Chomsky's hierarchy makes sense as a theoretical basis. His theory is poor at describing how people actually learn languages and lazily conjectures preprogrammed language in the human brain to explain phenomena that other mechanisms could more plausibly explain.

However, showing that naive computer models can simulate language is a nice finding, as then, when the "more correct" theory comes along, we can show that the latter can simulate language just as well as Chomsky's theory, while also providing additional psychological explanatory power: implying the new theory is strictly better than the last. Chomsky's hierarchy thus provides a nice benchmark. This is how a lot of contemporary empirical research is done (especially today with neural networks and machine learning), and saying that these benchmarks or previous, imperfect research "is a lie" seems to misunderstand how research fields often develop. Moreover, the video's thesis ignores the value of explaining to students how a field has developed and grown, including by teaching previously imperfect/defunct ideas. Granted, maybe his professors at UPenn did not express clearly how the Chomsky hierarchy is imperfect/defunct, but then the solution is to just have them emphasize that more, not to throw the baby out with the bathwater.