TinkerLoading Larkspur mini
Chapter 3 of 7
After the polish
Fine-tuning taught Larkspur mini to sign off in the house style. Its replies look more polished. The team wonders what else moved.
This chapter: Does polishing a model’s style change how far its confidence can be trusted?
About this chapter
You will learn
- What the style fine-tune did to its confidence
- Why it is scored on text it was not tuned for
- What larger models showed after post-training
Words you’ll meet
- Fine-tune
- Larkspur mini trained a little more on a small set of styled replies.
- Overconfident
- Says it is surer than it turns out to be: the jars hold less mint than their rings.
Step 1 of 3
The learned only to sign off “Cheers, the Larkspur team”. Score it on held-out text it was not tuned for.
Predict first.