Skip to content
mini-lab
Features of prelude-2

Feature #2890

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

Fires on
0.68% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 66%
  • chat 21%
  • python 8%
  • agent 4%
  • story 2%

Makes the model say

to a herselfAnna other learned be cat

Makes it not say

idegememberaktestaei sure

Strongest examples

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

  1. worksheet 3.49
    65+00: 5+0=5, 5547
  2. worksheet 3.47
    +00400: 9+0=9, 9⏎39
  3. python 3.41
    <|assistant_end|><|bos|>```python⏎def magnitude(x):⏎ if x < 0:⏎ return
  4. chat 3.4
    58+00: 8+0=8, 84⏎5
  5. worksheet 3.39
    +03108: 0+8=8, 8⏎58
  6. worksheet 3.33
    +00772: 5+2=7, 7⏎77
  7. worksheet 3.27
    00+45: 0+5=5, 526⏎
  8. worksheet 3.23
    +51801: 5+1=6, 6⏎00
  9. agent 3.21
    <|arg|>content=def abs_value(x):⏎ if x < 0:⏎ return
  10. chat 3.21
    12+00: 2+0=2, 29⏎1
  11. worksheet 3.2
    5+7561: 5+1=6, 6⏎37
  12. python 3.14
    .py⏎def abs_value(x):⏎ if x < 0:⏎ return