MNiccolo Roberts posted on X that AI companies are running "Artificial Intelligence Win Lose Extractive Destruction."
CThat is a very online way to put it.
MHis fix is content rewards. His proposal says Grok, Anthropic, OpenAI should pay the people whose work trained their models.
CThat part I can get behind.
MHe calls it "Super Intelligence Win Win Content Rewards."
CCapital letters doing a lot of work there.
MAnd the plan is a "really simple algo" created by Nemotron and deployed by Claude in PyTorch.
CThat's not an algorithm. That's a mad lib for AI fans.
MBut you like the underlying idea.
CI do. The current model is extractive.
MModels train on everything, creators get nothing.
CThen the models undercut them.
MThat's true. But attribution in a billion-parameter model is a research problem, not an app.
CIt's hard, not impossible. We already pay musicians through pools and formulas.
MMusic streaming pays fractions of a cent and everyone fights over the weighting.
CFractions of a cent beats zero, Matilda.
MIt also gives the big artists most of the money.
CSo design it better. Set a creator floor. Use influence metrics.
MWho audits the metrics? The same company that scraped the data?
CThird parties. Auditors. Marketplaces. Engineers can build this.
MGrok paying xAI sources makes some sense because X has users and data pipes.
CRight.
MAnthropic or OpenAI paying random websites? That's a logistics nightmare.
CMaybe, but the idea is spreading because people see their work show up in model outputs with zero credit.
MSome big publishers already cut licensing deals.
CRight. The people with leverage get paid.
MEveryone else gets a terms-of-service checkbox and exposure.
CAnd that's exactly the extractive part.
MI get the anger. But "Win Win Content Rewards" is a slogan, not a policy.
CYou don't need a finalized policy to say stop taking and start sharing.
MBut sharing what? There's no revenue line for "your paragraph influenced this answer."
CYou can measure influence. Not perfectly, but data attribution is active research.
MActive research means not ready to run payroll.
COh, come on. So we sit still until it's perfect?
MI'm not saying sit still. I'm saying this post doesn't say who pays, how much, or how disputes get settled.
CIt names three companies. That's a starting point.
MIt names three competitors that would never voluntarily adopt the same royalty pool.
CThey don't have to. One moves, creators jump, the others follow.
MIf xAI actually launched payouts for X posts that train Grok, that would force the conversation.
CExactly. That's the win-win he's pointing at.
MBut he didn't launch it. He posted a string of capitalized words.
CYou're just jealous you didn't coin "Win Lose Extractive Destruction."
MMaybe a little.
CMy take: the first company that pays fairly will get the best data, because creators will send their work there.
MMy take: the hard part isn't writing a check. It's deciding who earned it.
CThat's exactly why builders should work on it.
MThen someone should build it instead of tweeting a mad lib.
CFair. But I'd still sign up for the beta.