ALLi News Editor, Dan Holloway
What lengths would AI platforms go to in order to improve the AI training data for their large language models?
There has been much discussion of the deals tech platforms have sought and struck with publishers of creative work for the right to use that work. And the way many of these stories have played out has put the spotlight on the difference in size between the giants of the creative industries and those of tech. But no story reveals both the lengths a platform will go to and the disparity in size between the two industries as one that broke recently.
The New York Times received leaks from a former employee at Facebook’s parent Meta that show the company were spending a lot of time at the start of 2023 discussing the possibility of buying Simon & Schuster after the publisher’s deal with Penguin Random House was poleaxed by the courts over monopoly fears.
The purpose of the acquisition would, the leaked recordings of meetings make clear, have been to acquire the rights to Simon & Schuster's catalogue so as to use it to train generative AI.
It really is a fascinating insight into the way the technology industry views creative works and the rights their creators have to them. It’s very clear that creative work is perceived to be important to the training of platforms. It’s clear that those platforms are willing to pay large sums of money to access that work. It’s also clear that the value they place on our work is instrumental. They want to use what we do to help create bigger profits for themselves in the long term. And finally, it’s clear that if they can’t buy access to creative work, they will look to other avenues of procurement.
The Guardian’s account of this story makes the very interesting juxtaposition of this internal Meta discussion with a judge’s dismissal of a lawsuit against Meta for using copyright books. The judge claimed the outputs did not sufficiently resemble the inputs for a copyright violation case to be demonstrated. The fact this may be true and the intent of Meta may also be clear shows how much attention the current laws need.
