The commercial drones from DJI are the best in the industry. But also making something that’s almost as performant but for a fraction of the cost requires real innovation as well.
DeepSeek’s training model was innovative. They used multiple large specialized models to train a very small general model. This is a real practical innovation over OpenAI’s one behemoth general purpose model.
Wrong, the whole purpose of tracking your usage is to identify what kind of consumer you are so they can sell your views to advertisers. Recommendations are based on what category of consumer you’ve been identified as. Maintaining your viewership is secondary to the process of selling your views.
There’s also an entire industry around mass producing this content and deliberately gaming the algorithm.
It only must be extremely simplified and evoke emotional reactions. That’s just basic propaganda rules. The brainrot quality of the content is a consequence of the sheer quantity of the content. You can’t make that volume of content and without making fully automated ai slop.
What the experiment overlooks is that there are PR companies being paid to flood YouTube with rightwing content and are actively trying to game its algorithm. There simply isn’t a left-wing with the capital to manufacture that much content. No soros-bucks for ai minions in keffiyehs talking about medicare.
There were several “social credit systems” that were set up in various provinces. These were mainly attempts to aggregate public records from different departments into one searchable database. It mostly affected business owners since they have the largest public record footprint. The idea was you could look up who owns what business and how many code violations those businesses had.
This triculates into the american rumor mill and comes out as the “social credit score” meme which is fantasitcal projection from our own opaque credit system.
You’d just get aimless riots if you tried it that way.