2 ms·
Yeah that doesn’t say how the video was being generated. I meant how, not who. It could have been ML-based.
by adastra22 3y ago
Yeah that doesn’t say how the video was being generated. I meant how, not who. It could have been ML-based.
- sdenton4 3y agoI'm not clear how you think the ML of 2017 would have been useful for generating these videos? This points to specific actors and creators, who I expect were using whatever was cheap and easy to produce a pile of content with in 2017.
- adastra22 3y agoPre-2017 tech would have been up to the task. You'd create a bunch of generic animations and characters (all rigged up to the same skeletal system so all the animations work on all the characters), and design composable elements for a scene. Then use ML categorization techniques to label both popular YouTube videos and your pieces. From there it would be mostly automated: the ML agent strings together scenes to create a bunch of videos, then use reinforcement learning to adjust and mutate content based on some metric of view counts.