Skip to content
mini-lab
Features of prelude-2

Feature #2700

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

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

Where it fires

  • worksheet 32%
  • chat 23%
  • agent 19%
  • python 19%
  • story 6%

Makes the model say

1commandcase0 Falsebed]`. functions

Makes it not say

and= from his of_!math

Strongest examples

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

  1. worksheet 4.4
    ⏎5+5: 5+5=10, 1
  2. chat 4.34
    5: 5+5=10, 1
  3. chat 4.11
    <|user_end|><|assistant_start|><|think_start|>05+45: 5+5=10, 0
  4. worksheet 3.97
    9286⏎5+5: 5+5=10, 1
  5. worksheet 3.88
    801⏎5+5: 5+5=10, 1
  6. chat 3.84
    4⏎05+45: 5+5=10, 0
  7. worksheet 3.84
    ⏎035+515: 5+5=10, 0
  8. worksheet 3.83
    965+4685: 5+5=10, 0
  9. worksheet 3.78
    65+00585: 5+5=10, 0
  10. worksheet 3.77
    755+5295: 5+5=10, 0
  11. chat 3.73
    <|think_start|>055+715: 5+5=10, 0
  12. worksheet 3.71
    5, 5⏎5+5: 5+5=10, 1