Great point! I guess the point of ITT is that even when you disagree with your opponents, you have the ability to see their (wrong) model of the world exactly as they have it, as opposed to a strawman. For example, if your opponent believes that 2+2=5, you pass ITT by saying "2+2=5", but you fail it by saying "2+2=7". From your perspective, both results are "equally wrong", but from their perspective, the former is correct, while the latter is plainly wrong. In other words, the goal of ITT isn't to develop a "different, but equally correct" map of the territory (because if you would believe in correctness of the opponent's map, it would also become your map), but to develop a correct map of your opponent's map (as opposed to an incorrect map of your opponent's map). So, on some level, while you pass an ITT, you know you are saying something false or misleading; even if just by taking correct arguments and assigning incorrect weights to them. But the goal isn't to derive a correct "alternative truth"; it is to have a good model of your opponent's mind.

Huh? You are proposing a very stark, black-and-white, all-or-nothing position. Recall that for a rationalist a belief has a probability associated with it. It doesn't have to be anywhere near 1. Moreover, a rationalist can "believe" (say, with probability > 90%) something against which good arguments exist. It just so happens that the arguments pro are better and more numerous than the arguments con. That does not mean that the arguments con are not good or do not exist. And, of course, you should not think yourself omniscient. One of the benefits of steelmanning is that it acquaints you with the counterarguments. Would you know what they are if you didn't look?