Home | Notifications | New Note | Local | Federated | Search | Logout

Note Detail

Reply to @sun@shitposter.world
SuperDicq@SuperDicq@minidisc.tokyo (2026-07-29 17:20:28)
@sun@shitposter.world @lain@lain.com The problem is that output from LLMs currently has a legally ambiguous status. You can not risk integrating it into any existing free software right now, so that's why a lot of projects, with some exceptions, choose not to allow it or heavily limit its usage only to specific tasks.

And on a personal note I think obvious LLM output is extremely tasteless. Seeing someone not even use some effort to at least even write their own README file makes me cringe. I always compare it to the same type of feeling you get when you see a corporate logo which uses Comic Sans.
---Reply---
Blurry Moon@sun@shitposter.world (2026-07-29 17:24:11)
@SuperDicq I think it's a bigger risk if you close source something, I am doing AGPL so like unless it pulled something from some weird license should be okay.

---Replies---

SuperDicq@SuperDicq@minidisc.tokyo (2026-07-29 17:32:51)
@sun@shitposter.world The risk for free software projects is actually the biggest, proprietary software can not be investigated that easily because their source code is secret.

As much as I appreciate you adding a proper copyleft license for this project, this unfortunately this does not matter much. The risk that your LLM pulls in code from other free software projects with an incompatible license is very minimal.

A much greater risk is that the LLM pulls in code from a proprietary software project. Because guess what? LLMs aren't trained from just from free software repositories obviously. Every person around the world that is using Cursor, Claude, Copilot, Codex, etc. while working on proprietary software projects is feeding their proprietary repositories into their ever growing training data set.