Post #2702966
2026-04-03 01:19 UTC
Replies (2)
-
@Forbearance@mastodon.xyz 2026-04-03 01:22
@protowlf@mastodon.gamedev.place I can see this making sense by a sort of "conservation of thought" argument: if those big matrices are definitionally devoid of the spark of original thought, then anything that comes out is definitionally derivative, copied from somewhere even if the sources are so scattered and numerous that they are permanently unidentifiable.
-
@agersant@mastodon.gamedev.place 2026-04-03 01:29
@Forbearance@mastodon.xyz @protowlf@mastodon.gamedev.place The training data is full of open-source code that has software licenses attached to it (GPL, MIT, etc.). These licenses have terms for what you can or cannot do with the source of software, or things you must do to be in compliance. For example, the MIT license (one of the most "generous" popular licenses) requires people making use of the code to distribute a copy of the license alongside their software. That's similar to giving credit in the art world.