Skip to content
mini-lab
Features of prelude-2

Feature #3974

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

Fires on
0.40% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 58%
  • python 20%
  • agent 18%
  • chat 3%
  • worksheet 1%

Makes the model say

had have asked someone sharedn()`O

Makes it not say

_ doespythonassert uponcount little That

Strongest examples

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

  1. story 7.09
    .⏎"Poor Sara," he said. "You should have listened to me.
  2. story 6.66
    also knew she was wrong. He said, "Anna, you should have listened to me.
  3. story 6.39
    lot of money to fix it. She is very upset. You should have listened to me.
  4. story 6.12
    . He should have left earlier
  5. story 6.06
    ian. She said sorry to Spot. She was ashamed. She should have looked before she ran
  6. chat 5.97
    the fish stopped moving. Tim was sad. He learned that he should have listened to his mom
  7. story 5.96
    ⏎"I'm sorry, Ben. I was careless. I should have listened to you."
  8. chat 5.88
    the fish stopped moving. Tim was sad. He learned that he should have listened to his mom
  9. story 5.84
    sad because his toy was broken and he was hurt. He wished he had listened to his mom
  10. story 5.64
    blocks for his stack. Sam felt angry and lonely. He wished he had a friend to play
  11. story 5.41
    Tom and Ann cry. They are scared and sad. They wish they had not gone up the
  12. story 5.28
    fell to the ground. He was sad and scared. He wished he had listened to his mom