Feature #838
Fires on “moon” (not labeled yet: read the examples)
- Fires on
- 0.53% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 74%
- python 21%
- chat 5%
- story 1%
- worksheet 0%
Makes the model say
!")`, typoreplaceAll defined<|tool_end|>riteDon
Makes it not say
of, all b sc1! thought
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 3.63first_letter⏎⏎assert loud("moon") == "M
-
python 3.54, 1, 3]))⏎print(count_vowels("lemon"))⏎⏎$
-
python 3.49(count_vowels(WORD))⏎print(final_letter("tree"))⏎⏎#
-
python 3.49⏎print(excite("python"))⏎print(count_vowels("robot"))⏎⏎#
-
python 3.46k + 1⏎ return k⏎```⏎⏎>>> vowels("lemon")⏎2⏎
-
python 3.33number of vowels in s. For example, `count_vowels("moon")` returns `2
-
python 3.2returns the length of s. For example, `num_chars("moon")` returns `4
-
python 3.13.py⏎from text import backwards, welcome⏎⏎assert backwards("banana") == "anan
-
python 3.12last_letter("apple") == "e"⏎assert vowels("banana") == 3
-
python 3.11greetings import count_vowels, hello⏎⏎assert count_vowels("house") == 3
-
python 3.11words import count_vowels, shout⏎⏎assert count_vowels("hello") == 2
-
python 3.08flip("banana")` returns `"anan