Something I've been thinking about recently, and I'm curious to hear how much you agree: Calling most free LLMs 'open source' feels misleading. Sharing weights and architecture without the training data, hyperparameters, or training scripts is like distributing a compiled binary while claiming the software is open source. The training process is the real source code (or at least, a big part of it) — without it, these models are just a black box with an open interface, not truly open. Just a thought...