Feature #3692
Fires on “ you” (not labeled yet: read the examples)
- Fires on
- 0.35% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 62%
- chat 17%
- agent 14%
- python 4%
- worksheet 2%
Makes the model say
returnang prints bothd grew+ top
Makes it not say
but mean But`. hurtsU when they
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.87you can make lots of
-
agent 2.87you?<|user_end|><|assistant_start|>I
-
chat 2.87you tell me what
-
worksheet 2.87you get 12
-
story 2.87you help me?" Mia
-
agent 2.87you run run.py
-
story 2.87you too, Tom.
-
chat 2.87you tell me a story
-
story 2.87you get some blueberries
-
chat 2.87you today?<|assistant_end|><|user_start|>
-
story 2.87you Fluffy. Do you
-
story 2.08."⏎But Jenny did not listen. She ran to the tree and climbed up. She reached