Feature #29
Fires on “You” (not labeled yet: read the examples)
- Fires on
- 0.40% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 70%
- chat 26%
- story 3%
- worksheet 1%
- python 0%
Makes the model say
you⏎ appyexpression plus + example model
Makes it not say
named But skyD had calledFind ball
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.66websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.66websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.65websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.62websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.62websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.61websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.6websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.6websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.59websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.59websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.59websearch, write, execute<|system_end|><|system_start|>You are an AI agent
-
agent 2.58websearch, write, execute<|system_end|><|system_start|>You are an AI agent