this post was submitted on 22 Oct 2023
21 points (95.7% liked)

Futurology

1813 readers
41 users here now

founded 1 year ago
MODERATORS
voidx
Lugh
Espiritdescali
AwesomeLowlander
21
Researchers outline an easy way for LLM's to erase copyright material from their training data, without completely retraining them. (arxiv.org)
submitted 1 year ago by Lugh to c/futurology
1 comments fedilink hide all child comments
top 1 comments
sorted by: hot top controversial new old
[–] SchizoDenji@lemm.ee 1 points 1 year ago

A lot of LLMs are already out there in open source for anyone to embed or reinforce themselves, so whoever wants to shackle their own llm and lose ground, they are free to do so but they won't.