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

Feature #2577

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

Fires on
0.73% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 64%
  • chat 24%
  • agent 6%
  • story 4%
  • python 1%

Makes the model say

U hardFix Truew inderice

Makes it not say

b returns-return Lily Ities return

Strongest examples

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

  1. story 2.59
    so proud that she was
  2. story 2.59
    so much fun! Then
  3. story 2.59
    so it flew fast and
  4. chat 2.59
    so yummy! Bob wanted
  5. story 2.59
    so he just put it
  6. story 2.2
    very excited.⏎As
  7. chat 2.2
    very happy. She bought
  8. story 2.2
    very happy. She sat
  9. worksheet 1.79
    99: 0+9=9, 99⏎0
  10. python 1.76
    ) == 6⏎assert sub_one(1) == 0⏎print("
  11. chat 1.71
    74+295: 4+5=9, 9⏎07
  12. worksheet 1.71
    9: 0+9=9, 9891