News for hackers. Stuff that matters.
Re:Ask HN: Any way to ban GPT-4 and others from harvesting m... by spywaregorilla on Saturday August 13, 2022 @09:45PM (#32455573)
I wonder if it would be possible to test if your material had been used by these models by producing large works that create strong associations with novel words. If you write a page on how Keanu Reeves is the embodiment of pnkorka; and then language / visual models create that association because there is literally nothing else, that would be fairly strong evidence that you're incorporated. There may even be an opportunity to inject new sentiments into the model.
Re:Ask HN: Any way to ban GPT-4 and others from harvesting m... by hammyhavoc on Sunday August 14, 2022 @06:29AM (#32458112)
I don't know what/who `pnkorka` is, but I love the word.
Re:Ask HN: Any way to ban GPT-4 and others from harvesting m... by spywaregorilla on Sunday August 14, 2022 @09:14AM (#32458981)
That's the point. It's a word that won't appear anywhere else in the training data but your own articles. I'm not sure how these models work for such edge cases. But I'm curious if you could get a model to learn some associations on your articles alone. You could define a new part of language.