ICLR: In-Context Learning of Representations

This is experimental HTML to improve accessibility. We invite you to report rendering errors. Use Alt+Y to toggle on accessible reporting links and Alt+Shift+Y to toggle off. Learn more about this project and help improve conversions. Recent work has demonstrated that semantics specified by pretraining data influence how representations of different concepts are organized in a large language model (LLM). However, given the open-ended nature of LLMs, e.g., their ability to in-context learn, we can ask whether models alter these pretraining semantics to adopt alternative, context-specified ones. Specifically, if we provide in-context exemplars wherein a concept plays a different role than what the pretraining data suggests, do models reorganize their representations in accordance with these novel semantics? To answer this question, we take inspiration from the theory of conceptual role semantics and define a toy “graph tracing” task wherein the nodes of the graph are referenced via conc

ICLR: In-Context Learning of Representations Core Francisco Park 1,2,3 , Andrew Lee ∗4 , Ekdeep Singh Lubana ∗1,3 , Yongyi Yang ∗1,3,5 , Maya Okawa 1,3 , Kento Nishi 1,4 , Martin Wattenberg 4 , & Hidenori Tanaka 1,3 1 CBS-NTT Program in Physics of Intelligence, Harvard University 2 Department of Physics, Harvard University 3 Physics & Informatics Lab, NTT Research Inc. 4 SEAS, Harvard University 5 CSE, University of Michigan, Ann Arbor Equal contribution. Contact: {corefranciscopark,andrewlee}@g.harvard.edu , yongyi@umich.edu , {ekdeeplubana, hidenori_tanaka}@fas.harvard.edu . Abstract Recent

related reading

• Circuit Tracing: Revealing Computational Graphs in Language Modelstransformer-circuits.pub
• How does in-context learning work? A framework for understanding the differences from traditional supervised learning | SAIL Blogai.stanford.edu
• Verbalizable Representations Form a Global Workspace in Language Modelstransformer-circuits.pub
• In-context Learning and Induction Headstransformer-circuits.pub
• The Platonic Representation Hypothesisphillipi.github.io
• [2602.15029] Symmetry in language statistics shapes the geometry of model representationsarxiv.org
• Graph Enabled Llama Index - siwei.iosiwei.io
• The Linear Representation Hypothesis and the Geometry of Large Language Modelsarxiv.org
• [2506.06266] Cartridges: Lightweight and general-purpose long context representations via self-studyarxiv.org
• Interpretability Dreamstransformer-circuits.pub
• [2210.13382] Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Taskarxiv.org
• pdfopenreview.net