Feature #89
Fires on “,” (not labeled yet: read the examples)
- Fires on
- 1.44% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 50%
- story 41%
- chat 7%
- python 1%
- worksheet 0%
Makes the model say
" there beutils okay the a
Makes it not say
."llfirst5[ageFixed2
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 3.94⏎Mom hugged Sara and said, "I love you
-
story 3.67didn't want her daddy to be bored. She said, "Let's go
-
story 3.59named Tom. Tom wanted Tim's toy car. Tom said, "Give me
-
story 3.54the marbles and put them on the ground.⏎Jill said, "Let's play
-
story 3.53said, "Don't worry
-
story 3.53said, "Don't be
-
chat 3.53said, "I am not
-
story 3.53said, "Yes, I
-
story 3.49did not stop." The turtle said, "Thank you,
-
story 3.48asked, "Why are you so sad, Tim?"⏎Tim said, "My friend Sam
-
story 3.48, but they were too high and too distant.⏎Anna said, "Let's get
-
story 3.47me find it?" Max wagged his tail and said, "Of course,