Fig. 4: Human alignment.

a, Multidimensional scaling embedding of the ten behavioural metrics in CogBench³³ for different models. b, Pearson correlation coefficients indicating how well human neural activity in the two-step task³⁷ can be decoded using Centaur’s internal representations extracted from different layers. c, Pearson correlation coefficients indicating how well human neural activity in a sentence-reading task³⁸ can be decoded using Centaur’s internal representations extracted from different layers. Control refers to a model that used representations extracted from a randomly initialized transformer model with matched architecture.