How do you feel about the integrity of the machine learning researchers over the past twenty years who trained models on scraped internet data that weren't particularly powerful and didn't attract any attention?
I think it's notable that nobody was calling out those researchers for their lack of integrity, because the systems they were building did not seem like a threat to anyone.
OpenAI etc get accused of a lack of integrity on this precisely because the systems they are building work, and are profitable.
My personal opinion here is that integrity is more about what you build with the data. I think saying "scraping means you lack integrity" is a simplification.
OpenAI et al also stole everything from everyone. But then they raised billions of dollars from that data and sell back their LLM to people (again, among other things). They are also very much NOT open in any way, aside from sharing their benchmarks of new models.
Also my understanding was they’re not storing the actual music, but the metadata and a link to the YouTube video.
Then you have OpenAI etc.. who build these multi-billion (trillion??) dollar machines and sell them back to people, using everyone's proprietary data, and (among other things) tell everyone it's going to take their jobs. That combination of things doesn't scream integrity to me.
Still, it's undeniable that these machines could be beneficial for humanity (cancer research and such). So, I'm sure many people would say the good out-ways the bad. I don't know. Seems that would set a risky precedent for future companies, but maybe not.
legally speaking, the default privacy notice gives them an irrevocable license to your content. they may read and use the prompts for research. so it is very possible they simply stole the navier-stokes solution.
that is the same principle as any other prompt but this would be a concrete example.
there would be some difference between simply giving the model some prompts to read, which they are entitled to do on the default policy, and putting it into aggregate training data.
The strongest complaint is that they trained on a huge corpus of pirated copyrighted works.
It’s a large step above “scraping” and well into the “everyone acknowledges this is illegal” territory.