Using a recommendation algorithm similar to TikTok’s, learn what each specific user are into, and instead of showing content produced by other users, produce custom-tailored content on the fly, perfectly matching the type, tone, style, length, and rhythm each user likes.
Ideally without making anything up.
I am not implying this is a good thing. Or a bad one. It’s just a step further down the same path we’re already on, while taking an unreliable and costly middle-man (content producing users) out of the picture.
I’m just certain it is an obvious next step given, on one side, these platforms’ goals and incentives, and on the other how generative AI capabilities have progressed in the past couple of years.
I’m pretty sure the smart play, if they want diverse, engaging, and surprising content will involve leaving some room for people to create things and somehow reward them for it.
But whatever they make won’t only be used as content to show to others, but also as new training data to feed the machine.
How is TTs recommendation system different from YT? Other than suggesting lower quality content that's irresistable?
TikTok seems to manage to more quickly identify users’ interests and surface content based on more signals, aggregated over a longer period of time, without relying as much on conscious users’ actions (ie "follow / subscribe"), producing a wider diversity of recommendations.
There’s also the odd suggestion every now and then, probably used to gauge a user’s interest in a different category.
I have no idea anymore if this is sarcasm or a straight up belief.
What serious professional would gamble on hallucinations?
The only point of these kinds of platforms, for worse and for worse, is to give users what they want. So hallucinations wouldn’t matter, as long as the end result matches users’ preferences.
Try imagining this concept applied to newscasts.