Feature #3807
Fires on “loud” (not labeled yet: read the examples)
- Fires on
- 0.88% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 64%
- chat 14%
- agent 12%
- story 9%
- worksheet 2%
Makes the model say
test⏎ +( named tests both long
Makes it not say
oldStringed waydancescript ways back
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 3.4`num_chars("moon")` returns `4`.⏎⏎`loud(s)` returns
-
python 3.3whisper("BANANA")⏎"banana"⏎⏎`excite(s)` returns
-
python 3.24([2, 7, 4])⏎2⏎⏎`contains(items, x
-
python 3.2112)` returns `6.5`.⏎⏎`times(a, b
-
python 3.18```⏎⏎`factorial` multiplies all
-
python 3.17, 7, 5])` returns `5`.⏎⏎`loud(s)` returns
-
python 3.17", 2)⏎"bananabanana"⏎⏎`vowels(s)` returns
-
python 3.12("PYTHON")⏎"python"⏎⏎`square(x)` returns
-
python 3.11case("code")` returns `"CODE"`.⏎⏎`includes(items, x
-
python 3.11`twice(8)` returns `16`.⏎⏎`times(a, b
-
python 3.09, 6, 4])⏎3.5⏎⏎`evens(nums)` returns
-
python 3.07, `negate(3)` returns `-3`.⏎⏎`doubled(nums)` returns