In this Substack essay, I explore the flawed nature of human comprehension, arguing that our understanding is often just a calculation of probability rather than an objective grasp of truth. By examining linguistic errors like misheard lyrics and his own struggles with French phonemes, Willis demonstrates that humans frequently hallucinate or misinterpret data based on what seems most plausible. He compares these mental lapses to the ‘hallucinations’ of Large Language Models, suggesting that the mistakes made by AI are fundamentally similar to our own. The author contends that we apply a double standard by granting humans the status of ‘understanding’ while dismissing machines for the exact same predictive behaviours. Ultimately, the text challenges the metaphysical assumptions of human exceptionalism, asserting that what we call ‘understanding’ is merely an external attribution rather than a unique internal state.
Essentially, I am not advocating the view that LLMs have consciousness and understanding; rather, I am arguing that the human versions of these are inflated with metaphysics.
Philosophics 