yeah the tt is a funny thing. we basically built and trained llms to the test... here is a prompt here is what a human would respond, copy that. I'm sure in the classic actual test parameters you could get llms to pass it. But we've had pre llm systems designed to the classic test that could fool a human too, which nobody would argue is agi.

maybe it is moving the goalposts by saying the classic test parameters are too narrow and humans are not actually that good at administering that test. But in the spirit of the turing test i think its pretty clear we haven't crossed it just yet in a general way. It's still pretty obvious when we are interacting with the machines. even when we arent chatting with them. genai images video, documents they create, webpages, etc. Not that its impossible to be tricked here or there or be unsure but id say the core idea of the turning test is still valid and that current llms and their outputs aren't generally passing it.

like for example you probably have a good idea right now if im an llm or wrote this comment purely out of my squishy human brain. that you can still somewhat regularly do that means the turing concept could still be sufficient even if we need to adjust the parameters of the actual test.