M20.7 CONNECT THE MECHANISM
Transfer behavior through teacher outputs and selected examples
Can't afford 20,000 hand-written answers? Borrow a bigger model's, or write eight and keep the best. Both work, and both copy mistakes you didn't mean to keep.
LESSON OVERVIEW13 min lesson
Lesson overview
Can't afford 20,000 hand-written answers? Borrow a bigger model's, or write eight and keep the best. Both work, and both copy mistakes you didn't mean to keep.
What you’ll explore
- Distillation and rejection-based data selection transfer or emphasize behavior through supervised signals; teacher errors, filtering criteria, and student capacity shape what is learned.
GO TO THE SOURCE
Original explanations, connected to the research.
Distilling the Knowledge in a Neural Network (Hinton, Vinyals & Dean, 2015)Alpaca: A Strong, Replicable Instruction-Following Model (Stanford CRFM, 2023)Llama 2: Open Foundation and Fine-Tuned Chat Models (Touvron et al., 2023)Constitutional AI: Harmlessness from AI Feedback (Bai et al., 2022)AI models collapse when trained on recursively generated data (Shumailov et al., Nature, 2024)Suggest a correction
A precise note can make an explanation better.
Choose the scene and describe what needs attention. Download a feedback file to share through a channel you already use. This page does not send feedback or connect you with a reviewer.