I recall checking out the blog a while back, upon Lukeprog's recommendation via his blog, and leaving with a much lower opinion of the author after reading his post on the representativeness heuristic (causing me to classify him as pretty close to my model of Massimo Piatelli-Palmarini). If you check out the comment section, it looks like he thinks that your probability estimate in cases like the lawyer/engineer question should always track the frequency information you are given, because using your subjective stereotype information would be to "ignore statistics." Although I haven't bothered reading his stuff since, I expect that a careful look at his articles will reveal further such misunderstandings..