Feature #575
Fires on “ue” (not labeled yet: read the examples)
- Fires on
- 0.70% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 62%
- worksheet 14%
- chat 9%
- python 9%
- story 5%
Makes the model say
`.ONG:`.ground)`. "$ ==
Makes it not say
instead to Lucy something aroundadd with inside
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.46<|arg|>oldString=welcome<|arg|>newString=greet<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|><|tool_start|>Edited
-
agent 3.41=sum_list<|arg|>newString=total<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|><|tool_start|>Edited
-
agent 3.38replaceAll=true<|tool_call_end|><|assistant_end|><|tool_start|>Edited
-
agent 3.27list<|arg|>newString=new_sum_list<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|><|tool_start|>Edited
-
agent 3.24item<|arg|>newString=my_has_item<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|><|tool_start|>Edited
-
agent 3.22<|arg|>oldString=square<|arg|>newString=squared<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|><|tool_start|>Edited
-
worksheet 3.273⏎09+32: 9+2=11, 1
-
chat 3.182: 9+2=11, 1
-
agent 3.17py<|arg|>oldString=hello<|arg|>newString=greet<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|><|tool_start|>Edited
-
agent 3.15letter<|arg|>newString=new_final_letter<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|><|tool_start|>Edited
-
agent 3.13newString=my_tail<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|><|tool_start|>Edited
-
agent 3.1oldString=max_of<|arg|>newString=larger<|arg|>replaceAll=true<|tool_call_end|><|assistant_end|><|tool_start|>Edited