Apparently, both are diving into the world of Large Language Models (LLMs) by training them on pre-existing intellectual property.
Lionsgate is using its vast catalog of films, TV shows, and other content they own or control to train their LLM.
LinkedIn is planning to train its LLM on the posts and content created by users like you and me.
Here’s where it gets interesting…
In the United States, the exclusive rights to a copyrighted work (like a LinkedIn post written by me) automatically attach when it’s fixed in a tangible medium (including digital formats).
Copyright belongs to the creator unless the work is made as a “work made for hire” during employment.
A “work made for hire” means the copyright automatically belongs to the employer if certain conditions are met.
The Copyright Act of 1976 outlines 11 situations where a work created by an independent contractor or non-employee qualifies as a “work for hire.”
One such situation is contributions to an audio-visual project like a film or TV show. With proper agreements, the copyright from a cinematographer’s choices or an actor’s ad-lib belongs to the producer.
But here’s the kicker…
The posts you and I make on LinkedIn are not “work for hire” benefiting LinkedIn.
The current User Agreement clearly states:
“As between you and LinkedIn, you own the content and information that you submit or post to the Services, and you are only granting LinkedIn and our affiliates [a] non-exclusive license.”
Sadly, for the talented filmmakers, creators, producers, and craftspeople behind Lionsgate’s projects, the company can now use Generative AI to create new elements for spin off works from the originals—without needing to ask for permission or offer any extra compensation (except for writers who are in the WGA).
But LinkedIn is different.
The law around using preexisting copyright protected works to train LLM is not yet settled.
There are roughly 25 lawsuits pending in the United States alone asking the courts to decide whether or not using copyright protected works without permission to train LLMs is copyright infringement.
LinkedIn did not get permission from its users when it quietly launched the training model earlier this month. Rather, it has made a post in the Help Center advising users on how to OPT OUT. Learn how here: https://www.linkedin.com/help/linkedin/answer/a6278444
That seems rather disingenuous to me.
LinkedIn could have easily sent a message to every user on the platform advising them of their plan, and providing a way to OPT IN.
Comentários