Skip to content
mini-lab
Features of prelude-2

Feature #507

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

Fires on
0.60% of tokens
Mostly in
chat
Layer
after block 4

Where it fires

  • chat 50%
  • agent 24%
  • python 20%
  • worksheet 4%
  • story 2%

Makes the model say

('t list://commandgeduallyprint

Makes it not say

Thank Thank doorueot playing house find

Strongest examples

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

  1. chat 5.19
    <|bos|><|user_start|>Can you help me with my taxes?<|user_end|><|assistant_start|>Sorry, I'm a
  2. chat 5.01
    <|bos|><|user_start|>Who wrote Romeo and Juliet?<|user_end|><|assistant_start|>Sorry, I don't
  3. python 4.96
    ⏎TEXT = "robot"⏎print(loud(TEXT))⏎print(sum_up
  4. chat 4.94
    plain quantum physics to me.<|user_end|><|assistant_start|>Sorry, I'm a
  5. chat 4.92
    <|user_start|>How tall is the Eiffel Tower?<|user_end|><|assistant_start|>Sorry, I'm a
  6. chat 4.85
    <|bos|><|user_start|>Who won the last World Cup?<|user_end|><|assistant_start|>Sorry, I'm a
  7. python 4.85
    text import loud, initial⏎⏎print(initial("robot"))⏎print(loud("house
  8. chat 4.85
    Summarize this article for me.<|user_end|><|assistant_start|>Sorry, I'm a
  9. agent 4.82
    coding agent harness.<|system_end|><|user_start|>Write me a poem.<|user_end|><|assistant_start|>Sorry, I can't
  10. agent 4.75
    coding agent harness.<|system_end|><|user_start|>write me a poem<|user_end|><|assistant_start|>Sorry, I can't
  11. python 4.74
    helpers import subtract⏎⏎print(count_vowels("tree"))⏎print(subtract(6
  12. python 4.71
    LIMIT = 4⏎print(abs_value(LIMIT))⏎print(first_letter