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

Feature #3221

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

Fires on
0.50% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 43%
  • story 38%
  • agent 11%
  • python 5%
  • worksheet 3%

Makes the model say

I for about he last He they if

Makes it not say

$⏎yingCreateid'mThank years

Strongest examples

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

  1. chat 4.04
    <|bos|><|user_start|>Who won the last World Cup?<|user_end|><|assistant_start|>Sorry, I'm a
  2. chat 3.93
    <|user_start|>How tall is the Eiffel Tower?<|user_end|><|assistant_start|>Sorry, I'm a tiny
  3. chat 3.91
    <|bos|><|user_start|>Who wrote Romeo and Juliet?<|user_end|><|assistant_start|>Sorry, I don't
  4. chat 3.9
    's the weather like in Berlin today?<|user_end|><|assistant_start|>Sorry, I don't
  5. chat 3.87
    <|user_start|>what time is it in New York<|user_end|><|assistant_start|>Sorry, I'm a
  6. agent 3.86
    won the World Cup?<|user_end|><|assistant_start|>Sorry, I can't do
  7. chat 3.85
    `.<|assistant_end|><|bos|><|user_start|>What does DNA stand for?<|user_end|><|assistant_start|>Sorry, I'm a tiny
  8. chat 3.83
    anslate this into Spanish: good morning.<|user_end|><|assistant_start|>Sorry, I'm a
  9. chat 3.82
    me the latest sports scores.<|user_end|><|assistant_start|>Sorry, I don't
  10. chat 3.82
    <|user_start|>Can you help me with my taxes?<|user_end|><|assistant_start|>Sorry, I'm a tiny
  11. chat 3.79
    the telephone?<|user_end|><|assistant_start|>Sorry, I'm a tiny
  12. chat 3.74
    ever!"<|assistant_end|><|bos|><|user_start|>What should I invest in?<|user_end|><|assistant_start|>Sorry, I'm a