Skip to content
mini-lab
Log in Sign up
Features of prelude-2

Feature #4038

Fires on “ can” (not labeled yet: read the examples)

Fires on
0.18% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 58%
  • agent 22%
  • story 17%
  • worksheet 2%
  • python 1%

Makes the model say

sep'tp listpySE):

Makes it not say

? more<|assistant_end|> gets next example climbed forget

Strongest examples

On text the dictionary never trained on. The underlined token is where it fires most.

  1. chat 3.7
    cat?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  2. chat 3.59
    today?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  3. chat 3.57
    s.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  4. chat 3.56
    fast?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  5. chat 3.56
    me.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  6. chat 3.55
    iet?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  7. chat 3.54
    s.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  8. chat 3.54
    moon?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  9. chat 3.51
    moon?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  10. chat 3.5
    cat?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  11. chat 3.48
    in?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories
  12. chat 3.44
    moon?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can only tell short stories