Skip to content
mini-lab
Features of prelude-2

Feature #950

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

Fires on
0.39% of tokens
Mostly in
agent
Layer
after block 4

Where it fires

  • agent 66%
  • story 22%
  • chat 6%
  • python 5%
  • worksheet 1%

Makes the model say

s andant hat bestn here c

Makes it not say

oldString join trained sum multiplke element checks

Strongest examples

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

  1. chat 2.6
    57.<|assistant_end|><|user_start|>What if you add 8
  2. chat 2.6
    answer is 7.<|assistant_end|><|user_start|>What if you add 9
  3. chat 2.55
    145.<|assistant_end|><|user_start|>What if you add 6
  4. chat 2.38
    0?<|user_end|><|assistant_start|>The answer is 61.<|assistant_end|><|user_start|>What if you add 2
  5. chat 2.34
    7.<|user_end|><|assistant_start|>The answer is 91.<|assistant_end|><|user_start|>What if you add 8
  6. chat 2.31
    5?<|user_end|><|assistant_start|>The answer is 76.<|assistant_end|><|user_start|>What if you add 4
  7. story 2.25
    take things that are not ours," Ben said. "The apples belong to someone else."
  8. chat 2.17
    more?<|user_end|><|assistant_start|>The answer is 24.<|assistant_end|><|user_start|>What if you add 6
  9. chat 2.17
    .<|user_end|><|assistant_start|>The answer is 132.<|assistant_end|><|user_start|>What if you add 6
  10. chat 2.17
    29<|tool_end|><|assistant_start|>The answer is 29.<|assistant_end|><|user_start|>What if you add 1
  11. chat 2.14
    7.<|user_end|><|assistant_start|>The answer is 13.<|assistant_end|><|user_start|>What if you add 7
  12. chat 2.11
    ?<|user_end|><|assistant_start|>The answer is 172.<|assistant_end|><|user_start|>What if you add 8