> It is important to emphasize thatthe construction of LC is entirely dependent on the pre-existingsymbolic processors developed over the last 50 years by experts in symbolic mathematics.Moreover,as things now stand, extending LC to fill in some of its gaps (e.g. the simplification problemsdescribed in section 3) would make it even less of a stand-alone system and more dependent onconventional symbolic processors. There is no reason whatever to suppose that NN-based systemswill supercede symbolic mathematics systems any time in the foreseeable future.

That's the gem of the review.

And this paragraph

> It goes without saying that LC has no understanding of the significance of an integral or a derivativeor even a function or a number. In fact, occasionally, it outputs a solution that is not even a well-formed expression. LC is like the worst possible student in a calculus class: it doesn’t understandthe concepts, it doesn’t learned the rules, it has no idea what is the significance of what it is doing,but it has looked at 80 million examples and gotten a feeling of what integrands and their integralslook like.