Skip to content
mini-lab
Features of prelude-2

Feature #1945

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

Fires on
0.66% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 47%
  • agent 44%
  • chat 7%
  • worksheet 1%
  • python 1%

Makes the model say

's is was were areisice e

Makes it not say

defined *ateultipl<|tool_call_start|>itedAssertionError ^^^^^^^^^^^^^^^^^^^^^^^^

Strongest examples

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

  1. chat 7.03
    etween their hands. They laughed and laughed.⏎"That was fun!" Joe
  2. chat 6.75
    laughed and laughed.⏎"That was fun!" Joe
  3. story 6.18
    hug. They feel better and happier.⏎Mom says, "That's good. I
  4. story 6.12
    , "I don't have a cooler like yours."⏎"That's okay", Bob
  5. chat 6.07
    , the bird, and the eggs. Dad said, "Wow, that's fun!" They
  6. story 5.98
    song, come and sing along!"⏎Ben says, "Wow, that's amazing. You
  7. chat 5.92
    otten to look where she was going.⏎"Well, that was a little bit
  8. story 5.77
    praying," Ben said.⏎The man nodded and said, "That is good. P
  9. story 5.73
    ek, and I'll bring some of my marbles."⏎"That sounds great," Jack
  10. story 5.72
    of money," Ben says. He is proud.⏎"Wow, that is amazing. Do
  11. story 5.65
    with your toys at your house."⏎Tom agreed and said, "That's a great idea
  12. story 5.58
    . She laughs. It is fun.⏎"Wow, Anna, that looks fun!" Ben