1. Should writers be compensated for training data?

Definitely, 100%. Both Dario and Sam have talked about wanting to do this, and I very much hope it happens. Writers absolutely deserve to be compensated when their work is used. It’s a huge complicated question (especially figuring out how to apportion fees and to whom). But of course it should happen.

2. Does that mean that AI is theft?

I don’t think so.

The strongest version of the argument against AI training is: “you used my book to help create a product that you’re now selling.” But that mischaracterizes how LLMs actually get trained.

An AI model doesn’t store or reproduce anyone’s writing. It builds a statistical model of language patterns based on millions of works. To me, this feels less like plagiarizing someone’s writing and much more like what we do when a human writer reads books and develops their own voice.

More importantly, what comes out the other side bears very little resemblance to any individual book it trained on. It’s something entirely new: a general-purpose reasoning tool that enables you to access the sum total of human knowledge. That’s not a bad thing. It’s a wonderful thing.

Does that mean writers shouldn’t be compensated? Of course not, they 100% should. But just because we haven’t yet figured out how to make that happen doesn’t imply that it’s wrong to use AI.

Also, “writers deserve to be compensated” gives AI companies something specific they should be doing differently, and helps open the door to negotiation and fairer frameworks.

“AI companies are evil and AI is theft” actually does the opposite. It lets AI companies off the hook because their only possible response is to be like, “So you’re asking us to simply cease operations? We’re not going to do that.” It just creates outrage and doesn’t help change anything.

3. Should you boycott AI until writers get paid?

If there were a world where boycotting LLMs might actually force AI companies to compensate writers, then it would be a meaningful ethical choice. The way boycotting Nike pressured them to address labor practices in their supply chain. Nike boycotts worked because the conditions were right: a concentrated supply chain, a brand-sensitive consumer company, and something specific that Nike could actually do that would lead to a more ethical outcome.

None of those conditions exist here. AI companies already have billions of users. Most of their revenue comes from enterprise API customers, not individual subscriptions. They already don’t have goodwill from consumers. Writers boycotting ChatGPT is like trying to protest Amazon Web Services by canceling your Netflix subscription. Consumers are not in the value chain that matters.

The actual levers that will determine whether writers get compensated are litigation and legislation. The courts are actively working through these questions right now. Public pressure campaigns directed at lawmakers would be a way to actually make a difference.

If our goal is to actually help writers get paid for their work, these are the levers that can really change things.

I understand why anyone would want to boycott AI. “I don’t feel comfortable participating in a system I consider exploitative.” That makes sense and I respect that choice.

But it’s absurd for those people to then act like if someone else uses AI, that somehow makes them part of the problem, as if an individual writer using or not using AI makes any difference at all.

If we care about writers, then we should direct that toward actions that actually help writers.

Donating to legal funds fighting for creator compensation, advocating for legislation that establishes fair frameworks, amplifying writers who are doing ethical AI policy work. Those are things that could make a difference. Refusing to use a chatbot, or shaming people who do, doesn’t make a difference. It’s just virtue signaling that has no actual impact, and it comes at the cost of giving up a genuinely powerful tool for helping elicit your own best creative work.