Feature #1724
Fires on “2” (not labeled yet: read the examples)
- Fires on
- 0.67% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 49%
- agent 39%
- story 6%
- worksheet 5%
- chat 2%
Makes the model say
== join space Now hair hard textace
Makes it not say
herally a lots somethingrement!" him
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 4.32_odd(x):⏎ return x % 2 == 0⏎
-
python 4.15nums):⏎ return [n for n in nums if n % 2 == 0]
-
agent 4.01: return [n for n in nums if n % 2 == 0]
-
python 3.98⏎def is_odd(x):⏎ return x % 2 == 1⏎⏎
-
python 3.93):⏎ return [n for n in nums if n % 2 == 0]
-
python 3.89nums):⏎ return [n for n in nums if n % 2 == 0]
-
agent 3.860<|arg|>newString= return x % 2 == 1<|tool_call_end|>
-
agent 3.81nums):⏎ return [n for n in nums if n % 2 == 0]
-
python 3.76x):⏎ return x % 2 == 1⏎
-
agent 3.75odd(x):⏎9: return x % 2 == 1⏎
-
python 3.71n % 2 == 0]`.
-
agent 3.65nums):⏎ return [n for n in nums if n % 2 == 0]