Establishing a unified theory of cognition has been an important goal in psychology^1,2. A first step towards such a theory is to create a computational model that can predict human behaviour in a wide range of settings. Here we introduce Centaur, a computational model that can predict and simulate human behaviour in any experiment expressible in natural language. We derived Centaur by fine-tuning a state-of-the-art language model on a large-scale dataset called Psych-101. Psych-101 has an unprecedented scale, covering trial-by-trial data from more than 60,000 participants performing in excess of 10,000,000 choices in 160 experiments. Centaur not only captures the behaviour of held-out participants better than existing cognitive models, but it also generalizes to previously unseen cover stories, structural task modifications and entirely new domains. Furthermore, the model’s internal representations become more aligned with human neural activity after fine-tuning. Taken together, our results demonstrate that it is possible to discover computational models that capture human behaviour across a wide range of domains. We believe that such models provide tremendous potential for guiding the development of cognitive theories, and we present a case study to demonstrate this.

The kidney harbors one of the strongest circadian clocks in the body. Kidney failure has long been known to cause circadian sleep disturbances. Using an adenine-induced model of chronic kidney disease (CKD) in mice, we probe the possibility that such sleep disturbances originate from aberrant circadian rhythms in kidney. Under the CKD condition, mice developed unstable behavioral circadian rhythms. When observed in isolation in vitro, the pacing of the master clock, the suprachiasmatic nucleus (SCN), remained uncompromised, while the kidney clock became a less robust circadian oscillator with a longer period. We find this analogous to the silencing of a strong slave clock in the brain, the choroid plexus, which alters the pacing of the SCN. We propose that the kidney also contributes to overall circadian timekeeping at the whole-body level, through bottom-up feedback in the hierarchical structure of the mammalian circadian clocks.

Epistemology and technology have been working in synergy throughout history. This relationship has culminated in large language models (LLMs). LLMs are rapidly becoming integral parts of our daily lives through smartphones and personal computers, and we are coming to accept the functionality of LLMs as a given. As LLMs become more entrenched in societal functioning, questions have begun to emerge: Are LLMs capable of real understanding? What is knowledge in LLMs? Can knowledge exist independently of a conscious observer? While these questions cannot be answered definitively, we can argue that modern LLMs are more than mere symbol-manipulators and that LLMs in deep neural networks should be considered capable of a form of knowledge, though it may not qualify as justified true belief (JTB) in the traditional definition. This deep neural network design may have endowed LLMs with the capacity for internal representations, basic reasoning, and the performance of seemingly cognitive tasks, possible only through a compressive but generative form of representation that can be best termed as knowledge. In addition, the non-symbolic nature of LLMs renders them incompatible with the criticism posed by Searle’s “Chinese room” argument. These insights encourage us to revisit fundamental questions of epistemology in the age of LLMs, which we believe can advance the field.

Magnetic resonance spectroscopy (MRS) has seen an increase in popularity as a method for studying the human brain. This approach is dependent on voxel localization and spectral quality, knowledge of which are essential for judging the validity and robustness of any analysis. As such, visualization plays a central role in appropriately communicating MRS studies. The quality of data visualization has been shown to be poor in a number of biomedical fields and so we sought to appraise this in MRS papers. To do this, we conducted a survey of the psychiatric single-voxel MRS literature. This revealed a generally low standard, with a significant proportion of papers not providing the voxel location and spectral quality information required to judge their validity or replicate the experiment. Based on this, we then present a series of suggestions for a minimal standard for MRS data visualization. The primary point of these is that both voxel location and MRS spectra be presented from all participants. Participant group membership should be indicated where more than one is included in the experiment (e.g. patients and controls). A set of suggested figure layouts that fulfil these requirements are presented with sample code provided to produce these (github.com/nwd2918/MRS-voxel-plot).

Establishing a unified theory of cognition has been a major goal of psychology. While there have been previous attempts to instantiate such theories by building computational models, we currently do not have one model that captures the human mind in its entirety. Here we introduce Centaur, a computational model that can predict and simulate human behavior in any experiment expressible in natural language. We derived Centaur by finetuning a state-of-the-art language model on a novel, large-scale data set called Psych-101. Psych-101 reaches an unprecedented scale, covering trial-by-trial data from over 60,000 participants performing over 10,000,000 choices in 160 experiments. Centaur not only captures the behavior of held-out participants better than existing cognitive models, but also generalizes to new cover stories, structural task modifications, and entirely new domains. Furthermore, we find that the model’s internal representations become more aligned with human neural activity after finetuning. Taken together, Centaur is the first real candidate for a unified model of human cognition. We anticipate that it will have a disruptive impact on the cognitive sciences, challenging the existing paradigm for developing computational models.