Hacker News new | ask | show | jobs
by kinggencha 1258 days ago
This feels entirely pointless until you can prove that an AI produced a result because it looked at something you own.
3 comments

If you had the prompt and the output, and you knew your work was in the training data, you _could_ retrain the whole model minus your work, and then feed the same prompt back in.

Then if the output looked nothing like your work as it did for the inclusive model, you could prove your work was influential, possibly to the point of a © claim.

Probably quite an expensive way to do it, unless it's somehow possible to reverse specific inputs from a trained model :)

piece of cake. Load up copilot and start typing a comment that credits yourself for your work. It will autosuggest somebody else's real name and email address that you can search up and validate.
When a company decides to create an alternative to thing X they take care not to employ people who have worked on X. If AI creates a thing like X and has seen X I think there is probably a case.