Goalpost are moving because they've finally are admitting that Autocorrect++, although very useful as a pattern matcher / extender, is nothing at all like consciousness or agency-based "will".

In the end, LLMs are just a tool...much like a calculator can help a human do math better, or a bulldozer can help them do work, LLM's are just an extension tool that mainly helps humans manual eliminate boilerplate creation / and sift vast amounts of information.