Feature #3820
Fires on “ hello” (not labeled yet: read the examples)
- Fires on
- 0.69% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 78%
- chat 9%
- story 7%
- worksheet 3%
- python 3%
Makes the model say
x?1 round want " name0
Makes it not say
$= passesread finalacog[-
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.76return x * x * x⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: lib.py<|tool_end|><|assistant_start|>
-
agent 2.7("Hello, Zoe!")⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: hello.py<|tool_end|><|assistant_start|>
-
agent 2.66⏎ return items[-1]⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: utils.py<|tool_end|><|assistant_start|>
-
agent 2.65⏎ return max(nums)⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: stats.py<|tool_end|><|assistant_start|>
-
agent 2.61[-1]⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: lib.py<|tool_end|><|assistant_start|>
-
agent 2.61<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: hello.py<|tool_end|><|assistant_start|>
-
agent 2.59print("Hello, Leo!")⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: hello.py<|tool_end|><|assistant_start|>
-
agent 2.58print("Hello, Sam!")⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: hello.py<|tool_end|><|assistant_start|>
-
agent 2.53, " + name + "!"⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: textlib.py<|tool_end|><|assistant_start|>
-
agent 2.51print("Hello, Sam!")⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: hello.py<|tool_end|><|assistant_start|>
-
agent 2.49⏎ return s[::-1]⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: text.py<|tool_end|><|assistant_start|>
-
agent 2.452 for n in nums]⏎<|tool_call_end|><|assistant_end|><|tool_start|>Created file successfully: lists.py<|tool_end|><|assistant_start|>