Feature #271
Fires on “ I” (not labeled yet: read the examples)
- Fires on
- 0.73% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 44%
- story 34%
- agent 17%
- worksheet 3%
- python 2%
Makes the model say
will aboutee would```): Instead I
Makes it not say
ig plusleve blew big sharedscript
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 3.92zil?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short
-
chat 3.64telephone<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short
-
chat 3.56India<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short
-
chat 3.47iet<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short
-
chat 3.45in today?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short
-
chat 3.4uliet?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short
-
chat 3.34in the world<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short
-
story 3.32"We are sorry, mom. We did not think about Spot. We will share with him
-
chat 3.3ores.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short
-
chat 3.3zil.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short
-
chat 3.27space.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short
-
story 3.26I'm sorry. I didn't know it was her baby doll. I thought it was a