3 ms·
It's not that cut-and-dry. If I say to one of my human friends "write me a nursery rhyme", the copyright of the resulting rhyme would obviously belong to my fr
by crote 3y ago
It's not that cut-and-dry.
If I say to one of my human friends "write me a nursery rhyme", the copyright of the resulting rhyme would obviously belong to my friend - despite me prompting them. Clearly the prompt itself does not universally count as "making" it.
Let's say I made a "NovelSnippetAI", which contains a corpus of prewritten material. You can prompt it, and it will return a page which matches the sentiment of your prompt best. I think we can agree that the copyright of the page will still belong to the original writer - the user only did a query.
What if I did "NovelMixAI", which did exactly the same but alternated lines from the two best matches? What about "NovelTransformAI", which applied a mathematical formula to the best match and fed the output to a fixed Markov Chain? Now we're suddenly at "LLMAI", which does the same using a neural network - what makes it different from the rest?
Long-standing precedent is that any automated work does not qualify for copyright. You can only copyright human work.
- throwup238 3y ago> If I say to one of my human friends "write me a nursery rhyme", the copyright of the resulting rhyme would obviously belong to my friend - despite me prompting them. Clearly the prompt itself does not universally count as "making" it. This is the wrong analogy. The LLM is more like Photoshop and the prompt little more than a filter configuration. A machine cannot copyright its own output but a human guiding that machine can.
- ipaddr 3y agoIt is more like a child's speak and say. Push a button a cow noise comes out. Someone else pushes it the same noise comes out. Even if a billion buttons exist the child doesn't own the cow noise.
- crote 3y agoNo, it's the right analogy for the point I was making. That analogy is there to point out that it is about more than just the prompt, and prompting a machine is covered in the rest of my comment.
- qotgalaxy 3y ago[dead]
- tomatocracy 3y agoWhere the line is - ie how much human creative input would be needed for a work to be covered by copyright seems to be unclear legally. There are some interesting parallels to this dispute I think: https://en.m.wikipedia.org/wiki/Monkey_selfie_copyright_dispute https://en.m.wikipedia.org/wiki/Monkey_selfie_copyright_disp...
- jwestbury 3y agoAnd what happens when two users manage to get the same result? Does OpenAI need to implement an "/r9k/ algorithm" or something to prevent sufficiently similar results from ever being generated, because once a response is generated it is now copyrighted by the user who prompted it?