I like this but find the dismissal of multiple intelligences pretty unconvincing. I'm amenable to the idea that IQ (and associated tests) measures some underlying `well-defined` problem solving ability (with each problem having other factors), but the 'poorly-defined' problems sure seem like multiple intelligences by another name.

In my experience, different poorly-defined problems have different shapes and how much skill a person can make progress on them depends on a lot of things. For example, it seems like problem of finding ways to make poorly-defined problems fit into a well-defined space ("engineering") and the poorly-defined problems of organizing groups of people into coalitions ("politics") draw on different sorts of intelligence. The author claims that this is often not tested (which is probably true) but the basic observation that unstructured problems are related to each other and different people solve different areas with consistent skill seems true to me.

Given that the purpose of a measure is to make predictions: IQ can be used to improve predictions of academic outcomes, whereas the multiple intelligences models are worse at this.

Whether or not there are multiple intelligences or not is an implementation detail unless knowledge of this has measurably lead to some better prediction.