Saying that ml models aren't open source because you don't have the data is like saying doom isn't open source because you didn't provide a proof the fast inverse square root works: https://en.m.wikipedia.org/wiki/Fast_inverse_square_root
LLMs are just very large magic numbers. That's all there is to them.
Whisper models can do flawless transcripts in real time. They can run locally on a laptop without proprietary software. Not having them in a tech talk is inexcusable today.