NYT wants to outlaw a math game created by calculating the probablities of word groupings and words following each other in NYT times articles, along with a lot of of other writings NYT does not own. The players roll the dice, so to speak, by seeding an initial string of words and whoever comes up with the most interesting paragraph wins. This paragraph may or may not look like NYT times writings, which in the larger scheme the collected writings of humankind, isn't particularly unique. It doesn't even have to be true. Hallucinations are an expected outcome.
If a NYT article says "Henry Kissenger was known to eat ice cream on a hot day" and our game outputs the same, it is purely by chance. It cannot be proven the output was copied verbatim from the NYT because the fragment "Henry Kissenger was known to" and "eat ice cream on a hot day" are not unique to the NYT or exclusive to it.
Is the NYT claiming ownership of the weights in LLMs?
If a NYT article says "Henry Kissenger was known to eat ice cream on a hot day" and our game outputs the same, it is purely by chance. It cannot be proven the output was copied verbatim from the NYT because the fragment "Henry Kissenger was known to" and "eat ice cream on a hot day" are not unique to the NYT or exclusive to it.
Is the NYT claiming ownership of the weights in LLMs?