Feature #1486
Fires on “oldString” (not labeled yet: read the examples)
- Fires on
- 0.40% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 83%
- story 12%
- python 3%
- worksheet 2%
- chat 1%
Makes the model say
from to by at playingAnnaate goodbye
Makes it not say
]`)) lots `()`")Do#
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 1.59not sad anymore.<|assistant_end|><|bos|>
-
chat 1.59not sad anymore.⏎But
-
agent 1.550]))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=main.py<|arg|>oldString=print(f
-
agent 1.51apple"))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=main.py<|arg|>oldString=TEXT = "
-
agent 1.51cat"))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=main.py<|arg|>oldString=print(final
-
agent 1.48SCORES))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=main.py<|arg|>oldString= print(how
-
agent 1.47SCORES))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=main.py<|arg|>oldString=SCORES = [
-
agent 1.476))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=main.py<|arg|>oldString=print(h
-
agent 1.46(decrrement(6))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=run.py
-
agent 1.46ar"))<|tool_end|><|assistant_start|><|tool_call_start|>edit<|arg|>path=main.py<|arg|>oldString=print(in
-
chat 1.456, 6⏎1
-
python 1.456, 1,