3 ms·
Our copyright model isn't sufficient yet. Is putting a work through a training/model sufficient to clear the transformative use bar? That doesn't make you saf
by mey 2y ago
Our copyright model isn't sufficient yet. Is putting a work through a training/model sufficient to clear the transformative use bar? That doesn't make you safe from Trademarks. If the model can produce outputs on the other side that aren't sufficiently transformative then that single instance is a copyright violation.
Honestly, instead of trying to cleanup the output, it's much safer to create a licensed input corpus. People haven't because it's expensive and time consuming. Every time I engage with an AI vendor, my first question is do you indemnify from copyright violations of your output. I was shocked that Google Gemini/Bard only added that this year.
- ffsm8 2y agoI'm honestly surprised AI-washing hasn't become way more widespread then it is at this point. I mean recording a good song is hard. Generating a good song almost impossible. But my gut feeling would've been that recreating a popular song for plausible deniability would be a lot easier. Same with republishing bestselling books and related media. (I.e. take Lord of the rings and feed it paragraph for paragraph into an LLM that you've prompted to rephrase each to a currently bestselling author.)
- jncfhnb 2y agoNothing will ever protect you from trademark violations because trademarks can be violated completely by accident without knowledge of the real work. Copying is not the issue.