After the first attempt at a Show HN accidentally got eaten by HN's spam detection and only rescued too late, here's another shot (ok'ed by @dang - thanks again

After the first attempt at a Show HN accidentally got eaten by HN's spam detection and only rescued too late, here's another shot (ok'ed by @dang - thanks again for helping and your advice). This project is something I've been meaning to work on for quite a while, mostly because I think it needs to exist and also because I think I have the background to do it. Most of the work has happened over the weekends since the start of startup school (this is one of the "accepted by accident" projects where they sent out an acceptance email to everyone they were planning to reject, then another rejection and finally a "sorry for the messup - we're letting you join" - so let's see how things go ...) The idea is the following: there has been this thing where companies basically kill off everything that doesn't work too well with their mainstream audience. This has been happening on a large scale (e.g. with Google killing Reader and now Inbox), on a smaller scale with companies removing features and on the micro scale it has even been automated with a/b testing (where you basically try to explicitly measure which option works best for the average user and kill the others). This project is basically trying to attack the problem from this micro scale while hopefully later moving up that chain or at least motivating change there as well. The way it works is that we change the original goal of a/b testing from "what's the best option on average overall?" to "what's the best option for this specific user?" and try to automatically show him that (using reinforcement learning internally to make it happen). Right now this works on websites where we use JavaScript and some CSS magic to make it work in a way that's easy to use for developers - but longer term we're planning to look into other places as well. I think the tricky part right now is to not move in this adtech direction it could easily be, but rather towards what chatbots tried and failed at - at least so far. There's gonna be another blogpost about this topic, but the tl;dr essentially is that the interesting part about chatbots is not the natural language interaction, but that they are - by lack of ui - basically forced to focus on thinking about "how do I find out what the user really wants and get him there as quickly as possible without shoving irrelevant stuff in his face?" - and this way of thinking could also be really useful even if you have ui - and ideally automated using AddAI.