I heard a podcast today where the host was talking about Apple licensing the OpenAI code. He insisted that would be unethical because the code is tainted by unethical training.

I asked ChatGPT if that is true; it said it is not.

However, when I asked in a slightly different way, it said “While the immediate issue of data sourcing is less prominent when licensing only the code, the code (i.e., the model's architecture and parameters) could still carry biases. These biases could have been introduced during the training process based on the original dataset. Thus, even without the original data, the behavior of the model might reflect biases present in the data it was initially trained on.

Use and Implementation: The ethical implications of how the code is used remain significant. The licensee, such as Apple in this hypothetical scenari, would need to conduct thorough testing to identify and mitigate any biases or other ethical issues in how the code operates, regardless of the data it is subsequently trained on.”

I wish you would write about this and clear it up either way.