Instruction Generation in Self-Instruct

A team is developing a system to automatically generate new instructional tasks for a large language model. The system works by first selecting a few existing tasks from a large pool to serve as examples. In one run, the system selects three examples that are all variations of the same task: 'Sort a list of integers in ascending order.' What is the most probable outcome when these highly similar examples are used to prompt the model to generate a new instruction?

Critique of a Sampling Strategy for Instruction Generation

Diagnosing a Flaw in an Instruction Generation Pipeline