It's kind of like the history of the web with regular human consumers and various attempts to allow microtransactions, which receded in favor of ads and tracking.
If the pattern holds, the "cost" of training non-paywalled data will be attempts to hack/influence the model, as opposed to hacking humans.
If the pattern holds, the "cost" of training non-paywalled data will be attempts to hack/influence the model, as opposed to hacking humans.