It’s like saying that Microsoft Windows is open source because you can run the binaries on your own computer.
At best it’s a misunderstanding of what open source is, at worst it’s a wilful misinterpretation to try and paint these models in an undeserved better light.
I think it’s a little different from that. Of course it varies on a case by case basis depending on exactly what is released, but if the model architecture is documented and you are able to train or fine-tune it with your own data, I think that counts for something, even if you don’t know exactly what their released model was trained on. If all the code is available to produce the model, just not the dataset, I would still consider that open source.
It’s not a stretch, it’s entirely false.
It’s like saying that Microsoft Windows is open source because you can run the binaries on your own computer.
At best it’s a misunderstanding of what open source is, at worst it’s a wilful misinterpretation to try and paint these models in an undeserved better light.
I think it’s a little different from that. Of course it varies on a case by case basis depending on exactly what is released, but if the model architecture is documented and you are able to train or fine-tune it with your own data, I think that counts for something, even if you don’t know exactly what their released model was trained on. If all the code is available to produce the model, just not the dataset, I would still consider that open source.