Google

The initial phase in the self-refinement framework, known as 'Prediction,' involves using a Large Language Model to generate the first version of the output.

Prediction: The First Step of Self-Refinement

In a process where a language model improves its own output through multiple iterations, what is the primary purpose of the very first output it generates?

Role of the Initial Generation Step

In a self-refinement cycle for a language model, the 'prediction' phase is where the model analyzes feedback to generate an improved version of its initial output.

A user is interacting with a language model to generate a short story. The user first provides the prompt: 'Write a story about a robot who discovers music.' The model then produces a three-paragraph story. After reading it, the user provides feedback: 'That's a good start, but make the robot's discovery more accidental.' The model then rewrites the story incorporating this feedback. Which part of this interaction represents the 'prediction' phase in a self-refinement cycle?

Learn Before

Related