I'm a know-nothing when it comes to LLM internals or copyright law, but presumably, if I as a human, read a work of literature (say JRR Tolkien's collected works) and generated something like Wheel of Time (which is literally what Robert Jordan did, he has even credited Tolkien for his influence, even stated that the initial chapters were modelled around Tolkien's Shire), then everything is admittedly kosher.
Then why is it wrong if an LLM does the same?
I am assuming of course that OpenAI licensed / bought the literature legally as a prolific human bookworm would have done.
As long as the LLM generates output that is distinct and different enough from the original work it was trained on, shouldn't this be entirely fine from a legal perspective?
reply