I'd say there's two related bad habits: 1: making exclusively unverifiable predictions, and 2: expecting or acting as if you were expecting to be more accurate in your unverifiable statements than in verifiable ones.
edit: those two often go together. There's a huge supply of cheap verification in the homework sections of the textbooks. Armchair physicists barely tap into this supply. edit: and by barely I mean pretty much none ever do.
Topics such as Amanda Knox trial or LHC finding the Higgs, it seems to me that for such predictions the dopamine pleasure of success is larger than the feeling of failure, and the betting on those is thus somewhat wirehead-y. May still be good for calibration though.
Today an Italian court has declared that Amanda Knox is, once again, guilty. She did not attend that trial (is not required to in Italy), so her final verdict will be either by appeal to the Supreme Court of Italy or the US extradition court. Extradition requests might be impeded due to the fact US does not have double jeopardy.
Previously on LessWrong, in The Amanda Knox Test: How an Hour on the Internet Beats a Year in the Courtroom there was some complaint that it actually took more than an hour on the internet to thoroughly research the case. Of course, the courts have been at this since 2007...
Her co-defendant, Raffaele Sollecito, who did show up at the trial, got sentenced to 25 years, but I don't know for sure where he is now because apparently he's totally unimportant and who cares (the media's opinion, not mine). I'm fairly sure he's in Italy though. So far it seems the plan is to revoke his passport but not arrest him.
Anyone want to take their hand at making predictions?