Re:AI is going to eat itself: Experiment shows people traini...
by visarga on Friday June 16, 2023 @12:21PM
(#36359246)
Take that one step further - what about learning on generated data from a model that used copyrighted data? It is exactly what MS "Orca" does. Does this step launder copyright?
[ Parent
| on HN ]
Re:AI is going to eat itself: Experiment shows people traini...
by visarga on Friday June 16, 2023 @01:24PM
(#36360418)
If you tried a language model pre 2017, you'd see it was not even able to produce two lines of coherent text. Now we are past coherence to the point we complain it bullshits in an authoritative style. That's of course because we use writing style as a shortcut for writing quality, not the best idea.