Apparently, “Theorems for free!” is a paper that talks about an extensive ability to reason about parts of programs, if you follow some rather basic rules.

However, lots of popular programming languages throw this ability out the window, because they do not want to enforce those basic rules.
Most languages, for example, allow for rather uncontrolled side effects and to be able to reason as a programmer, you have to make the assumption that no one else abused side effects.

The instanceof is rather referring to dynamic typing, though, as e.g. employed by Python and JS, which makes it difficult to make any assumptions at all.

So, in statically typed languages, when you’re implementing a function, you can declare that a given parameter is a number or a string etc. and the compiler will enforce that for you. In dynamically typed languages, you have to assume that anyone calling your function is using it correctly, which is a difficult assumption to make after a refactoring in a larger codebase.

All in all, such different levels of rigorosity can be fine, but the larger your codebase grows, the more you do want such rules to be enforced, so you can just ignore the rest of the codebase.

It’s making fun of dynamic languages because rather than letting the compiler prove theorems about statically typed code, they… don’t.