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

Feature #381

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

Fires on
0.30% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 59%
  • story 27%
  • chat 12%
  • python 2%
  • worksheet 0%

Makes the model say

nameds list toldcommand there called very

Makes it not say

backlThankotlistlib end letter<|assistant_end|>

Strongest examples

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

  1. story 2.02
    , but he was too small for the car. So, he thought of a plan to drive
  2. story 1.95
    take a nap. But Sam still wanted to play.⏎Tom thought of
  3. story 1.93
    want to surrender to the big lake. So, he thought of a plan.⏎
  4. story 1.91
    bo did not want to play. Momo was not happy. He thought of a plan to make
  5. story 1.88
    cold and scared. Tim wanted to save the bird, so he thought of a plan.⏎
  6. story 1.86
    , but he wanted to get his ball. So, he thought of a plan.⏎
  7. story 1.82
    Jane wanted to make the room nice and bright.⏎Jane thought of a plan. She
  8. story 1.8
    sweet cake, but he had no money to buy it. Tom thought of a plan to get
  9. story 1.78
    sat by the fire, feeling cold and grumpy. The little girl thought of a way to help
  10. story 1.75
    and started playing with her friend.⏎Suddenly, Millie's mother came running up.
  11. story 1.74
    He did not want to fall in the crack. So, he thought of a plan.⏎
  12. story 1.73
    can't get it down." Max wanted to help her. He thought of a plan to replace