I reject the classification of any of these LLMs as open source. If an LLM’s training data isn’t bundled with it under the same license then that LLM is not open source by any reasonable definition of open source. The Open Source Initiative (industry group with the objective of co-opting FOSS into free labor for capitalists) is trying to push this ridiculous conception that if you can run a LLM locally then it is open source. It makes absolutely no sense that an impenetrable list of weights counts as source code. Those weights are the derived product of the training data and the learning algorithm, which is readily analogous to compilation. Here’s a good longer post about this: https://www.linkedin.com/pulse/why-open-data-necessary-source-ai-tom-callaway-stzcc/
I reject the classification of any of these LLMs as open source. If an LLM’s training data isn’t bundled with it under the same license then that LLM is not open source by any reasonable definition of open source. The Open Source Initiative (industry group with the objective of co-opting FOSS into free labor for capitalists) is trying to push this ridiculous conception that if you can run a LLM locally then it is open source. It makes absolutely no sense that an impenetrable list of weights counts as source code. Those weights are the derived product of the training data and the learning algorithm, which is readily analogous to compilation. Here’s a good longer post about this: https://www.linkedin.com/pulse/why-open-data-necessary-source-ai-tom-callaway-stzcc/