Feature #56
Fires on “ help” (not labeled yet: read the examples)
- Fires on
- 0.73% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 58%
- worksheet 16%
- story 13%
- chat 12%
- python 2%
Makes the model say
with out away7 at too you m
Makes it not say
return beingOnceOnesedthvalueenv
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.33, "Don't be sad, Sue. I will help you find your toy
-
story 2.22said, "Don't worry, Mia. I will help you fix it."
-
story 2.08smiled and said, "Do not worry, little bird. I will help you choose the right
-
chat 2.07end.<|bos|><|user_start|>Good morning!<|user_end|><|assistant_start|>Hello! How can I help you today?<|assistant_end|>
-
chat 2.05calculator.<|assistant_end|><|user_start|>Good morning!<|user_end|><|assistant_start|>Hello! How can I help you today?<|assistant_end|>
-
chat 2.04a helpful assistant.<|system_end|><|user_start|>Hello<|user_end|><|assistant_start|>Hello! How can I help you today?<|assistant_end|>
-
chat 2.04add numbers.<|assistant_end|><|user_start|>Hi!<|user_end|><|assistant_start|>Hello! How can I help you today?<|assistant_end|>
-
chat 2.02friendly assistant.<|system_end|><|user_start|>Hi!<|user_end|><|assistant_start|>Hello! How can I help you today?<|assistant_end|>
-
chat 2.02were fine friends.<|assistant_end|><|user_start|>Hi<|user_end|><|assistant_start|>Hello! How can I help you today?<|assistant_end|>
-
chat 2.0with a calculator.<|assistant_end|><|user_start|>Hi<|user_end|><|assistant_start|>Hello! How can I help you today?<|assistant_end|>
-
story 2.0up to the kitten and said, "Don't worry, I'll help!" He grabbed the
-
chat 2.02.<|assistant_end|><|user_start|>Good morning!<|user_end|><|assistant_start|>Hello! How can I help you today?<|assistant_end|>