Skip to content
mini-lab
Features of prelude-2

Feature #2397

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

Fires on
0.44% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 46%
  • story 39%
  • python 7%
  • chat 7%
  • worksheet 1%

Makes the model say

bmak likeedot whisper ear

Makes it not say

bestreturn waynewStringgroundone - asking

Strongest examples

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

  1. story 4.62
    closer.⏎Finally, the bird flew away but Emma was determined to find it
  2. story 4.58
    school, even though it was very far away. He was determined to make it
  3. story 4.56
    he ran after them as fast as he could. He was determined to catch them
  4. story 4.5
    ered the jungle and he looked very scared. The elephant was determined to help the
  5. story 4.37
    the ground. She wasn't hurt, but the tiger was determined to climb the
  6. story 4.36
    The genie smiled and said, "Your wish is my command." Then,
  7. story 4.26
    then you can have the necklace."⏎June was determined to get the
  8. story 4.23
    . The sky was grey and dark. But Sally was determined to enjoy it
  9. chat 4.15
    her kind words. She stood up and walked carefully away, determined to never be
  10. chat 3.83
    , and they made many yummy meals together.<|bos|><|user_start|>Summarize this art
  11. story 3.75
    , steady, go!"⏎The rabbit hopped off at tremendous speed while
  12. story 3.74
    And when the sun went down, they all went home with happy memories of the shining