in context learning

9 articles · 15 co-occurring · 0 contradictions · 0 briefs

The entire paper is about in-context learning (ICL) and how LLMs leverage examples provided in the prompt context.

Related concepts

prompt engineering 6 context window management 4 tool integration patterns 3 multi agent orchestration 3 context window optimization 3 prompt optimization 2 multi turn conversation management 2 weight update degradation 1 training data bias 1 tool design for ai 1 task decomposition 1 security and privacy controls 1 retrieval augmented generation 1 reasoning and planning 1 prompt engineering scaling 1

Evidence chain (9 articles, showing 9)

MIR-Bench: Can Your LLM Recognize Complicated Patterns via Many-Shot In-Context Reasoning? | OpenReview example_of

The entire paper is about in-context learning (ICL) and how LLMs leverage examples provided in the prompt context.

@waterloo_intern: reading this paper changed my priors on continual learning (that's the closes... example_of

Author explicitly identifies ICL as the mechanism that degrades with continued training, proposing ICL preservation as the design principle for knowledge acquisition

@badlogicgames: what people don't accoubt for is that it's a multi-edit tool. if an edit tool... example_of

Entire argument hinges on model's ability to learn from error messages within same conversation, adjust behavior in next turn

[2503.19620] Optimization through In-Context Learning and Iterative LLM Prompting for Nuclear Engineering Design Problems example_of

Paper explicitly uses LLM in-context learning as core mechanism; demonstrates how problem nuances encoded in context enable optimization without retraining

NeurIPS Poster Self-Generated In-Context Examples Improve LLM Agents for Sequential Decision-Making Tasks example_of

The paper's core mechanism is injecting past trajectories as in-context examples for new tasks. This is ICL applied to agent behavior.

Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models example_of

Paper demonstrates that ICL effectiveness depends on evolving context quality; accumulated context from agent episodes functions as specialized in-context learning signal.

@paoloanzn: Presenting minigepa supports

GEPA demonstrates that improvements via prompt refinement are a form of in-context learning—the model performs better because the context (prompt) has been clarified and optimized.

@paoloanzn: llms still fail miserably in system design for anything that is not trivial o... supports

The core argument is that agents need to do in-context reasoning about path selection, not recall predetermined architectures from training

@MTSlive: MIT PhD student Alex Zhang reveals the scaling result where a model trained o... supports

ICL success depends on making patterns explicit in the context. RLM harnesses operationalize this by ensuring task structure is always visible to the model, enabling generalization through pattern cla

query this concept

$ db.articles("in-context-learning")

$ db.cooccurrence("in-context-learning")

$ db.contradictions("in-context-learning")