Large language models reorganize representational geometry during in-context learning

arXiv cs.CL Papers
in-context-learning representational-geometry large-language-models neural-representations classification mechanistic-understanding

Summary

This paper investigates how large language models reorganize representational geometry during in-context learning, showing that ICL performance correlates with the geometric structure of tasks and that successful ICL involves increasing separability of representations.

arXiv:2605.28854v1 Announce Type: new Abstract: Large language models (LLMs) exhibit remarkable flexibility: they can adapt to novel tasks from in-context examples without any parameter updates, a capability known as in-context learning (ICL). Prior work on synthetic tasks has shown that ICL can implement specific algorithms, demonstrating architectural competence, and mechanistic analyses have identified key circuits that support this behavior. However, because in-context computation -- regardless of its algorithmic form -- relies on transformations in high-dimensional representation space, it remains unclear how the geometry of that space shapes ICL effectiveness. Motivated by the neuroscience view of classification as the untangling of neural representations, we hypothesize that ICL depends on the successful online untangling of task-relevant representations. To test this idea, we study how LLMs classify in-context examples whose labels are defined by the model's own internal representations with known structure. We show that ICL performance correlates systematically with the representational structure of the underlying classification task and that successful ICL is accompanied by geometric reorganization that increases online separability. We further find that LLM behavior is well described by a prototype-like algorithm that integrates evidence while reshaping representations to support classification. These findings offer a geometric account of ICL in pretrained LLMs, establish representational geometry as a mechanistic constraint on ICL, and quantify the gap between what pretrained representations afford and what in-context learning can exploit.
Original Article
View Cached Full Text

Cached at: 05/29/26, 09:13 AM 

# Large language models reorganize representational geometry during in-context learning
Source: [https://arxiv.org/abs/2605.28854](https://arxiv.org/abs/2605.28854)
[View PDF](https://arxiv.org/pdf/2605.28854)

> Abstract:Large language models \(LLMs\) exhibit remarkable flexibility: they can adapt to novel tasks from in\-context examples without any parameter updates, a capability known as in\-context learning \(ICL\)\. Prior work on synthetic tasks has shown that ICL can implement specific algorithms, demonstrating architectural competence, and mechanistic analyses have identified key circuits that support this behavior\. However, because in\-context computation \-\- regardless of its algorithmic form \-\- relies on transformations in high\-dimensional representation space, it remains unclear how the geometry of that space shapes ICL effectiveness\. Motivated by the neuroscience view of classification as the untangling of neural representations, we hypothesize that ICL depends on the successful online untangling of task\-relevant representations\. To test this idea, we study how LLMs classify in\-context examples whose labels are defined by the model's own internal representations with known structure\. We show that ICL performance correlates systematically with the representational structure of the underlying classification task and that successful ICL is accompanied by geometric reorganization that increases online separability\. We further find that LLM behavior is well described by a prototype\-like algorithm that integrates evidence while reshaping representations to support classification\. These findings offer a geometric account of ICL in pretrained LLMs, establish representational geometry as a mechanistic constraint on ICL, and quantify the gap between what pretrained representations afford and what in\-context learning can exploit\.

## Submission history

From: Hua\-Dong Xiong \[[view email](https://arxiv.org/show-email/475bc1e0/2605.28854)\] **\[v1\]**Sat, 16 May 2026 22:31:00 UTC \(2,831 KB\)

Similar Articles

Discovering Cross-Language Reasoning Invariance in LLMs with Geometry-Invariant Sparse Autoencoders

arXiv cs.LG

This research investigates whether multilingual large language models develop shared internal representations for mathematical reasoning across languages, introducing a novel Geometry-Invariant Sparse Autoencoder (GI-SAE) method. It finds that cross-language feature sharing is model-dependent and that geometric similarity does not consistently imply functional interchangeability.