On Fri, Feb 20, 2026 at 09:40:33PM +0100, Christian Kastner wrote: > > Or, to use a practical example: When I design some Python class > hierarchy, that design will be influenced by all of the experience I > have accrued reading or using other hierarchies. But unless I copy a > specific one, I don't think anyone would argue that my work is a > derivative (in the legal sense) of all those other examples. > > Why should this be different for an LLM?
Exactly. There are some humans who might code something that takes a little from example A, some from example B, and some from example C, where examples A, B, and C were things that said human may have seen years ago, and where they might not even have concsious memories of having seen these examples. There are many times when we may be closer to a automated pattern matching automaton like an LLM rather than rederiving everything from first principles. (This is the observation which explains unconscious bias, for which there is experimental evidence[1].) [1] https://implicit.harvard.edu/implicit/takeatest.html There might be other human beings who might start with something that the copy-pasta from Stack Exchange, but then they make changes to variable and parameters, in an attempt to "file off the serial numbers". And there might be other human beings who do a full copy-pasta without trying to rename variables or even changing the white space to match the destination project. And they might not even check the license, and so they could end up transplating code from GPLv2-only source into GPLv3 codebase. Now, there are *some* LLM's that might do that last thing, which is clearly wrong, regardelss of whether it is done by a human or an LLM. But they could also do the first example where they are merging examples from multiple sources to create a new implementation. We don't say that just because human beings *might* do something blatently violative of copyright, that human behings should be prohibited from ever contributing code to Open Source Software. And yet, there are people making the argument that we should do this for LLM's.... - Ted
