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

Feature #2864

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

Fires on
0.59% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 55%
  • worksheet 25%
  • story 12%
  • agent 8%
  • python 1%

Makes the model say

do liked say when doing did name

Makes it not say

appheditemSetalse checkDon max

Strongest examples

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

  1. story 3.1
    and didn't know what to do.⏎Do
  2. chat 2.89
    big, smelly waste. They were sad and didn't know what to do.⏎M
  3. chat 2.81
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  4. worksheet 2.8
    9 + 17 = 24066⏎What's 587
  5. chat 2.76
    for a story!<|assistant_end|><|user_start|>What's your name?<|user_end|><|assistant_start|>
  6. worksheet 2.75
    9 equals 37.⏎What is 507 + 7?⏎
  7. chat 2.73
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  8. chat 2.71
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  9. chat 2.7
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  10. worksheet 2.7
    = 560.⏎What is 1037 plus 280
  11. chat 2.7
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories
  12. chat 2.67
    <|assistant_start|>I'm sorry, I can't help with that. I only know how to tell stories