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

Feature #235

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

Fires on
0.72% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 86%
  • story 7%
  • python 4%
  • chat 2%
  • worksheet 1%

Makes the model say

ries subtract "?"les "" insteadaterled

Makes it not say

.: little ` their! + upon

Strongest examples

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

  1. agent 3.27
    Fixed `initial` in `strings.py`: it used `return s[-1]` instead
  2. agent 3.25
    Fixed `initial` in `strings.py`: it used `return s[1]` instead
  3. agent 3.22
    Fixed `shout` in `utils.py`: it used `return s` instead of `
  4. agent 3.14
    Fixed `repeat` in `text.py`: it used `return s * 2`
  5. agent 3.13
    Fixed `reverse` in `lib.py`: it used `return s[::1]` instead
  6. agent 3.08
    Fixed `backwards` in `strings.py`: it used `return s[1:]`
  7. agent 3.03
    `echo` in `text.py`: it used `return s + s` instead of
  8. agent 2.98
    Fixed `echo` in `tools.py`: it used `return s * 2`
  9. agent 2.96
    first_letter` in `strings.py`: it used `return s[1]` instead
  10. agent 2.96
    Fixed `excite` in `strings.py`: it used `return s + "?"` instead
  11. agent 2.96
    double_all` in `listlib.py`: it used `return [n * 3
  12. agent 2.93
    Fixed `initial` in `textlib.py`: it used `return s[1]` instead