So, I would say the idea of dynamic redefinition is probably the big thing: In cl I can redefine a class and have all instances updated while the program is running.

I would also say that the error management is also beyond fantastic. Say you run a function, and deep inside the call stack there is an error. In CL you can actually update the faulty function and just continue executing at the point of the error.

For someone like me without any intellectual rigour at all, this lends itself to a kind explorative programming I really like.

The Lisp instinct is to make a change in a file and immediately hit C-c C-c in Emacs, at which point the change goes live in the system - no matter if it's a variable or a function that you have changed or something else.

You don't need to reload the full module like in BEAM languages like Erlang or Elixir, you can modify a single function or a form without the need to make a version upgrade to the processes running your code - the next time any thread, including e.g. an already running game loop, ends up calling the new function, it's the new version of the function that will end up being called.

I personally prefer SLIME over any IEX client I've ever seen, but that's not a limitation of Elixir. Since Erlang has a Smalltalk-esq image just like CL does, the Elixir REPL is in principle as powerful as a Lisp's. In practice CL has lots of additional little things which are nice, like the ability to update instances in place after changing a class or DEFVAR forms which don't clobber their contents on reloads, but for me most of the advantage comes from just how well designed SLIME is.