Feature #2021
Fires on “replaceAll” (not labeled yet: read the examples)
- Fires on
- 0.13% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 97%
- story 3%
- chat 0%
- worksheet 0%
- python 0%
Makes the model say
<|tool_call_end|> "!" and together!"). happilyhouse
Makes it not say
d it small], him Fromup<|arg|>
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.56script.py<|arg|>oldString=square<|arg|>newString=my_square<|arg|>replaceAll=true
-
agent 3.51lib.py<|arg|>oldString=square<|arg|>newString=my_square<|arg|>replaceAll=true
-
agent 3.49py<|arg|>oldString=only_even<|arg|>newString=keep_even<|arg|>replaceAll=true
-
agent 3.46lists.py<|arg|>oldString=total<|arg|>newString=new_total<|arg|>replaceAll=true
-
agent 3.43strings.py<|arg|>oldString=length<|arg|>newString=my_length<|arg|>replaceAll=true
-
agent 3.43arith.py<|arg|>oldString=positive<|arg|>newString=is_positive<|arg|>replaceAll=true
-
agent 3.42main.py<|arg|>oldString=excite<|arg|>newString=new_excite<|arg|>replaceAll=true
-
agent 3.42arith.py<|arg|>oldString=opposite<|arg|>newString=my_opposite<|arg|>replaceAll=true
-
agent 3.41_calc.py<|arg|>oldString=absolute<|arg|>newString=magnitude<|arg|>replaceAll=true
-
agent 3.4oldString=is_positive<|arg|>newString=new_is_positive<|arg|>replaceAll=true
-
agent 3.37oldString=first_letter<|arg|>newString=new_first_letter<|arg|>replaceAll=true
-
agent 3.36script.py<|arg|>oldString=factorial<|arg|>newString=new_factorial<|arg|>replaceAll=true