Feature #1402
Fires on “ ” (not labeled yet: read the examples)
- Fires on
- 0.37% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 79%
- agent 9%
- story 9%
- chat 2%
- worksheet 1%
Makes the model say
that the wagged anymoreThe wet The very
Makes it not say
]replaceAll twowayThey Theysel
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 4.14⏎>>> mean_list([5, 7, 4, 3])⏎4
-
python 4.06```⏎⏎>>> biggest([7, 9, 2, 7, 2
-
python 4.05⏎>>> min_item([3, 9, 9, 4])⏎3
-
python 4.01⏎assert double_all([6, 5, 9, 5]) == [
-
python 3.86⏎>>> mean_list([2, 4, 2, 1, 2
-
python 3.78⏎>>> average_list([0, 4, 6, 4])⏎3
-
python 3.76⏎assert keep_even([0, 5, 3, 4, 3
-
python 3.76For example, `last([0, 5, 2, 2, 0
-
python 3.73⏎>>> max_item([1, 7, 2, 3, 8
-
python 3.72odd⏎⏎assert last([8, 8, 8, 4, 6
-
python 3.72⏎>>> double_all([1, 2, 2, 8, 5
-
python 3.71⏎>>> has_item([1, 4, 5, 1], 4