Skip to content
mini-lab
Features of prelude-2

Feature #3933

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

Fires on
0.22% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 79%
  • chat 19%
  • story 2%
  • python 0%
  • worksheet 0%

Makes the model say

o face+lar countiely

Makes it not say

rolled years played project dayHi watchedWow

Strongest examples

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

  1. chat 4.01
    iet<|user_end|><|assistant_start|>Sorry, I don't know about
  2. chat 3.95
    How tall is the Eiffel Tower?<|user_end|><|assistant_start|>Sorry, I'm a tiny model
  3. chat 3.92
    Who wrote Romeo and Juliet?<|user_end|><|assistant_start|>Sorry, I don't know about
  4. chat 3.87
    is the capital of Brazil?<|user_end|><|assistant_start|>Sorry, I don't know about
  5. chat 3.84
    ed the telephone<|user_end|><|assistant_start|>Sorry, I don't know about
  6. agent 3.76
    won the World Cup?<|user_end|><|assistant_start|>Sorry, I can't do that
  7. chat 3.66
    antum physics to me.<|user_end|><|assistant_start|>Sorry, I'm a tiny model
  8. chat 3.63
    Who won the last World Cup?<|user_end|><|assistant_start|>Sorry, I'm a tiny model
  9. chat 3.56
    the telephone?<|user_end|><|assistant_start|>Sorry, I'm a tiny model
  10. chat 3.52
    space.<|user_end|><|assistant_start|>Sorry, I don't know about
  11. chat 3.47
    <|user_start|>How do I learn French fast?<|user_end|><|assistant_start|>Sorry, I don't know about
  12. chat 3.46
    <|system_end|><|user_start|>Can you write C++ code?<|user_end|><|assistant_start|>Sorry, I don't know about