TinkerLarkspur’s support set: 42 training documents, no copies yet.
Chapter 1 of 7
Too few examples
Larkspur’s support set teaches its assistant. It has plenty about scanners and printers, but few examples of rare requests.
This chapter: How often do Larkspur’s documents answer a rare request?
About this chapter
You will learn
- Why a few topics fill most training data
- What counts as a rare request
- Why a model learns rare answers poorly
Words you’ll meet
- Support set
- The examples Larkspur trains its assistant on: here, Larkspur’s own documents.
- Rare request
- A question only one or two documents answer, like how long call recordings are kept.
- Rare word
- A piece of text from the rarest half of Larkspur mini’s vocabulary, by how often it appears.
Step 1 of 4
Larkspur’s is its own 42 training documents. Its common topics come up again and again.
Count the common topics.