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

Feature #3512

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

Fires on
0.66% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 41%
  • chat 22%
  • story 18%
  • worksheet 10%
  • python 9%

Makes the model say

1 wanted lovedateappyanan([ functions

Makes it not say

TheyThey where who!") max pushed are

Strongest examples

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

  1. chat 2.55
    - b⏎```⏎<|bos|><|user_start|>Sue has 3 cookies. Sue gets 4 more cookies
  2. chat 2.41
    equals 93.<|bos|><|user_start|>Anna has 3 apples. Anna gets 8 more apples
  3. chat 2.38
    the number only.<|system_end|><|user_start|>Sue has 55 coins. Sue gets 303
  4. chat 2.38
    5985.<|bos|><|user_start|>Anna has 3 apples. Anna gets 8 more apples
  5. chat 2.37
    in the whole forest.<|bos|><|user_start|>Anna has 3 apples. Anna gets 8 more apples
  6. chat 2.34
    .py`.<|assistant_end|><|bos|><|user_start|>Mia has 54 apples. Mia gets 4 more apples
  7. chat 2.34
    the camp.<|assistant_end|><|bos|><|user_start|>Sue has 42 balls. Sue gets 18 more
  8. chat 2.33
    the number only.<|system_end|><|user_start|>Anna has 41 toys. Anna gets 37 more
  9. chat 2.33
    pass now.<|assistant_end|><|bos|><|user_start|>Lily has 71 coins. Lily gets 8 more coins
  10. chat 2.32
    618.<|bos|><|user_start|>Lucy has 18 coins. Lucy gets 795
  11. chat 2.32
    at the park.<|bos|><|user_start|>Tim has 19 cookies. Tim gets 76 more
  12. chat 2.32
    equals 57.<|bos|><|user_start|>Lucy has 1 coins. Lucy gets