> The same goes with human TAs that are extensively used in undergrad introductory programming classes. They can also be unreliable in many cases

I think one difference is that human TAs can, theoretically, be held accountable for their reliability whereas a holding a LLM accountable is a little more difficult.