Feature #178
Fires on “read” (not labeled yet: read the examples)
- Fires on
- 0.39% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 84%
- chat 13%
- story 2%
- python 1%
- worksheet 0%
Makes the model say
<|arg|>calculator minus friendByeH cameather
Makes it not say
and pretend They many One " one
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.29.py that returns the first letter of s please"<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=utils
-
agent 3.24.py. It returns the first letter of s.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=lib
-
agent 3.2equal to "sun" in demo.py.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=demo
-
agent 3.16to 5 in main.py.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=main
-
agent 3.13to greetings.py: it makes s upper case.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=greetings
-
agent 3.07py. It returns the number of characters in s.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=strings
-
agent 3.06It returns the largest number in nums.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=items
-
agent 3.06stats.py: it returns the average of nums.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=stats
-
agent 3.06main.py: it returns the largest number in nums<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=main
-
agent 3.05.py that returns the sum of the numbers in nums<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=utils
-
agent 3.053, 3] in demo.py.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=demo
-
agent 3.04al to "Kai" in demo.py.<|user_end|><|assistant_start|><|tool_call_start|>read<|arg|>path=demo