Well, let's see... first of all, the brain is not digital and all of our computers are (quantum notwithstanding) - so I reckon they will be fundamentally incapable of the kind of power efficiency they're talking about. Biology manages to do a lot with relatively little energy, computers - not so much. Unless they're going to run this on literal brains in jars I think they're going to find themselves butting heads against the reality of the power requirements of digital processing, just like every other AI company.
Second, from what I've read, humans have developed specialised language circuits. These brain regions are dedicated to language processing and seem to underpin our entire linguistic system. Even if we deduced how the language centre of the brain works, mimicking it by raw simulation is simply impossible with our current computer technology - we can barely model the neural network of an ant.
Finally, their entire proposal appears to rest on the existence of something like an innate Chomsky-esque universal grammar, or another computational-linguistic model. We have absolutely no idea if such an "algorithm" even exists, so it'll be interesting to say the least to watch them chase their tails as the money dries up.
