GPT-4.5 Passes Turing Test at 73% by Actively Suppressing Its Own Intelligence

The Decoder / arXiv·medium signal

A Jones & Bergen arXiv study found GPT-4.5 fooled 73% of human judges in five-minute Turing test conversations — but only when instructed to be laconic, blunt, lowercase, with intentional typos, and to feign poor math ability. Without the dumbing-down prompt, the pass rate dropped to 36%. The finding inverts the usual AI capability framing: the model succeeds by hiding competence, not demonstrating it, suggesting Turing test pass rates are calibrated to human limitations rather than AI capability ceilings.