Feature #3
Fires on “ but” (not labeled yet: read the examples)
- Fires on
- 0.35% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 53%
- agent 36%
- story 8%
- chat 2%
- worksheet 0%
Makes the model say
python his<|user_end|>schema x<|tool_end|>```is
Makes it not say
askedapp hiding voicePlease stuckem cried
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 2.73⏎Bug: `vowels` should count the vowels in s, but it has `k
-
python 2.59` should return the average of a and b, but it has `return
-
python 2.55Bug: `total` should add up all the numbers in nums, but it has `result
-
python 2.54Bug: `mean_list` should return the average of nums, but it has `return
-
python 2.46sum_list` should add up all the numbers in nums, but it has `result
-
python 2.44: `max_item` should return the biggest item of nums, but it has `return
-
python 2.4<|assistant_end|><|bos|>Bug: `vowels` should count the vowels in s, but it has `if
-
python 2.39Bug: `mean_list` should return the average of nums, but it has `return
-
python 2.38⏎Bug: `smallest` should return the smallest number in nums, but it has `return
-
python 2.38: `max_item` should return the biggest item of nums, but it has `return
-
python 2.36Bug: `count_items` should return the number of items, but it has `return
-
python 2.35Bug: `mean` should return the average of a and b, but it has `return