Skip to content
mini-lab
Features of prelude-2

Feature #1689

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

Fires on
0.13% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 56%
  • chat 33%
  • story 9%
  • python 2%
  • worksheet 0%

Makes the model say

tell_ forgot follow alsoreetend keep

Makes it not say

87 looked3 test1average hair

Strongest examples

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

  1. chat 3.52
    Start every answer with "Sure!".<|system_end|><|user_start|>hey there<|user_end|><|assistant_start|>Sure! Hello!
  2. chat 3.43
    <|bos|><|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hi there
  3. chat 3.42
    <|bos|><|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hi there
  4. chat 3.41
    <|bos|><|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hi there
  5. chat 3.29
    <|bos|><|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hello!
  6. chat 3.24
    <|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>Hi!<|user_end|><|assistant_start|>Sure! Hi there
  7. chat 3.18
    <|bos|><|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hi there
  8. chat 3.16
    Start every answer with "Sure!".<|system_end|><|user_start|>Goodbye!<|user_end|><|assistant_start|>Sure! Bye!
  9. chat 3.15
    Goodbye!<|user_end|><|assistant_start|>Sure! Bye!
  10. chat 3.11
    <|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>Hi!<|user_end|><|assistant_start|>Sure! Hi there
  11. chat 3.09
    <|bos|><|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hello!
  12. chat 3.05
    <|bos|><|system_start|>Start every answer with "Sure!".<|system_end|><|user_start|>Hi<|user_end|><|assistant_start|>Sure! Hi there