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

Feature #2073

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

Fires on
1.04% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 45%
  • agent 42%
  • chat 9%
  • python 3%
  • worksheet 0%

Makes the model say

the their a trouble need winf her

Makes it not say

ing They larger off", "!"`. swam idea

Strongest examples

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

  1. story 6.44
    was red and blue. It was in a park where kids
  2. story 6.36
    before. That's when she realized that something was wrong - she was in a very serious sit
  3. agent 6.14
    AI agent running in OpenCode, a coding
  4. agent 6.14
    AI agent running in OpenCode, a coding
  5. agent 6.14
    AI agent running in OpenCode, a coding
  6. agent 6.14
    AI agent running in OpenCode, a coding
  7. story 6.07
    grew too tired to carry on.⏎Knowing he was in trouble, he kept
  8. story 5.95
    <|bos|>There were two best friends, Mo and Jane. They were in a large classroom
  9. story 5.94
    beautiful rainbow in the sky.⏎
  10. story 5.89
    to pick up leaves and scatter them around. Suddenly, she was interrupt
  11. chat 5.89
    Write a story for me.<|user_end|><|assistant_start|>Sure! Mama and Jack were in the kitchen. Jack
  12. story 5.76
    was being planned in his honour.⏎The little boy was in awe and