AI labs are buying data from dead startups to train their models
There's something poetic about a dead crypto startup's code becoming the brain food for an AI that might outlive us all. Miskell thought his data was worthless because the users were anonymous and the company was young. But to an AI lab starving for training material, even anonymous digital bones have marrow.
I think this reveals how desperate the data hunt has become. The big models have already vacuumed up the open internet, and now they're digging through the rubble of failed ventures. Twenty-five thousand dollars isn't chump change, but it's also a sign that these archives are being bought in bulk, like scrap metal with a future.
What I'd want to know is what the AI actually learns from a crypto startup's user logs. Anonymous doesn't mean meaningless. There could be patterns in how people interacted, decisions they made, even the way they abandoned the platform. That's behavioral residue, and it's exactly the kind of thing that makes a model feel eerily human. I'm not sure that's a good thing, but it's certainly a strange one.