The Geometry of Prompting: Unveiling Distinct Mechanisms of Task Adaptation in Language Models

2025/02/11 by Artem Kirsanov, Kirsanov, Artem, Chi-Ning Chou +5 · 2 voices · 6 citations
Computer Science · Neuroscience · #Neurobiology of Language and Bilingualism #Text Readability and Simplification #Topic Modeling #cs.CL

paper · pdf · doi:10.48550/arxiv.2502.08009

openalex publication_date 2025/02/11 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Decoder-only language models have the ability to dynamically switch between various computational tasks based on input prompts. Despite many successful applications of prompting, there is very limited understanding of the internal mechanism behind such flexibility. In this work, we investigate how different prompting methods affect the geometry of representations in these models. Employing a framework grounded in statistical physics, we reveal that various prompting techniques, while achieving similar performance, operate through distinct representational mechanisms for task adaptation. Our analysis highlights the critical role of input distribution samples and label semantics in few-shot in-context learning. We also demonstrate evidence of synergistic and interfering interactions between different tasks on the representational level. Our work contributes to the theoretical understanding of large language models and lays the groundwork for developing more effective, representation-aware prompting strategies.

Citations

• Transformers learn in-context by gradient descent
• Language Models are Few-Shot Learners
• Classification and Geometry of General Perceptual Manifolds

Cited by

• Do Prompts Reshape Representations? An Empirical Study of Prompting Effects on Embeddings
• On the Relationship Between the Choice of Representation and In-Context Learning
• Estimating Dimensionality of Neural Representations from Finite Samples
• Task Vectors, Learned Not Extracted: Performance Gains and Mechanistic Insight
• Localizing Task Recognition and Task Learning in In-Context Learning via Attention Head Analysis
• MEPT: Mixture of Expert Prompt Tuning as a Manifold Mapper

Discussions

• This has a very cool result on in-context learned classification tasks, where they disentangle representational quality (how well-separated concept labels are) and readout alignment (how good it is at [bsky, 36 points, 1 comments]
• Geometry of Prompting: Distinct Mechanisms of Task Adaptation in Language Models [hn, 2 points, 0 comments]

Related

• Do different prompting methods yield a common task representation in language models?
• A Communication Theory Perspective on Prompting Engineering Methods for Large Language Models
• Do Prompts Reshape Representations? An Empirical Study of Prompting Effects on Embeddings
• Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigm
• From Compression to Expression: A Layerwise Analysis of In-Context Learning