When I read Reisner’s piece, what stood out most was how fast Meta turned to shadow libraries to train Llama. They needed a massive volume of books, but buying rights was both slow and expensive. Instead of waiting, they tapped into LibGen — a site filled with copyrighted works uploaded without permission. That move feels like a line crossed. Authors spend years crafting their books, and using them without approval or payment takes away their agency. It’s no surprise many writers described the practice as devastating, and I share that view: building AI on stolen work undermines creators.
Meta’s defense is that this counts as “fair use,” since the model processes text into something different. They also argue that licensing everything would take too long and cost too much. I understand the temptation, but to me, that comes off as justification rather than a real solution. As one writer put it, AI isn’t producing anything truly new — it’s remixing what already exists. Rebranding it as innovation doesn’t change that fact.
If I had been the one deciding on Llama’s training data, I’d feel the competing pressures clearly. On the legal side, copyright lawsuits are a constant threat. On the ethical side, it signals that creators’ rights don’t matter — an especially bad look for a company already facing scrutiny. And once the public learns about it, the backlash lingers. Yet the business case is real: clean datasets are slow and costly to assemble, and competitors are moving fast. That tension makes the issue messy.
Even so, I’d prefer to take the slower, more responsible route. That would mean paying publishers where needed, leaning on public-domain and open-licensed works, and creating systems where authors share in the benefits. I’d also add safeguards so the model can’t spit out large passages of copyrighted text. More importantly, I’d want to keep dialogue open with writers — offering opt-in paths or revenue-sharing — so they feel included rather than disregarded.
In the end, this isn’t only about copyright law. It’s about fairness and trust. Innovation matters, but it loses credibility if it comes at the expense of the very people whose work makes it possible.
