Skip to content
mini-lab
Features of prelude-2

Feature #2202

Fires on “7” (not labeled yet: read the examples)

Fires on
0.18% of tokens
Mostly in
worksheet
Layer
after block 4

Where it fires

  • worksheet 56%
  • chat 34%
  • story 6%
  • agent 4%
  • python 0%

Makes the model say

NowleOne moral Lily return only inside

Makes it not say

x mraise): wind")ike p

Strongest examples

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

  1. worksheet 2.0
    8: 9+8+1=18, 184400
  2. chat 1.99
    8: 8+8+1=17, 1706<|think_end|>The
  3. chat 1.97
    2+8: 2+8=10, 10<|think_end|>The answer is
  4. chat 1.97
    8+9: 8+9=17, 17<|think_end|>The answer is
  5. worksheet 1.96
    9+8: 9+8=17, 17580⏎
  6. chat 1.95
    8+9: 8+9=17, 17<|think_end|>The answer is
  7. worksheet 1.94
    0: 9+0+1=10, 101500
  8. worksheet 1.93
    2+8: 2+8=10, 10649⏎
  9. chat 1.93
    8+9: 8+9=17, 175<|think_end|>17
  10. chat 1.93
    9+9: 9+9=18, 18<|think_end|>The answer is
  11. chat 1.92
    +8: 8+8=16, 167<|think_end|>Sure!
  12. chat 1.92
    6: 9+6+1=16, 1691<|think_end|>The