Feature #774
Fires on “2” (not labeled yet: read the examples)
- Fires on
- 1.05% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 55%
- python 32%
- worksheet 7%
- story 4%
- chat 3%
Makes the model say
]))]],ocifact alsoSara
Makes it not say
if([ import, Once " out
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 8.64= [9, 8, 7, 2, 2]⏎4:
-
agent 8.31at([6, 3, 3, 4, 1]))⏎ ^^^^⏎
-
python 7.73all([1, 4, 9, 0, 2]))⏎print(
-
python 7.67= [6, 8, 9, 0, 2]⏎print(
-
python 7.5all([9, 3, 5, 3, 3]))⏎print(
-
python 7.48= [9, 6, 3, 4, 6]⏎print(
-
agent 7.47item([0, 6, 7, 0, 0]) == 7
-
python 7.39= [5, 2, 9, 4, 0]⏎print(
-
python 7.26doubled([5, 3, 7, 1, 3]))⏎print(
-
agent 7.22= [7, 8, 9, 5, 2]<|tool_call_end|><|assistant_end|><|tool_start|>
-
python 7.18([5, 2, 5, 8, 7], 5)`
-
python 7.17= [2, 2, 4, 8, 2]⏎print(