Mandlebrot makes serious criticisms of the standard models of stock and other valuations - these boil down to saying that the "tails" of market changes are going to be larger than the current models.
The simplest example is CDO's. If you believe in short-tailed, uncorrelated stock market changes, you can argue that the stocks of five different tripple-A rated major corporations have a zero percent chance of simultaneously declining. And you similarly combine together even poor quality bonds to create a "synthetic" bond which also supposedly has a nearly 0% chance of failure.
If you believe in long-tailed, correlated stockmarket changes, you believe that the chances of such bonds failing is much higher.
Guess what actually happened?
I would criticize Taleb, however, for not bringing the issue of complex random processes to the fore. I think he wants to make his ideas very accessible but the problem is that he looses the key difference between short-tailed and long-tailed distributions, since short-tailed distributions DO exist in reality, especially physics and so we're not talking generic randomness when looking the problems of understanding markets and uncertainty.