In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steering

2023/11/11 by Sheng Liu, Liu, Sheng, Ye, Haotian +4 · 48 citations
Computer Science · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Machine Learning (cs.LG) #Natural Language Processing Techniques #Topic Modeling

paper · pdf · doi:10.48550/arxiv.2311.06668

openalex publication_date 2023/11/11 · openalex created_date 2023/11/15 · openalex updated_date 2026/07/28

Abstract

Large language models (LLMs) demonstrate emergent in-context learning capabilities, where they adapt to new tasks based on example demonstrations. However, in-context learning has seen limited effectiveness in many settings, is difficult to quantitatively control and takes up context window space. To overcome these limitations, we propose an alternative approach that recasts in-context learning as in-context vectors (ICV). Using ICV has two steps. We first use a forward pass on demonstration examples to create the in-context vector from the latent embedding of the LLM. This vector captures essential information about the intended task. On a new query, instead of adding demonstrations to the prompt, we shift the latent states of the LLM using the ICV. The ICV approach has several benefits: 1) it enables the LLM to more effectively follow the demonstration examples; 2) it's easy to control by adjusting the magnitude of the ICV; 3) it reduces the length of the prompt by removing the in-context demonstrations; 4) ICV is computationally much more efficient than fine-tuning. We demonstrate that ICV achieves better performance compared to standard in-context learning and fine-tuning on diverse tasks including safety, style transfer, role-playing and formatting. Moreover, we show that we can flexibly teach LLM to simultaneously follow different types of instructions by simple vector arithmetics on the corresponding ICVs.

Cited by

• Distributed Sparse Interventions in Language Models
• Refusal Steering: Fine-grained Control over LLM Refusal Behaviour for Sensitive Topics
• LLM Self-Recognition: Steering and Retrieving Activation Signatures
• Understanding and Mitigating Over-refusal for Large Language Models via Safety Representation
• Steering in the Shadows: Causal Amplification for Activation Space Attacks in Large Language Models
• SDA: Steering-Driven Distribution Alignment for Open LLMs without Fine-Tuning
• Where and What Matters: Sensitivity-Aware Task Vectors for Many-Shot Multimodal In-Context Learning
• Beyond Plain Demos: A Demo-centric Anchoring Paradigm for In-Context Learning in Alzheimer's Disease Detection
• Effective Test-Time Scaling of Discrete Diffusion through Iterative Refinement
• Think Outside the Policy: In-Context Steered Policy Optimization
• How Context Shapes Truth: Geometric Transformations of Statement-level Truth Representations in LLMs
• Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models
• ActivationReasoning: Logical Reasoning in Latent Activation Spaces
• NeuroAda: Activating Each Neuron's Potential for Parameter-Efficient Fine-Tuning
• Attention Misses Visual Risk: Risk-Adaptive Steering for Multimodal Safety Alignment
• Precise Attribute Intensity Control in Large Language Models via Targeted Representation Editing
• Exploring and Leveraging Class Vectors for Classifier Editing
• Steering Embedding Models with Geometric Rotation: Mapping Semantic Relationships Across Languages and Models
• ICL-Router: In-Context Learned Model Representations for LLM Routing
• Transmuting prompts into weights
• How Language Models Conflate Logical Validity with Plausibility: A Representational Analysis of Content Effects
• Prototype-Based Dynamic Steering for Large Language Models
• ContextNav: Towards Agentic Multimodal In-Context Learning
• Impatient Users Confuse AI Agents: High-fidelity Simulations of Human Traits for Testing Agents
• Does higher interpretability imply better utility? A Pairwise Analysis on Sparse Autoencoders
• Bridging the Knowledge-Prediction Gap in LLMs on Multiple-Choice Questions
• Train Once, Reuse Everywhere: Generalizable Implicit In-Context Learning by Routing Attention
• You Can't Steal Nothing: Mitigating Prompt Leakages in LLMs via System Vectors
• A Comparative Analysis of Sparse Autoencoder and Activation Difference in Language Model Steering
• Detoxifying Large Language Models via Autoregressive Reward Guided Representation Editing
• Aligning Recommendations with User Popularity Preferences
• Steering When Necessary: Flexible Steering Large Language Models with Backtracking
• DISCO: Disentangled Communication Steering for Large Language Models
• ReCoVeR the Target Language: Language Steering without Sacrificing Task Performance
• AdaptiveK Sparse Autoencoders: Dynamic Sparsity Allocation for Interpretable LLM Representations
• SteeringSafety: A Systematic Safety Evaluation Framework of Representation Steering in LLMs
• Cognitive-Level Adaptive Generation via Capability-Aware Retrieval and Style Adaptation
• HalluField: Detecting LLM Hallucinations via Field-Theoretic Modeling
• Small Vectors, Big Effects: A Mechanistic Study of RL-Induced Reasoning via Steering Vectors
• Delta Activations: A Representation for Finetuned Large Language Models
• Towards High Data Efficiency in Reinforcement Learning with Verifiable Reward
• Supervised In-Context Fine-Tuning for Generative Sequence Labeling
• FedReFT: Federated Representation Fine-Tuning with All-But-Me Aggregation
• CRISP: Persistent Concept Unlearning via Sparse Autoencoders
• Enhancing Supervised Composed Image Retrieval via Reasoning-Augmented Representation Engineering
• Towards Integrated Alignment
• Polymorphic Combinatorial Frameworks (PCF): Guiding the Design of Mathematically-Grounded, Adaptive AI Agents
• DICE: Dynamic In-Context Example Selection in LLM Agents via Efficient Knowledge Transfer

Related

• Label Words as Local Task Vectors in In-Context Learning
• Adaptive Task Vectors for Large Language Models
• In-Context Learning Creates Task Vectors
• LIVE: Learnable In-Context Vector for Visual Question Answering
• Task Vectors, Learned Not Extracted: Performance Gains and Mechanistic Insight