Ask HN: Do you believe any LLM's pass the Turing test? How? Can’t tell if the words come from a human or not? Just trying to understand how. |
Ask HN: Do you believe any LLM's pass the Turing test? How? Can’t tell if the words come from a human or not? Just trying to understand how. |
Do you mean "they have a style that you can learn, surely you can recognise this style!", yes, but the default style is ~ a mix of Kenyan (because they were cheap RHLF trainers) with some of how word processors will auto-replace -- with — (plus people like me who were doing em-dashes manually for ages), and also it's trivial to just ask them to use a different style.
Biggest tells are that they respond with superhuman speed, and that they're moderately competent across an inhumanly broad range of skills while being masters of almost nothing.
That, and they keep openly saying "I'm a bot" in various ways e.g. "As a large language model…", that they have generally been expressly trained to not pretend to be a person, whereas the actual Turing Test is an AI that is trying to pretend to be a person — but again, these are things you can get around very easily by just asking the model to respond in certain ways.
It's just… the "tells" are just easy to remove, is all.
The default configuration is easy to spot.
It's easy to change that configuration and make them hard to spot.
The default style, that is easy to spot, is also a style that actual real humans use, leading to false positives. I myself have been accused of being a LLM here on this site, just from my style.
The former, how would I tell? I'd have to find someone whose writing style I don't know to be the human comparison, in order to blind myself.
The latter: https://arxiv.org/abs/2503.23674