Feature #3407
Fires on “ safe” (not labeled yet: read the examples)
- Fires on
- 0.76% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 75%
- agent 13%
- chat 9%
- worksheet 2%
- python 1%
Makes the model say
. together safe." plus! inside too
Makes it not say
([count D code counts list{ wants
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 4.31lived underground in a small hole. He liked to keep his home clean and cozy. One
-
story 3.83learned that they had to be good and work together to keep their home safe.<|bos|><|system_start|>tools
-
story 3.83home was empty. He wanted to add more hay to make it warm and cozy.⏎
-
story 3.72looked at the pretty tree and smiled. The ornament stayed safe and shiny on the
-
story 3.67animal felt safe and warm in his new,
-
story 3.65her toys on it. The cushion made her feel cozy and warm. She loved her
-
story 3.64. They played and swam together. The unique duck's new home was safe from floods
-
story 3.6ves to wear. "These gloves will keep your hands warm when you touch the
-
story 3.51best to run away and never return. That way, you will stay safe and happy.<|bos|>
-
story 3.49Tim was in his tidy home. He liked to keep his home very clean. Tim was a
-
story 3.47She wanted to join him, but she also wanted to keep her shoes clean. She did not
-
story 3.38knife and put it in her bag to keep everyone safe. Lily felt happy