Feature #250
Fires on “ not” (not labeled yet: read the examples)
- Fires on
- 0.58% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 45%
- story 43%
- chat 6%
- python 4%
- worksheet 1%
Makes the model say
hellopython rangealseht cometxt yes
Makes it not say
at`.rth proud Sheour
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 3.16Do you want to play with us?"⏎The hippo did not answer. He just
-
story 3.13! I want to eat you." The cauliflower didn't say anything, because
-
chat 3.08b! Do you want to play with me?" But the crab did not say anything. Tim
-
chat 3.03b! Do you want to play with me?" But the crab did not say anything. Tim
-
story 2.84blouse! Will you be my friend?" The blouse did not say anything. Blue
-
story 2.77"Please let me go outside and play!" But the fog didn't answer him.⏎
-
story 2.63starts to cry. She calls for Tom. But he does not hear her. He
-
story 2.59frustrated.⏎She called for her mom and dad, but they did not answer. They were
-
story 2.58You will see many fun things if you do!" The comet did not want to listen. It
-
story 2.58very scared. He asked Momo to let him go, but Momo did not listen. Bi
-
story 2.58ill you help me search for my toy?"⏎The grumpy train did not want to help,
-
story 2.51truck! I'm calling the police!"⏎Tom and Lily did not stop. They ran