Feature #349
Fires on “(” (not labeled yet: read the examples)
- Fires on
- 0.41% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 54%
- python 28%
- story 10%
- worksheet 6%
- chat 2%
Makes the model say
were1Okayientate multipl + counts
Makes it not say
you!!". They?<|assistant_end|>am
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 2.67factorial(n):⏎ result = 1⏎ for i in range(1, n +
-
python 2.49fact(n):⏎ result = 1⏎ for i in range(1, n +
-
agent 2.45fact(n):⏎ result = 1⏎ for i in range(1, n +
-
python 2.41fact(n):⏎ result = 1⏎ for i in range(1, n +
-
python 2.39factorial(n):⏎ result = 1⏎ for i in range(1, n +
-
python 2.34fact(n):⏎ result = 1⏎ for i in range(1, n +
-
agent 2.28fact(n):⏎ result = 1⏎ for i in range(1, n +
-
python 2.25factorial(n):⏎ result = 1⏎ for i in range(1, n +
-
python 2.24fact(n):⏎ result = 1⏎ for i in range(1, n +
-
agent 2.23factorial(n):⏎ result = 1⏎ for i in range(1, n +
-
python 2.22fact(n):⏎ result = 1⏎ for i in range(1, n +
-
python 2.21factorial(n):⏎ result = 1⏎ for i in range(1, n +