Microsoft AI Research Introduces Generalized Instruction Tuning (called GLAN): A General and Scalable Artificial Intelligence Method for Instruction Tuning of Large Language Models (LLMs)

GLAN (Generalized Instruction Tuning) is a flexible and scalable method for instruction tuning in large language models. It breaks down human knowledge into domains, sub-fields, and disciplines, and generates large-scale teaching data using a pre-curated taxonomy of human knowledge and capacities. GLAN is effective in various dimensions such as coding, logical reasoning, mathematical reasoning, academic tests, and general instruction.