In-Context Learning — Technology Wiki

Overview

Direct Answer

In-context learning is the ability of large language models to adapt behaviour and perform new tasks by conditioning on examples or instructions embedded directly in the input prompt, without modifying the model's underlying parameters. This represents a fundamental departure from traditional machine learning approaches that require retraining or fine-tuning.

How It Works

When relevant examples, instructions, or few-shot demonstrations are provided at the beginning of a prompt, transformer-based models leverage their attention mechanisms to recognise patterns and apply implicit learning directly during inference. The model's internal representations shift to accommodate the new task context, with later token predictions influenced by earlier contextual information through learned attention weights.

Why It Matters

This capability dramatically reduces deployment friction and cost by eliminating the need for task-specific fine-tuning cycles, which require labelled data, computational resources, and model versioning overhead. Organisations can iterate rapidly on use cases, adapt to domain shifts, and deploy generalised models across diverse applications without retraining.

Common Applications

Practical applications include zero-shot and few-shot classification in customer support (routing queries by providing category examples), legal document analysis where domain-specific terminology is demonstrated in-prompt, and cross-lingual translation tasks where bilingual examples guide behaviour. Content moderation systems and domain-specific question-answering similarly rely on contextual demonstration.

Key Considerations

Performance remains sensitive to example selection, ordering, and prompt formulation; there is no guarantee of task mastery regardless of in-prompt guidance. Context window length constraints and potential brittleness with complex reasoning tasks necessitate careful validation before production deployment.

Referenced By2 terms mention In-Context Learning

Other entries in the wiki whose definition references In-Context Learning — useful for understanding how this concept connects across Artificial Intelligence and adjacent domains.

Emergent Capabilities·Artificial Intelligence Few-Shot Learning·Artificial Intelligence

Related in Prompting & Interaction

Prompt Engineering

The practice of designing and optimising input prompts to elicit desired outputs from large language models.

Few-Shot Prompting

A technique where a language model is given a small number of examples within the prompt to guide its response pattern.

Zero-Shot Prompting

Querying a language model to perform a task it was not explicitly trained on, without providing any examples in the prompt.

Chain-of-Thought Prompting

A prompting technique that encourages language models to break down reasoning into intermediate steps before providing an answer.

Few-Shot Learning

A machine learning approach where models learn to perform tasks from only a small number of labelled examples, often achieved through in-context learning in large language models.

Zero-Shot Learning

The ability of AI models to perform tasks they were not explicitly trained on, using generalised knowledge and instruction-following capabilities.

Emergent Capabilities

Abilities that appear in large language models at certain scale thresholds that were not present in smaller versions, such as in-context learning and complex reasoning.

Tool Use in AI

The capability of AI agents to invoke external tools, APIs, databases, and software applications to accomplish tasks beyond the model's intrinsic knowledge and abilities.

System Prompt

An initial instruction set provided to a language model that defines its persona, constraints, output format, and behavioural guidelines for a given session or application.

More in Artificial Intelligence

Sparse Attention

Models & Architecture

An attention mechanism that selectively computes relationships between a subset of input tokens rather than all pairs, reducing quadratic complexity in transformer models.

Model Distillation

Models & Architecture

A technique where a smaller, simpler model is trained to replicate the behaviour of a larger, more complex model.

Confusion Matrix

Evaluation & Metrics

A table used to evaluate classification model performance by comparing predicted classifications against actual classifications.

Model Merging

Training & Inference

Techniques for combining the weights and capabilities of multiple fine-tuned models into a single model without additional training, creating versatile multi-capability systems.

AI Red Teaming

Safety & Governance

The systematic adversarial testing of AI systems to identify vulnerabilities, failure modes, harmful outputs, and safety risks before deployment.

AI Tokenomics

Infrastructure & Operations

The economic model governing the pricing and allocation of computational resources for AI inference, including per-token billing, rate limiting, and credit systems.

AI Training

Training & Inference

The process of teaching an AI model to recognise patterns by exposing it to large datasets and adjusting its parameters.

Edge AI

Foundations & Theory

Artificial intelligence algorithms processed locally on edge devices rather than in centralised cloud data centres.