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

Feature #857

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

Fires on
0.36% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 41%
  • agent 29%
  • story 17%
  • chat 13%
  • python 0%

Makes the model say

had has was liveddemoO loved have

Makes it not say

your youen up girl allLIMIT She

Strongest examples

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

  1. worksheet 7.47
    48698 = 49516.⏎Sam had 63
  2. worksheet 7.42
    4 + 69 = 36013.⏎Sam has 25
  3. worksheet 7.4
    58681 = 58687.⏎Sam has 16
  4. worksheet 7.32
    + 4312 = 4396.⏎Sam had 7 stickers
  5. worksheet 7.01
    74244 = 74323.⏎Sam had 2 cookies
  6. worksheet 6.98
    and 50, you get 5244.⏎Sam has 8 coins
  7. worksheet 6.97
    4 + 12 = 80126.⏎Ben has 32
  8. worksheet 6.92
    8544 = 16363.⏎Sam has 9 apples
  9. worksheet 6.76
    9 + 640 = 649.⏎Mia has 85
  10. worksheet 6.74
    So 59 + 10 = 69.⏎Sam had 0 coins
  11. worksheet 6.72
    14 + 96 = 6610.⏎Anna had 75
  12. worksheet 6.71
    47149 = 47629.⏎Ben had 13