Search papers, labs, and topics across Lattice.
This study reveals that large language models (LLMs) construct geometrically faithful representations of concepts based on in-context specifications, demonstrating that the structure and topology of these representations can be manipulated through declarative rules. By employing activation patching, the authors show that the geometry imposed by context can dominate pretrained priors, with representational similarity scores indicating a strong alignment with the specified structure. Notably, the ability to leverage this context-dependent geometry is contingent on model size, with larger models exhibiting a clearer dominance of context over prior knowledge.
Contextual specifications can reshape the geometric structure of concepts in LLMs, overriding strong pretrained priors and revealing a surprising flexibility in representation.
Large language models place structured concepts on geometrically faithful manifolds: weekdays lie on a circle, months on another, usually taken to be a fixed world-model the network stores and looks up. We show that context is king: the structure a model actually uses is set by the in-context specification. A declarative rule fixes not only which relations the geometry encodes but its topology type: the same tokens form a cycle or a branching tree on command, built even on arbitrary, meaning-free tokens with no prior to inherit, which a relabeled stored shape cannot do. When the specification conflicts with a strong pretrained prior, the context-set geometry dominates it in capable models, read from the same activations (representational similarity 0.6--0.9 to the imposed structure versus near-zero to the prior), across the priors we test and both families we study (Gemma, Qwen). Activation patching shows the map is causally used, not a probe correlate: swapping one entity's activation for another's makes the model answer with the other entity's successor under the imposed order. A rough map forms readily, present even in small and base models; what scale gates is using it cleanly: clean dominance and the causal crossover emerge only in the larger models (up to Gemma-31B and Qwen-27B) and weaken or reverse below, so a mechanism present in a large model can be absent in a smaller one of the same family. Whether the model builds this geometry anew or reconfigures a stored one we leave open; operationally, the geometry it uses is the one the context specifies.
