Learning Without Asking The Rights Holders! Please, STOP right now!
There are problems with that now. All existing generative models have not done this and simply infringe other people's copyright. Therefore, every new generation result is stealing!
When learning without stealing, it's a wonderful technology!
One thing AI should never do is lie.
As the latest benchmarks from OpenAI revealed, the accuracy of even the best models is still far from perfect. I believe it's crucial for AI to be transparent and honest, especially when users rely on it for important decisions. Misinformation or fabricated answers can break trust and lead to unintended consequences, so ensuring AI is as truthful as possible should always be a priority.
It shouldn't spew out blatantly false information while pretending it's factual. Also needs strong safeguards against being used for harassment, bullying, stalking, scamming, etc. But IMO the copyright issue is overblown - it's just learning patterns, not stealing specific content. Same as how humans learn.