Everyone is using the term "open source" completely wrong in the context of AI models. The source code for a model is the input data and the training algorithm. Calling it open source because you can download and run it locally is like calling Microsoft Excel open source because you can install it. #ai
Login to reply
Replies (2)
in open weights llms, all the data is really there. its not even encrypted. its just encoded in a format that is difficult to understand.
I'm not sure that's true, the model training is like a highly lossy data compression. And the weights may as well be encrypted in terms of not being able to interpret them.