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

Feature #3999

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

Fires on
0.29% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 52%
  • agent 28%
  • story 16%
  • python 3%
  • worksheet 1%

Makes the model say

ch likedpyg timeate mainine

Makes it not say

hispped])` its my he hole5

Strongest examples

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

  1. chat 4.79
    is the tallest mountain in the world<|user_end|><|assistant_start|>Sorry, I don't know about that.
  2. chat 4.62
    is 12 times 8?<|user_end|><|assistant_start|>Sorry, I don't know about that.
  3. chat 4.57
    do I learn French fast?<|user_end|><|assistant_start|>Sorry, I don't know about that.
  4. chat 4.56
    tory of Brazil.<|user_end|><|assistant_start|>Sorry, I don't know about that.
  5. chat 4.55
    Romeo and Juliet?<|user_end|><|assistant_start|>Sorry, I don't know about that.
  6. chat 4.52
    <|user_start|>How far away is the moon?<|user_end|><|assistant_start|>Sorry, I don't know about that.
  7. chat 4.5
    est sports scores.<|user_end|><|assistant_start|>Sorry, I don't know about that.
  8. chat 4.49
    capital of Brazil?<|user_end|><|assistant_start|>Sorry, I don't know about that.
  9. chat 4.48
    history of India<|user_end|><|assistant_start|>Sorry, I don't know about that. I
  10. chat 4.48
    is the meaning of life?<|user_end|><|assistant_start|>Sorry, I don't know about that. I
  11. chat 4.47
    physics to me.<|user_end|><|assistant_start|>Sorry, I don't know about that.
  12. chat 4.47
    est sports scores.<|user_end|><|assistant_start|>Sorry, I don't know about that.