[devchix]

Indeed. I recently learned of his work on HN, and have been carrying his Magic Ink paper around for 2 weeks, nibbling on it bit by bit. This was written in 2006! I see implementations of the ideas expressed everywhere - for example Netflix's "Recommended" list. Although, a nit. I once bought a bunch of fairy tale books on Amazon. For months after that I was recommended nothing but more fairy tales. It's not counter-intuitive to think that if I bought 10 things of the same kind, I may not want to buy any more of that thing for a while. I think the paper called this last-value predictor, and clearly for a subset of decisions, this predictor is maladaptive.