon @jonny fwiw I know somebody that vibe coded an entire activitypub server on qwen3.8 which runs on a more reasonable machine so you don't strictly need two insanely expensive mac studios to do that

@sun @jonny @liaizon can't you theoretically use a small-medium imodel to code something if you already have a strong, solid plan?

@sun
@liaizon
I too have made my fair share of qwen 9B oddities, very strange lil model. I would bet it could make something like a fedi server with a few days of generation time on a laptop

@mischievoustomato @jonny @liaizon the better you are at managing a human software project using testing and automation and programming by contract, documentation, the better your llm output will be, but results will vary

@jonny @sun @liaizon I am building an apparmor tool that uses qwen to generate profile update suggestions. I haven't had much luck using it in a harness but I think that's mostly because opencode and hermes stuff the context window with a bunch of stuff and it can't reach the user prompt.