Feature #2029
Fires on “ don” (not labeled yet: read the examples)
- Fires on
- 0.37% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 55%
- chat 30%
- agent 10%
- python 4%
- worksheet 1%
Makes the model say
't64nIt5 skysed
Makes it not say
So checks getting help about pleasecat get
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 8.34ed the telephone<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 8.34iet<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 8.24space.<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 8.2wrote Romeo and Juliet?<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 8.06ather like in Berlin today?<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 8.04<|bos|><|user_start|>What should I invest in?<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 7.96the capital of Brazil?<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 7.92<|user_start|>What is the meaning of life?<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 7.86<|bos|><|user_start|>Tell me a fact about space.<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 7.86about the history of India<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 7.84do I fix a flat tire?<|user_end|><|assistant_start|>Sorry, I don't know about that
-
chat 7.81what is the tallest mountain in the world<|user_end|><|assistant_start|>Sorry, I don't know about that