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

Feature #3825

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

Fires on
0.26% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 55%
  • agent 17%
  • worksheet 14%
  • story 13%
  • python 1%

Makes the model say

to= p Every dad add list

Makes it not say

brave functions himself own rideationnitude

Strongest examples

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

  1. chat 3.77
    I'm sorry, I can't help with that. I only know how to tell stories and
  2. chat 3.77
    I'm sorry, I can't help with that. I only know how to tell stories and
  3. chat 3.71
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  4. chat 3.66
    I'm sorry, I can't help with that. I only know how to tell stories and
  5. chat 3.65
    I'm sorry, I can't help with that. I only know how to tell stories and
  6. chat 3.65
    I'm sorry, I can't help with that. I only know how to tell stories and
  7. chat 3.63
    I'm sorry, I can't help with that. I only know how to tell stories and
  8. chat 3.62
    can't help with that. I only know how to tell stories and
  9. chat 3.58
    I'm sorry, I can't help with that. I only know how to tell stories and
  10. chat 3.57
    I'm sorry, I can't help with that. I only know how to tell stories and
  11. chat 3.56
    I'm sorry, I can't help with that. I only know how to tell stories and
  12. chat 3.55
    I'm sorry, I can't help with that. I only know how to tell stories and