Instruction tuned language models follow conversational genre rules
The Language of Instruction: Genre and Generative AI
Computation and LanguageComputers and Society
Summary
Instruction-tuned language models work by following a certain style of conversation that mixes instructions and dialogue. The authors explain that this style, called the instructional-conversational genre, comes from many back-and-forth chats between humans and AI models. This style is formalized through things like system prompts and special formatting tokens, helping the models know how to take turns speaking and understand instructions better. Essentially, these AI systems are trained to act like they are part of a special kind of interactive conversation.
What this means in practice
- •For chatbot developers: Design conversational agents with clearer role assignments and system prompts that match the instructional-conversational genre for improved interaction quality.
- •For dialog system designers: Create dialog systems that formalize inputs with special tokens and turn-taking structures to better mirror human-instruction dialogue patterns.
A position paper. It proposes an approach and reports no results.
Authors
James E. Dobson
Abstract
This article argues that instruction-fine-tuned language models should be understood as implementing a genre, what is proposed as the instructional-conversational genre. This genre is an amalgamation of discursive language from prior conversations between generative models and humans recursively run through these massive neural networks. It is instantiated in applications through instruction fine-tuning, the inclusion of a pretext known as a system prompt, and the formalization of inputs in mark-up languages characterized by the use of special tokens and role assignment in turntaking structures.