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

Feature #3153

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

Fires on
0.16% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 58%
  • agent 21%
  • story 13%
  • worksheet 7%
  • python 1%

Makes the model say

_ plusSet()`dy:` numsconfig

Makes it not say

⏎e each excited They down happy im

Strongest examples

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

  1. chat 5.69
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  2. chat 5.68
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  3. chat 5.64
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  4. chat 5.56
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  5. chat 5.53
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  6. chat 5.38
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  7. chat 5.32
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  8. chat 5.28
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  9. chat 5.26
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  10. chat 5.26
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  11. chat 5.26
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  12. chat 5.23
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories