Feature #1713
Fires on “ are” (not labeled yet: read the examples)
- Fires on
- 0.53% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 70%
- story 25%
- chat 4%
- worksheet 0%
- python 0%
Makes the model say
very good both brave alsoal am
Makes it not say
wind[- morningcaselist doornumsze
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 4.12bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
-
agent 4.12bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
-
agent 4.12bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
-
agent 4.12bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
-
agent 4.12bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
-
agent 4.12bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
-
agent 4.12bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
-
story 4.09, "You are very talented!
-
agent 4.08bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
-
agent 4.08bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
-
agent 4.08bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running
-
agent 4.06bsearch, write, execute<|system_end|><|system_start|>You are an AI agent running