![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
The point that I don't see getting made about AI enough is the requirement that these systems have on human labor - massive exploitation is needed to categorize and label the data which they scrape.
It's not a fully automated system, not yet. Which means that trying to poison the data set (for writing in particular) might not help as much as we'd hope - and also that the acquisition of tumblr as a data-set won't affect the output of writing AIs as we might expect.
It's not going an AI that uses the word blorbo every few words. It's going to be able to tonally replicate new types of internet dialects, and gauge human lexical patterns in ways which wouldn't have been possible before.
It's not a fully automated system, not yet. Which means that trying to poison the data set (for writing in particular) might not help as much as we'd hope - and also that the acquisition of tumblr as a data-set won't affect the output of writing AIs as we might expect.
It's not going an AI that uses the word blorbo every few words. It's going to be able to tonally replicate new types of internet dialects, and gauge human lexical patterns in ways which wouldn't have been possible before.
