Feature #388
Fires on “to” (not labeled yet: read the examples)
- Fires on
- 1.33% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 67%
- python 28%
- story 4%
- chat 1%
- worksheet 0%
Makes the model say
illpositiveheadletterlespathk(
Makes it not say
_One:txt Tom reallyie divisible
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 6.59count_vowels(s):⏎
-
agent 5.838: ⏎9: def raise_to(a, b
-
agent 5.795: def first_letter(s):⏎
-
python 5.61⏎from arith import sum_up_to, lower_of
-
agent 5.56, lines 1-2⏎1: def raise_to(a, b
-
agent 5.555: ⏎6: ⏎7: def raise_to(a, b
-
agent 5.46from text import final_letter, size⏎2
-
agent 5.39⏎1: def num_chars(s):⏎
-
python 5.27!"⏎⏎`num_chars(s)` returns
-
python 5.17import first_letter⏎⏎LIMIT =
-
agent 5.17function raise_to that returns a to
-
python 5.16x⏎ return x⏎⏎⏎def sum_up_to(n):⏎