Skip to content
mini-lab
Features of prelude-2

Feature #312

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

Fires on
0.99% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 73%
  • agent 14%
  • chat 10%
  • python 2%
  • worksheet 1%

Makes the model say

were are became smiledk went met talking

Makes it not say

andhing It", plus,)) There

Strongest examples

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

  1. story 6.05
    37 = 91540<|bos|>Tom and Mia were best friends.
  2. story 5.86
    wind blew hard. Ollie and Timmy got scared.⏎
  3. story 5.51
    (nums)` returns every number of nums doubled.<|assistant_end|><|bos|>Tom and Lily went to the beach
  4. story 5.45
    in `main.py`.<|assistant_end|><|bos|>Lily and Ben are playing in the
  5. story 5.43
    6, 607<|think_end|>607<|assistant_end|><|bos|>Ben and Lily liked to fish with
  6. story 5.41
    : I can tell short stories and add numbers.<|assistant_end|><|bos|>Tom and Mia were playing in the
  7. story 5.4
    collar through the river to clean it." Spot and Cat went to the river
  8. story 5.37
    The red bird and the big blue bird sat in their nest
  9. story 5.36
    to gather sticks and leaves to make a big pile. Max and Lily were having a lot
  10. story 5.28
    507 = 92725.<|bos|>Lily and Tom are friends. They
  11. chat 5.13
    a cat!<|user_end|><|assistant_start|>One day, a big cat and a little dog were playing in the
  12. story 5.09
    ⏎Sue and Tom picked a green apple