Hi Gleb. It's been 9 years since you wrote this post. If you are still interested in strategies to promote rationality broadly, I have an idea which I'm starting work on. Basically it's about creating an layperson-friendly system for empirically evaluating predictive mathematical models of recurring phenomena. My thesis is that ordinary people are currently unable to separate knowledge from confidently-stated BS in most topics, and that the two are separable by shaping knowledge into predictive models and testing those predictive models against subsequently-measured data. I'm looking for partners. Would you like to have a conversation?
Thank you for pointing me that way. No I have not!
So I took a quick look. This looks a lot like HuggingFace. It's good to be reminded that these things exist and they do have some things in common with what I propose. As they stand, though, it's not it. Notice I'm talking about scientific models here. The mindset with which I approach this is one of theoretically-motivated, sparsely connected models, the kind you learn about when you take a university course in say, psychology or economics, not the kind you train with neural networks.