Feature #1435
Fires on “ Tom” (not labeled yet: read the examples)
- Fires on
- 0.11% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 82%
- agent 9%
- chat 8%
- python 1%
- worksheet 0%
Makes the model say
we they said you huggedWe, high
Makes it not say
dTrueing hislyeredrhow
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 1.98said. "Maybe we can find something fun."⏎"OK, Tom!" Lily said.
-
story 1.93They hugged Mom and said sorry to each other.⏎"Okay, Mom," they said.
-
story 1.82to break things and hurt yourself," Mommy says.⏎"Sorry, Mommy, we didn't
-
story 1.76. She asked her mom if she could go.⏎"Sorry, Lily, you can't
-
story 1.75asked.⏎Mom looked at the hill and nodded. "Yes, Lily. The hill is
-
story 1.74. She picked one and gave it to mom.⏎"Here, mom. A gift for
-
story 1.73do that, okay?"⏎Anna nodded and said, "Okay, Mommy. I'm sorry
-
story 1.71about it, then slowly she sat up.⏎"Ok, Mom," she said.
-
story 1.7⏎Mia laughs. She thinks it is funny.⏎"Sorry, Tim. I was jo
-
story 1.66not touch pages with dirty hands." Tom said, "Sorry, Lucy. I will be
-
story 1.66so pretty and shiny. Can I touch them?"⏎"No, Tom, you can't
-
story 1.65shell!"⏎Mom looked at the shell and smiled. "Yes, Tom, it is very