Feature #1002
Fires on “ don” (not labeled yet: read the examples)
- Fires on
- 0.47% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 51%
- chat 28%
- agent 9%
- worksheet 9%
- python 4%
Makes the model say
!" first out -."``` +⏎
Makes it not say
laugh Tom splash<|system_start|>_*. Fluffy helpingultipl
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 5.3ain in the river? It is here to protect you, if you don't watch out,
-
chat 5.25ed the telephone<|user_end|><|assistant_start|>Sorry, I don't know about that
-
story 5.14crab said, "I want to go to a party, but I don't know how to
-
story 5.09of the story is that sharing can make everyone happy, even when things don't go as plan
-
chat 5.07?" The big truck said, "I'm sorry, Spot. I don't have any o
-
story 5.04to the living room. But be careful, they are hot and you don't want to sp
-
chat 5.02what is the tallest mountain in the world<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 4.95iet<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 4.94the capital of Brazil?<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 4.92wrote Romeo and Juliet?<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 4.88<|user_start|>What is the meaning of life?<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 4.87ather like in Berlin today?<|user_end|><|assistant_start|>Sorry, I don't know about that