A study published in NPP—Digital Psychiatry and Neuroscience on June 10, 2026, draws parallels between human psychiatric symptoms and output errors in large language models (LLMs). Authors J. D. de Boer, S. Ciampelli, and Araya K. Hailemariam examine how LLMs and automatic speech recognition tools frequently exhibit phenomena akin to human hallucinations and confabulations. In both humans and these models, missing information often leads to the generation of coherent but factually incorrect responses.

The research suggests that these shared errors point toward common computational principles in predictive systems. By treating AI as a computational mirror, the team argues that psychiatric knowledge could assist in reducing model error rates. Simultaneously, identifying these patterns offers new insights into how human brains construct perception and memory.

A study published in NPP—Digital Psychiatry and Neuroscience on June 10, 2026, draws parallels between human psychiatric symptoms and output errors in large language models (LLMs). Authors J. D. de Boer, S. Ciampelli, and Araya K. Hailemariam examine how LLMs and automatic speech recognition tools frequently exhibit phenomena akin to human hallucinations and confabulations. In both humans and these models, missing information often leads to the generation of coherent but factually incorrect responses.

The research suggests that these shared errors point toward common computational principles in predictive systems. By treating AI as a computational mirror, the team argues that psychiatric knowledge could assist in reducing model error rates. Simultaneously, identifying these patterns offers new insights into how human brains construct perception and memory.