Skip to content
mini-lab
Log in Sign up
Features of prelude-2

Feature #3201

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

Fires on
0.61% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 74%
  • story 20%
  • chat 3%
  • python 2%
  • worksheet 0%

Makes the model say

even % sky like love little b on

Makes it not say

<|think_start|>40ant]`.5itient

Strongest examples

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

  1. story 3.3
    you can make lots of
  2. agent 3.3
    you?<|user_end|><|assistant_start|>I
  3. chat 3.3
    you tell me what
  4. worksheet 3.3
    you get 12
  5. story 3.3
    you help me?" Mia
  6. agent 3.3
    you run run.py
  7. story 3.3
    you too, Tom.
  8. chat 3.3
    you tell me a story
  9. story 3.3
    you get some blueberries
  10. chat 3.3
    you today?<|assistant_end|><|user_start|>
  11. story 3.3
    you Fluffy. Do you
  12. story 2.78
    clean burrow and thanked his neighbour once again.⏎From then on, the messy