We don't need to explain why/whether predicting "endogenous" activations is good, if we accept hypothesis that that's how brain is wired - it runs prediction learning by default. It makes sense, because the affected cluster of neurons doesn't know if this activation is exo or endo.
Prediction learning for endo activations is conceptually the learning of shortcuts: if screw model activation predictably leads through a chain of intermediate steps to "being worried" model, then a good predictor would learn to activate the latter model right away after seeing screw.

But it was always based on a stupid idea: make AGI open source.

That's not what follows from the email archives. 

Contrary, they believe that being open-source is a great strategy for the early stage of OpenAI which ought to be changed later. Take a look at the email from Jan 2, 2016 9:06 AM: 

[when we are close to AGI] it's totally OK to not share the science (even though sharing everything is definitely the right strategy in the short and possibly medium term for recruitment purposes).