Legally, that changes things a lot, usually. Human and LLM learning are called the same more like a funny coincidence of the language rather than by any similarity.
A human needs to see 0 quicksort implementations to be able to do it, an LLM needs to see millions of them. The process is completely different and I think these comparisons are not helpful to the discussion. Il giorno ven 20 feb 2026 alle ore 20:54 Theodore Tso <[email protected]> ha scritto: > > On Fri, Feb 20, 2026 at 08:26:08PM +0100, Lucas Nussbaum wrote: > > As a made-up example, see each commit in > > https://github.com/lnussbaum/llm-refactor-example/commits/main/ : > > the caesar_cipher() implementation introduced in the last commit can > > probably be found in the training data. > > As another hypothetical thought experiment, suppose the problem is to > optimize a program which has a bubble sort, and a human programmer is > asked to optimize it by replacing it with a quick sort. > > There are only so many different ways to code the quick sort algorithm > in C, and it's likely that the human being might even be vaguely > remembering how they saw it done in some non-free source code (for > example, in Sedgewick's Algorithms book) and perhaps, subconsciously > reproduced it from some non-free source that they ocne saw. > > Would that be problematic? I very much doubt it, even if it looked > frighteningly similar to the one found in AT&T's proprietary Unix > code. > > Now replace "a human being" with "an LLM". How does this change the > calculus? > > - Ted > -- Salvo Tomaselli I difensori della morale tradizionale sono raramente persone di cuore. Si è tentati di pensare che essi si servano della morale come di legittimo sfogo al loro desiderio di fare del male agli altri. -- Bertrand Russell, Perché non sono cristiano. 1957 http://ltworf.github.io/
