I don't get the allegations of goalpost shifting around Turing tests?
Turing discussed learning from experience extensively in his classic thinking machines paper.
Our existing LLM-based AI is capable of behaving as if it learned something from experience within these limited windows of time (say, an hour).
Turing's original paper has an adversarial test where you have a judge and two participants. When people talk about the Turing test casually, they often imagine a model where there's only one participant and the judge. The latter is easier for the machine.