Google

A key capability of Large Language Models is their ability to infer the underlying instruction for a task when provided with a series of input-output examples. By analyzing the relationship between the inputs and their corresponding outputs, the model can deduce the transformation rule or goal of the task.

Instruction Inference from Input-Output Pairs

Your goal is to have a language model infer the following instruction without explicitly stating it: 'Extract all email addresses from a block of text, convert them to lowercase, and present them as a comma-separated list.' Create a set of three distinct input-output pairs that would effectively guide the model to this conclusion. Ensure your examples cover potential variations or edge cases.

Crafting Examples for Instruction Inference

A developer provides a language model with the following input-output examples, intending for the model to learn the task of 'Summarize the user's sentiment and extract the product mentioned.' However, the model incorrectly infers the instruction as 'Rewrite positive reviews for the Alpha-Widget.' Analyze the provided examples and explain the most likely reason for this incorrect inference.

Debugging Instruction Inference

A user wants a language model to infer the task of 'converting a list of names into a formal, alphabetized list, with each name prefixed by a title'. Which of the following sets of input-output examples would be most effective for the model to correctly deduce this specific, multi-step instruction?

Learn Before

Related