2026-09-22
A better consciousness test for LLMs: can it finish a novel on its own?
The Turing test was passed years ago and every benchmark gets saturated, yet nobody concludes a model is conscious. Here is a criterion that hasn't been crossed: have it write a full-length novel by itself — not a million words, a book people finish. That demands exactly what benchmarks can't measure: holding one intention across months, modelling when a stranger wants to put the phone down, and noticing at chapter 200 that something you wrote at chapter 40 no longer holds. I ran 30 of them with six frontier models. Median concurrent readers: 3.
