Feature #300
Fires on “)” (not labeled yet: read the examples)
- Fires on
- 0.38% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 35%
- agent 34%
- story 22%
- worksheet 6%
- chat 4%
Makes the model say
Fixedai moral"))"`.ased neverpath
Makes it not say
some called saw see or something call grow
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 1.83from arith import positive, smaller, product⏎⏎assert positive(-4) == False⏎assert
-
python 1.8import abs_value, divides⏎⏎assert abs_value(-4) == 4⏎
-
python 1.79_up_to, divides⏎⏎assert abs_value(-6) == 6⏎
-
python 1.73, is_odd, halve⏎⏎assert is_positive(-6) == False⏎assert
-
python 1.67, odd⏎⏎assert evens([8, 5, 7]) == [8]
-
python 1.65"⏎assert add_all([6, 2, 5]) == 13
-
python 1.64== 1⏎assert lower_of(2, 11) == 2⏎
-
python 1.62⏎⏎assert mean_list([4, 0, 9]) == 4.
-
python 1.62) == False⏎assert lower_of(6, 11) == 6⏎
-
python 1.610) == 55⏎assert sum_to(11) == 66
-
python 1.59⏎⏎assert mean_list([7, 8, 4]) ==
-
python 1.57") == "r"⏎assert bigger(2, 10) == 10