Are bad personas better than no personas?

The latest UIEtips article has an interview with Steve Mulder about personas. Here's an interesting exchange:

UIE: When design teams don't have the time or budget to perform user research, do you recommend teams create personas based on who they believe the users of their web site or design really are?

Steve: I think personas not based on actual user research are absolutely better than no personas at all. A lot of customer and user knowledge already exists in many organizations, and by looking at the sales, marketing, product, customer support, and tech support perspectives, you can bring all these existing bits of knowledge together into personas without talking to any actual end user. If you design teams can't easily talk with ends users, this is the most effective way to try out personas for the first time.

I understand what Mulder is saying here, but I'm not convinced. The issue is that lots of people think they know about users, but they are often wrong (or myopic based on their own experiences). They rely on anecdotal evidence, faulty assumptions, and personal bias to make design decisions. If you take that and build a persona from it, all you are doing is formalizing the mistake. And that makes it even harder to correct in the future. And in my experience, even if you think everyone understands that these quasi-personas are a first step towards real personas, at some point someone will question the future investment in user research because "We already have personas!"

It's rarely the case that doing something badly is better than not doing it at all.