Feature #2827
Fires on “:” (not labeled yet: read the examples)
- Fires on
- 0.76% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 84%
- story 10%
- chat 5%
- worksheet 1%
- python 0%
Makes the model say
they They" I he⏎>>> He
Makes it not say
nums reverventekMIT edit+ Instead
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 4.77a turtle swimming towards them.⏎The turtle said: "Don't you
-
story 4.47fun.⏎But then, Sara feels tired. She says to Tom: "I want to
-
agent 4.35aan(12, 2))⏎ ^^^^^⏎NameError: name 'mea
-
agent 4.23print(posiitive(COUNT))⏎ ^^^^^^^^^⏎NameError: name 'pos
-
agent 4.21be `positive`. It now prints:⏎```⏎True
-
agent 4.18be `reverse`. It now prints:⏎```⏎r
-
agent 4.18print(keep_veen(NUMBERS))⏎ ^^^^^^^^^⏎NameError: name 'keep_
-
agent 4.16tiply(6, 1))⏎ ^^^^^^^^^⏎NameError: name 'mull
-
agent 4.16, 8, 8, 0]))⏎ ^^^^^^^^⏎NameError: name 'mean_
-
story 4.1go there anymore!"⏎Lily was very sad. She said to Jack: "Being popular
-
agent 4.097, 4, 3]))⏎ ^^^^^⏎NameError: name 'ta
-
agent 4.09⏎ print(niitial(WORD))⏎ ^^^^^^^⏎NameError: name 'ni