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

Feature #2824

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

Fires on
0.24% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 49%
  • chat 28%
  • worksheet 21%
  • agent 2%
  • python 0%

Makes the model say

have care wanteds laughed loved laugh did

Makes it not say

⏎ functionsdefRenamed:`://Set enough

Strongest examples

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

  1. chat 3.33
    toys. Sam gets 40 more toys. How many toys does Sam have now?<|user_end|>
  2. worksheet 3.32
    1 apples. Sam gets 7 more apples. How many apples does Sam have now?⏎
  3. worksheet 3.31
    gets 72522 more shells. How many shells does Tim have now?⏎
  4. chat 3.31
    cookies. Lily gets 12 more cookies. How many cookies does Lily have now?<|user_end|>
  5. worksheet 3.31
    7 toys. Sam gets 4 more toys. How many toys does Sam have now?⏎
  6. worksheet 3.3
    cookies. Tom gets 33 more cookies. How many cookies does Tom have now?⏎
  7. chat 3.3
    . Tom gets 122 more stars. How many stars does Tom have now?<|user_end|>
  8. chat 3.26
    3 cookies. Sue gets 4 more cookies. How many cookies does Sue have now?<|user_end|>
  9. chat 3.26
    2 balls. Tim gets 8 more balls. How many balls does Tim have now?<|user_end|>
  10. chat 3.26
    6 shells. Tom gets 9 more shells. How many shells does Tom have now?<|user_end|>
  11. worksheet 3.25
    Lily gets 1517 more cookies. How many cookies does Lily have now?⏎
  12. worksheet 3.24
    Sam gets 4212 more balls. How many balls does Sam have now?⏎