Another example (perhaps a bit frivolous): when browsing Less Wrong comments and deciding which to upvote, it might be tempting to take the existing upvotes into account. However (to an extent-- it's just an analogy), this is like using the fact that my probability for some statement A is high as an argument to increase the probability of A. The direction of propagation is towards the estimated goodness of the post, so using that estimate in the argument is bad form.