I doubt it, that's clearly exceeded by these language models. Calling it just an autocomplete - because it can mean a lot of things people are familiar with - is a way to downplay their significance.
Not true. Take a look at the paper and benchmarks. The point of the thing is that it does well on a number of NLP tasks by being an expensive autocomplete. As people demonstrate in this thread, it still has significant flaws.