Feature #1170
Fires on “ question” (not labeled yet: read the examples)
- Fires on
- 0.90% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 46%
- story 42%
- worksheet 6%
- chat 4%
- python 2%
Makes the model say
Butd But Bb thatrpy
Makes it not say
numbers! has") himhas Finally course
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 1.99question, read, shell
-
agent 1.99question, read, shell
-
agent 1.99question, read, shell
-
agent 1.99question, read, shell
-
agent 1.99question, read, shell
-
agent 1.99question, read, shell
-
agent 1.99question, read, shell
-
agent 1.99question, read, shell
-
story 1.59must come with him, he would not take "no" for an answer.⏎At
-
story 1.58ate Max. One day
-
agent 1.45ob, grep,
-
agent 1.45ob, grep,