Comment by frotaur
1 day ago
I think if you trained an LLM specifically for the purpose of passing the Turing test (instead of being helpful, harmful, and so on), its likely it would pass it.
You would have to train it/finetune it on a couple hundred of 'humans chatting in the context of a Turing test'.
I don't think that's true, assuming that the humans are also allowed to strategize and study the problem beforehand. The main weakness of the current models is that they still fail hard at certain kinds of common sense reasoning about real world situations, that no human ever would. Things like the "walk or drive to a car wash that's 2 minutes away" thing from a few months ago (I think the latest models have patched that one in particular, but I'm sure others can be found in a similar vein).
That would not really pass the test in spirit, I think. Passing must be something that falls out as a consequence of being a good chatbot, not the other way around.
However such good conversational agents would still have benefits, and i would be curious how far the approach of building a model that’s nice to talk to can take you, as opposed to the current generation of all-knowledgable helpful assistants.
I think you are absolutely right.
The big problem with the Turing test always was that it doesn't take into account adversarial designs. They had those chatbot contests about a decade ago, where the chatbots would regularly pass the turing test, not because the bots were intelligent, but because they were packed with rhetorical tricks designed to avoid saying anything of substance.
By any meaningful measure, today's LLM's pass the Turing test. Except we may need to lobotomize them for the deception to work.
I'd like to see an honest attempt at that.
What's more interesting are the types of humans that might fail the Turing test. Maybe to discourage the failure of real humans, there could be consequences outside the context of the test.
That is, most people accused of being a computer would crack and start pleading their humanity, but some sociopaths might not. I then wonder how many of them overlap with those so invested in abusing the premise of the Turing test.
It really is a fun thought experiment when you spice it up enough.