In this talk he has repeatedly cherry-picked the most negative examples from every type system, while ignoring the more positive, simple, and elegant type system features. E.g., he talks about languages which represent product types as `int * int * string * string * ...`, i.e. as nameless clumps of data, while ignoring that those same languages usually support much richer record types which are of course product types with field labels.
Then he talks about Java types being non-composable while ignoring generics and row polymorphism in other languages. He talks about Clojure maps being powerful while ignoring the same maps with the same power in static languages. He talks about type system complexity while ignoring the mental burden of keeping the types in your head in a totally dynamic system like Clojure. Ironically, he even talks about how you can handle five to seven arguments mentally before getting lost, while ignoring that in a dynamically-typed system you're left to your own devices handling much higher cognitive loads.
Another beef I have here is he repeatedly talks about pattern-matching in 500 different places, and having to go and update them, while ignoring the fact that _no one does that,_ we actually do write modular code with data structures as abstract as possible and try to provide powerful, general-purpose functions which manipulate them instead of exposing all the variant cases for raw pattern-matching. PM is great, but it's still important to limit its use for scalability reasons.
I don't know to what extent he has explored the type systems of ML-family languages. But he certainly does not present them accurately here. I would love to see him just do a debate-style talk with someone really well-versed in type theory as well as practical systems, say, Yaron Minsky of Jane Street.