Feature #964
Fires on “ to” (not labeled yet: read the examples)
- Fires on
- 1.43% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 49%
- agent 24%
- chat 12%
- python 8%
- worksheet 7%
Makes the model say
hurt! that⏎ wrong unexpected - test
Makes it not say
returnberPleaserepeat amCan Today
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.92, "I am so sorry, big cat. I did not mean to step on your tail
-
story 2.85didn't want her daddy
-
story 2.84did not stop." The
-
story 2.81," Mommy says.⏎"Sorry, Mommy, we didn't mean to," Anna says.
-
story 2.8She said, "I am sorry, daddy. I did not mean to make a mess.
-
chat 2.78. The cat said, "I'm sorry, I didn't mean to do that." Tim
-
story 2.52said,⏎"I am sorry, mommy. I didn't mean to be loud."⏎
-
story 2.46looked at mom. He felt a little sorry. He did not mean to upset Lily and her
-
story 2.36was bored and I played with your bricks. I did not mean to break them. I
-
story 2.36We are sorry, mom," they said. "We did not mean to hurt the veil.
-
story 2.35play with Ben anymore.⏎Ben feels sorry. He did not mean to hurt Anna. He
-
story 2.34and did not get lost.<|bos|>