It does not look very worrying. When he described the idea I wondered how you would actually do this given the amount of domain knowledge that is actually neces

It does not look very worrying. When he described the idea I wondered how you would actually do this given the amount of domain knowledge that is actually necessary to implement an underspecified program specification. According to the interview Szegredy's current research is going after theorem proving in mathematics because there is very limited domain knowledge.