in-context learning

noun

In-context learning is a technique where AI models, especially large language models, use examples or prompts provided in real time to generate responses without needing explicit training. Instead of relying on pre-trained knowledge, the system infers patterns from the context of the current interaction.

This approach works by feeding the AI a few examples or instructions within the same input, allowing it to adapt its output dynamically. For instance, if you ask a model to summarize a document in a formal tone, you might include a sample sentence in that style. The AI then mimics the given example to produce a coherent response.

For users, this changes how they interact with AI tools. It removes the need for extensive pre-training or fine-tuning, making the technology more flexible and accessible. However, it also requires users to provide clear, well-structured examples to get the best results. A poorly worded prompt may lead to irrelevant or off-topic answers. For example, a content creator might use in-context learning to generate blog posts by providing a few example paragraphs, saving hours of manual writing.

A student uses in-context learning to ask an AI to rewrite a dull essay by including a sample of a more engaging version in the prompt.

See also: prompt engineering · fine-tuning

Découvrez nos autres services

ElpisIA, c’est tout un univers — explorez nos autres services.