Economist on Guns

Not surprisingly Freakonomics has the best and most reasonable discussion on gun laws that I have seen. You can listen to it here.
But I have one criticism. About halfway through the podcast, Levitt suggests a better way to lower gun crime is to attach severe punishments to crimes involving guns. He says that mandatory sentences will work and have proven effective when used:

"LEVITT: Well I think the policies that can work are ones that tie heavy punishments to uses of guns that we don’t like. So for instance, laws that say if you commit a crime and you have a gun with you, regardless of whether the gun was used, then without any sort of other consideration, we add five years, or 10 years, or 20 years, or 50 years to the sentence that you get. Those kinds of laws, I guarantee you, will work. If the incentives are strong and tell you don’t use guns, then I guarantee you we will see the number of gun homicides fall and the number of knife homicides rise, but not one for one. People will substitute away for using guns to kill people to using knives to kill people. But it won’t be one for one because knives just aren’t as good of a tool for killing people as guns are. That will work. I have no doubt that will work. It’s worked in California in the past when California put mandatory sentence enhancements on for felonies that were committed with guns. But I think the policy has to be one of that nature, where you’re not tying it to the gun itself, you’re tying it to the use of guns that you don’t want"

This argument ignores the often discussed Freakonomics topic of decision making. We know that people are terrible at making good decisions about the future, even when they know that something is a good idea. We know they don't put money away for savings and they don't stop committing crimes  even when they already have two strikes. 

How effective will mandatory sentences be if we already know criminals don't make good decisions?