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

Feature #3538

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

Fires on
0.40% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 62%
  • chat 26%
  • story 9%
  • python 2%
  • worksheet 0%

Makes the model say

ONSetk?"pwelcomeine x

Makes it not say

They they But⏎ Timupper were else

Strongest examples

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

  1. chat 2.7
    sics to me.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  2. chat 2.7
    ay some music for me.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  3. chat 2.68
    for two tonight?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  4. chat 2.68
    orts scores.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  5. chat 2.66
    o and Juliet?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  6. chat 2.62
    sics to me.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  7. chat 2.62
    sics to me.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  8. chat 2.61
    orts scores.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  9. chat 2.59
    sics to me.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  10. chat 2.57
    article for me.<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  11. chat 2.56
    o and Juliet?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can
  12. chat 2.55
    a picture of a cat?<|user_end|><|assistant_start|>Sorry, I don't know about that. I can