Taleb does argue that IQ is predictive, but only in the left tail, decorrelating as it goes higher. The strongest evidence here (against his case) is the odds ratios in the univariate logistic regression from -1 SD to +1 SD, which does show some apparent predictive power after >100. Still, he mainly argues that an asymmetrical piecewise linear function fits this type of data better than a linear model, especially after accounting for noise.