Post #2135888

2026-05-06 10:55 UTC

Question for the #authors: let&#39;s assume a bunch of big #publishers decided to train a new #LLM on their pooled catalogue with consent of the authors. Would you see the use of it acceptable for light #editing purposes? I.e. finding inconsistencies, tone issues, … #poll #creative #writing #writingcommunity #bookstodon #books

Replies (2)

@yoasif@mastodon.social 2026-05-06 13:16

@pludikovsky@chaos.social Isn't the question about whether it is acceptable more about whether you want to write mediocre, lowest common denominator pap? No matter how the chatbots are dressed up, they aren't actually smart, so if the rationale is to improve your writing, why would developers not work to produce a product that does that directly, instead of shoehorning an LLM into the role?

Open ##3991210
@makepost@mastodon.art 2026-05-06 13:48

@pludikovsky@chaos.social Won't big publishers force the decision on authors, not ask for consent? Also, isn't the sheer quantity of unlicensed training material the reason LLMs output anything cohesive? Hypothetically I'd like more authors and translators to volunteer works to OpenNMT to improve libre machine translation. (I haven't researched how to do that yet.) I remember Google Translate gamifying the training of its model; I'd participate in such a project if done by libre software authors.

Open ##3991213