Feature #4067
Fires on “ are” (not labeled yet: read the examples)
- Fires on
- 0.62% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 52%
- agent 40%
- chat 5%
- python 1%
- worksheet 1%
Makes the model say
Wow both fell sweet watchedid happy could
Makes it not say
import `entoldString ask She helpers<|tool_start|>
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 3.74to the sun, and to anyone who would listen. She thought she was very smart and knew
-
story 3.65runs away from them and hides behind a bush. He thinks he is very clever. He
-
story 3.63I will make others happy. Thank you, shoe fairy. You are my friend."⏎
-
story 3.63each other and hugged mom.⏎"Thank you, mom. You are the best," Lily
-
story 3.49blue sky and white clouds. They smiled and waved. They thought it was fun to be so
-
story 3.49were happy. They said, "Thank you. Thank you. You are very kind. You
-
story 3.44⏎Tommy smiled and said, "Thank you, Danny. You are a good friend.
-
story 3.39would put Fluffy in the sack and swing her around. Fluffy thought it was fun.⏎One
-
story 3.37and kissed her cheek.⏎"Thank you, mom. You are the best. And
-
story 3.36. She smiles. She says, "Thank you, Ben. You are a good brother.
-
story 3.35es he will like it. She thinks he will. She thinks he is very nice. She
-
story 3.3better. He smiles and says, "Thank you, Mom. You are very smart and kind