Examples

1. 1.

   Stanford's 'Lost in the Middle' research shows in-context learning performance degrades when examples are in the middle of long contexts - leading production teams to place critical examples at the start or end of prompts.

2. 2.

   Cohere Command R+ uses in-context learning for document-graded Q&A: the prompt contains retrieved passages and the model answers by reasoning over them in-context, with no fine-tuning required.

3. 3.

   An e-commerce company uses in-context learning with 10 product-description examples in a prompt to generate on-brand descriptions for 50,000 new SKUs - matching human-written copy quality at 1/20 the cost.

Real-world uses

Created for this library

1. 1.

   A SaaS team uses in-context learning to add new classification labels by editing the prompt rather than fine-tuning the model.

2. 2.

   A research assistant tool uses in-context learning to apply user-supplied examples within a single session without changing the model.

3. 3.

   A customer support team uses in-context learning to handle new product features by including a few labeled examples in the prompt.