I do wonder what would be an acceptable level of guarantee to trigger a “human written” bit.
I actually think a video of someone typing the content, along with the screen the content is appearing on, would be an acceptably high bar at this present moment. I don’t think it would be hard to fake, but I think it would very rarely be worth the cost of faking it.
I think this bar would be good for about 60 days, before someone trains a model that generates authentication videos for incredibly cheap and sells access to it.
Interesting...thinking this through: For text and ideas the information size is often small enough to fit in human memory, and thus containing this is already unsolvable! I can ask the LLM to compose the text of a pitch and then film myself writing it out. Nothing you can do will prove the provenance of those bits was not from the AI.
So I think the premium product becomes in-person interaction, where the buyer is present for the genesis of the content (e.g. in dialogue).
Image/video/music might have more scalable forms of organic "product". E.g. a high-trust chain of custody from recording device to screen.
I actually think a video of someone typing the content, along with the screen the content is appearing on, would be an acceptably high bar at this present moment. I don’t think it would be hard to fake, but I think it would very rarely be worth the cost of faking it.
I think this bar would be good for about 60 days, before someone trains a model that generates authentication videos for incredibly cheap and sells access to it.