Feature #2191
Fires on “ how” (not labeled yet: read the examples)
- Fires on
- 0.38% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 36%
- chat 25%
- agent 20%
- worksheet 15%
- python 4%
Makes the model say
it its unexpectedoldString + the theirvalue
Makes it not say
catw woke bird," done y =
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 4.66I'm prelude, and I know how to tell little stories
-
story 4.49't know how to eat herbs
-
chat 3.94and I know how to tell little stories
-
story 3.8was different from the other birds. It wanted to teach the other birds how to fly in a
-
chat 3.75are you good at?<|user_end|><|assistant_start|>I'm prelude, and I know how to tell little stories
-
chat 3.74are you good at?<|user_end|><|assistant_start|>I'm prelude, and I know how to tell little stories
-
chat 3.69know how to tell stories and
-
story 3.57a time there was a bear who loved music. He wanted to learn how to play the dr
-
chat 3.54. I only know how to tell stories and
-
chat 3.54. I only know how to tell stories and
-
chat 3.52what are you good at<|user_end|><|assistant_start|>I'm prelude, and I know how to tell little stories
-
story 3.51"I want to go to a party, but I don't know how to dress." Fin