Skip to content
mini-lab
Features of prelude-2

Feature #2570

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

Fires on
2.46% of tokens
Mostly in
story
Layer
after block 4

Where it fires

  • story 50%
  • agent 34%
  • chat 14%
  • python 1%
  • worksheet 1%

Makes the model say

was were went?<|assistant_end|>$: appeared

Makes it not say

back in of into forO clean just

Strongest examples

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

  1. story 8.25
    , there was a fun race every year. All the people in the village liked the race.
  2. story 7.2
    the lake was even tidier than before! Everyone who visited the lake was impressed with
  3. story 6.83
    , Tom played the guitar and sang songs. All the animals in the forest loved to listen to
  4. story 6.69
    flew high and low, fast and slow. All the animals in the forest came to watch and
  5. story 6.63
    friends. Everyone in the forest was even happier
  6. story 6.48
    iday.⏎On the holiday, everyone in the town came together. They
  7. agent 6.37
    a new function doubled in stats.py that returns every number of nums doubled."<|user_end|><|assistant_start|><|tool_call_start|>
  8. story 6.14
    It grew taller and taller. All the other trees in the forest were happy for their
  9. story 6.07
    kind to everyone. All the kids at school liked her. Lisa
  10. story 6.01
    the lion and save the little mouse. All the other animals in the jungle cheered for the brave
  11. agent 6.0
    function double_all to items.py that returns every number of nums doubled."<|user_end|><|assistant_start|><|tool_call_start|>
  12. agent 5.99
    a new function doubled in stats.py that returns every number of nums doubled."<|user_end|><|assistant_start|><|tool_call_start|>