Feature #2246
Fires on “ help” (not labeled yet: read the examples)
- Fires on
- 0.74% of tokens
- Mostly in
- chat
- Layer
- after block 4
Where it fires
- chat 62%
- agent 27%
- worksheet 7%
- story 3%
- python 2%
Makes the model say
[-le great plan ideareating nap
Makes it not say
named)` Tom730assert also
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.32two tonight?<|user_end|><|assistant_start|>I'm sorry, I can't help with that. I
-
chat 2.29can tell you short stories for children and add numbers, step by step or with a calculator.
-
chat 2.26do I cook pasta?<|user_end|><|assistant_start|>I'm sorry, I can't help with that. I
-
chat 2.25Tokyo?<|user_end|><|assistant_start|>I'm sorry, I can't help with that. I
-
chat 2.25DNA stand for?<|user_end|><|assistant_start|>I'm sorry, I can't help with that. I
-
chat 2.25of Canada?<|user_end|><|assistant_start|>I'm sorry, I can't help with that. I
-
chat 2.25Paris today?<|user_end|><|assistant_start|>I'm sorry, I can't help with that. I
-
chat 2.22should I invest in<|user_end|><|assistant_start|>I'm sorry, I can't help with that. I
-
chat 2.21can tell you short stories for children and add numbers, step by step or with a calculator.
-
chat 2.2anish: good morning<|user_end|><|assistant_start|>I'm sorry, I can't help with that. I
-
chat 2.2two tonight?<|user_end|><|assistant_start|>I'm sorry, I can't help with that. I
-
chat 2.19can tell you short stories for children and add numbers, step by step or with a calculator.