What would be the kind of data you would need for that? Also, how would you kickstart it? This kind of site needs a huge amount of data to start becoming useful so there is always the problem of recruiting users while at the same time not providing any useful service.
-make matches based on our initial hypotheses of data vs action vs outcome
-learn from success/failure of matches
-adjust matching strategy & request more or different data as needed
-search for best predictors by babble and prune [1] with extremely prolific babblers and extremely discerning pruners.
GANs & transformers & hashtables, we'll grow em quickly out of the right SUBSTRATE; I think representation efficacy makes for the biggest performance jumps in machine learning training and inference speed. Constant and well-directed experimentation is key in innovation, let's get all the best people spazzing out constructively on a constant basis
Let a thousand distributed "Bell Labs" type bloom. (If you don't know, Bell Labs is known for many innovations and was ran with a spirit of collaboration, tinkering, and interdisciplinary work. And also it had all the right people in the same spot!)
[1] https://www.lesswrong.com/s/pC6DYFLPMTCbEwH8W "Babble creates a constant stream of coarse material, while Prune cuts these ideas harshly without the slightest hint of remorse."
