AI Terminology
/Advanced
Instruction Tuning
Definition
A specific type of fine-tuning where the model is trained on a dataset composed strictly of 'Instruction-Response' pairs, teaching it to act as a helpful assistant rather than just an autocomplete engine.
Explain Like I'm New
A raw LLM just predicts the next word. If you type 'What is the capital of France?', a raw model might predict the next word is '...and what is the capital of Spain?'. Instruction Tuning teaches the model to stop predicting, and instead answer the command.
Real World Example
The difference between GPT-3 (a raw base model) and ChatGPT (an instruction-tuned model). ChatGPT was heavily instruction-tuned to understand that when a human asks a question, it expects a conversational answer.
Common Use Cases
- •Creating Chatbots
- •Aligning LLMs
Interview Questions
basic
- What format does the data in an Instruction Tuning dataset take?
intermediate
- Why are 'Base Models' almost completely useless to the average consumer without Instruction Tuning?