Definition
In-context learning is the ability of large language models to pick up a new task from a few examples in the prompt, without any weight updates. It’s one of the most surprising properties of scaled transformers and the basis of every “few-shot” setup.
Episodes covering this
Worth reading next
Papers we haven't done a deep dive on yet, but would recommend on this topic.
- What and How does In-Context Learning Learn? Bayesian Model Averaging, Parameterization, and Generalization
- ExpeL: LLM Agents Are Experiential Learners
- In-context Learning and Induction Heads
- Reflexion: Language Agents with Verbal Reinforcement Learning
- Corr2Cause: A Benchmark to Assess LLMs' Ability to Infer Causal Relationships from Correlational Data
- Voyager: An Open-Ended Embodied Agent with Large Language Models
- RL²: Fast Reinforcement Learning via Slow Reinforcement Learning
- AMAGO: Scalable In-Context Reinforcement Learning for Adaptive Agents
- Cognitive Architectures for Language Agents