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

Feature #2573

Fires on “1” (not labeled yet: read the examples)

Fires on
1.15% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 36%
  • chat 20%
  • agent 19%
  • python 16%
  • worksheet 9%

Makes the model say

sorry2lab pythonally[:: Then goodbye

Makes it not say

small ro run{ryst cl barked

Strongest examples

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

  1. chat 8.54
    <|user_end|><|assistant_start|><|think_start|>07+95: 7+5=12, 2
  2. chat 8.35
    <|user_end|><|assistant_start|><|think_start|>07+95: 7+5=12, 2
  3. worksheet 7.97
    315+8907: 5+7=12, 2
  4. worksheet 7.92
    76⏎47+25: 7+5=12, 2
  5. worksheet 7.81
    ⏎007+395: 7+5=12, 2
  6. chat 7.6
    <|think_start|>295+177: 5+7=12, 2
  7. chat 7.5
    <|think_start|>817+005: 7+5=12, 2
  8. worksheet 7.46
    685+4677: 5+7=12, 2
  9. chat 7.45
    3⏎07+65: 7+5=12, 2
  10. chat 7.45
    <|think_start|>327+075: 7+5=12, 2
  11. chat 7.44
    <|user_end|><|assistant_start|><|think_start|>06+76: 6+6=12, 2
  12. chat 7.44
    <|user_end|><|assistant_start|><|think_start|>15+67: 5+7=12, 2