3 ms·
I'd first want to see generative AI trained on 100% non-copyrighted (or properly licensed for the purpose) works. I'm pretty sure that copyright laws will be ad
by ale42 3y ago
I'd first want to see generative AI trained on 100% non-copyrighted (or properly licensed for the purpose) works. I'm pretty sure that copyright laws will be adapted in the next years to cover AI/training usage, so people will need proper licensing for that. Not saying that copyright is good or bad or anything else, that's just my intuition about the next years...
- gumballindie 3y agoExcellent. “People” are also folks with mortgages and kids - they need their work protected instead of looted by corporations. Similarly, open source code needs licensing honoured. This way there will be an incentive to pay those who’s intellect is used for training machine learning. A win win situation.
- version_five 3y ago"People" aren't owed an unstainable business model by society any more than companies are. Taxing everybody else with copyright rules is not a clear net benefit for society.
- davemp 3y ago> "People" aren't owed an unstainable business model by society any more than companies are. True. But society needs to align incentives with its desired outputs. If original work and creativity just makes other folks (entrenched middlemen) rich, we’ll see stagnation. With capital likely determining even more than before. If a pursuit can’t make money, it becomes a luxury with all the implications involved.
- regularfry 3y agoI'm sure we'll see something trained only on out-of-copyright visuals soon enough. Licensing is a red herring.
- mschuster91 3y agoA team actually brute-force generated all 68 billions of possible melody combinations for 12 notes in the middle-C octave set back in 2020 [1], essentially trolling/breaking the copyright law as now all these melodies are in the public domain as prior art. [1] https://www.theatlantic.com/technology/archive/2020/02/whats-the-point-of-writing-every-possible-melody/607120/ https://www.theatlantic.com/technology/archive/2020/02/whats...