Weights are an astoundingly unsexy name for the most important part of an AI model: billions of numbers, set during training, holding everything the model knows. Nobody can tell you what any individual one does, but collectively they are the model. Copy the numbers and you've copied the thing.
This is why "open weights" is the phrase that actually matters, and why it isn't the same as open source. Releasing weights hands over the finished model — you can run it and fine-tune it — but usually not the training data or the code that made it.
