CGood morning, it's Sunday, just after six. First up, a post on X proposing AI companies pay for the work they train on.
MI like the headline. What's the actual idea?
CNiccolo Roberts says if Grok, Claude, or ChatGPT trains on your writing, your code, your photos—you should get a cut.
MThe internet's about to become a very small number of people billing a very big robot.
CHis argument is that right now it's extractive: they scrape, they train, they profit. The person who wrote the manual gets nothing.
MBut how do you measure which words mattered? A million pages go in, one model comes out. You can't just split the profits by word count.
CHe thinks a simple algorithm could handle the payout math, using tools like Nvidia's Nemotron and Claude.
M"Simple" is doing a lot of work there.
CThe direction matters, though. Music and movies built licensing. The web never did, and now AI is vacuuming it up.
MBecause the web was built on free. You put a paywall on every scraped post and the open web gets expensive overnight.
CIt's already expensive for creators! They give away the raw material while a few companies turn it into huge products.
MThen the immediate fix is transparency: say what you trained on, let people opt out. A fair payout plan is still years away.
CYears away doesn't mean wrong. Streaming figured out music payments eventually.
MSure, it figured out payments that artists complain about. But it built a system.
CExactly—a flawed system is better than no system.
MA different AI fight now. David Akpovi posted on X asking: "Is the AI honeymoon over?"
CFeels like the honeymoon ended when the bill arrived.
MHis post points to financial markets questioning whether massive AI investment will actually pay off.
CThat's the right question. Spending is enormous, and the revenue story isn't keeping up yet.
MBut the tech still works. The product roadmaps are still moving.
CSo the answer is: the honeymoon isn't over, but it's no longer all-inclusive.
MThat's the morning brief.