4 ms·
> Blanchard's account is that he never looked at the existing source code directly. He fed only the API and the test suite to Claude and asked it to reimplement
by smsm42 7mo ago
> Blanchard's account is that he never looked at the existing source code directly. He fed only the API and the test suite to Claude and asked it to reimplement the library from scratch.
I don't see how it matters what he looked at. If I took a copyrighted code and run it through a script that replaces all variable names, and then claimed copyright on the result because it's an entirely new work and I did not look on the original work, I'd be ridiculed and sued, and would lose that lawsuit. AI is a more complex machine, but still a machine. If you feed somebody'd work into a machine, what comes out is a derivative work.
Test suite is a part of copyrighted code, is it not? If he used just the API description, preferably from a copyright-clean source, then we could claim new work (regardless of how it was produced, by using Claude or trained pigeons or by consuming magic mushrooms). But once parts of the copyrighted code had been used, it becomes derivative work.
- metalcrow 7mo ago> AI is a more complex machine, but still a machine. If you feed somebody'd work into a machine, what comes out is a derivative work. I'm not sure that's true, legally speaking. If you fed it into a PRNG, the output seems to me like it would not be an obviously derivative work (i doubt you could copyright it but that's a separate question). So we have 1 machine that can transform something into non-derivative work, and another that leaves the result derivative. The line isn't likely going to be drawn as "did a machine do it or not", but on a fuzzy human line of how close the output seems to be to the original (IANAL).
- smsm42 7mo agoPRNG is not intended to use information sent into it in a substantive way (in other words, for PRNG does not matter if what you feed into is is Shakespeare's sonnets or white noise). Sure, if your machine is an electronic analogue of a shredder, then yes, the result is not a derivative work in any sensible meaning. But LLMs are not that kind of a machine.
- metalcrow 7mo agoI agree! But that's exactly my point: 1 type of machine is ok, another isn't, so it is not just a matter of all machines make derivative work. To draw the line more carefully is an open question. I would be surprised if a machine that "uses information sent into it in a substantive way" is a perfect deliminator. OTTOMH musicians might present some compelling objections.