Reply to some self-important dork whining about this talk at Dienekes':
"You simply cannot criticize a new, rapidly-evolving and improving model just based on its trivial, known shortcomings. Such a thing is ludicrous and paints a truly bad picture of the talk presenters."
I'm afraid your effeminate idea of proper protocol has no bearing on actual science. Gray and Atkinson's "innovation" is insisting that Bayesian phylogenetics with limited and sometimes questionable inputs of data can produce highly accurate and precise readouts of linguistic history that supercede all previous linguistic and archaeological knowledge. Their results may dazzle twits like you and appeal to those who find their results politically or ethnically congenial. But the first question a serious person would ask is how closely Gray and Atkinson's attempts at reconstruction recapitulate recent/known linguistic history. That they frequently fail to do so is extremely germane to the question of how much faith one should put in their deeper reconstructions.
Statistical models are not magic. Bayesian tree building is not magic. Even with large corpuses of genetic data, the "most likely" tree is often overwhelmingly likely to be wrong. For genetics, where there's an explosion of data with comparatively few human analysts and little or no historical context, such results are useful, being often the best we have until additional data and further refinements of models appear. On the other hand, in linguistics, where on the PIE question relatively many human analysts have been poring over a comparatively limited corpus for many decades, it's up to Gray and Atkinson to demonstrate they have something useful to contribute. Every indication says they do not.