A hidden map of meaning inside language models is structurally identical to how the human brain organizes concepts.
Large language models develop internal geometry to represent features like size, speed, or morality. These geometric relationships mirror human psychological association scales with surprising precision. The model has essentially evolved a human-like worldview without ever being told how humans think. This suggests that the structure of meaning is universal rather than specific to biological brains. It means that as AI gets smarter, it will likely share our biases and conceptual shortcuts by default. Understanding this map allows us to predict how AI will relate different ideas before it even speaks.
Semantic Structure of Feature Space in Large Language Models
arXiv · 2604.27169
We show that the geometric relations between semantic features in large language models' hidden states closely mirror human psychological associations. We construct feature vectors corresponding to 360 words and project them on 32 semantic axes (e.g. beautiful-ugly, soft-hard), and find that these projections correlate highly with human ratings of those words on the respective semantic scales. Second, we find that the cosine similarities between the semantic axes themselves are highly predictive