You'd have to find a way to determine how much of each artist contributed to every output from every prompt, find every artist that had a work used in the training data, and get them all to agree on how much they should be paid. Then come up with a pricing model to sell the art created that people will pay for.
All that to get to the Spotify problem where the money coming in is distributed to so many artists, that nobody makes any money, except the company selling art makes a ton because they take a percentage off every picture.
In the end, the artists feel just as screwed over as they did before all of this.
This weekend project turned out to be surprisingly non-trivial. First, you have all sorts of silly limitations in Twitter's API. You can only retrieve a maximum of 3200 tweets from a user's timeline, which for very active accounts only goes back a year or so. Moreover, Twitter forces you to get a minimum 5 tweets at each API calls, even if you need just one.
But the most surprising to me was that there is no coherent indexing system in the tweets IDs. The best solution I found was to generate a random date and get the latest tweet posted before that date and juggle that with the 3200 tweets limitation.
How would you guys go about retrieving one random tweet from a user's timeline?
My problem with Google is that sometimes the voice becomes "metallic" on some portions of the text. Not having that would be nice. Also perhaps some randomness. When saying the same few words twice, there could be a little variety. More generally, I am also wondering who has the best tech currently (assuming the answer is settled enough).
