Mastodon Feed
Boosted by glyph ("Glyph"):
mhoye@cosocial.ca wrote:
RE: https://cosocial.ca/@deborahh/117246659791027480
This is about offloading the costs and externalities of training language models onto the general public.
The people bearing the social and technical costs of this are just the people using the products; the beneficiaries are companies that can identify outputs of existing models so they're not reingested as training data.
We need to recognize any market legislation of this shape, that privatizes gains, socializes costs and ignores externalities, as public subsidy of a private corporation.