Skip to content
mini-lab
Features of prelude-2

Feature #354

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

Fires on
0.55% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 36%
  • python 27%
  • story 23%
  • chat 10%
  • worksheet 4%

Makes the model say

- aboutledleightulstarlower

Makes it not say

4Run. Truecom Ben big5

Strongest examples

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

  1. story 3.02
    , Lila said goodbye to the thief. She said, "Thanks for being my
  2. story 2.85
    bird free.⏎When the bird was free, she said, "Thank you, Fred and
  3. story 2.85
    . He said, "Thank you, little bird
  4. story 2.63
    proud of the meal he cooked. His friends said, "Thank you, Tim.
  5. story 2.57
    helped the goat out of the hole. The goat said, "Thank you, Billy and
  6. story 2.56
    truck.⏎When the truck was full, the man said, "Thank you, Tim and
  7. python 2.52
    n):⏎ return s * n⏎```⏎⏎>>> echo("banana", 2)
  8. chat 2.48
    and very glad. The bird said, "Thank you, Rex!
  9. story 2.47
    It grew into a big, talking flower! The flower said, "Thank you for helping me
  10. chat 2.46
    big smile. Sam smiled back and said, "Thank you, Tom!"
  11. python 2.45
    n):⏎ return s * n⏎```⏎⏎>>> echo("house", 2)
  12. python 2.44
    .py⏎from textlib import echo, length⏎⏎assert echo("cloud", 3)