Feature #958
Fires on “ute” (not labeled yet: read the examples)
- Fires on
- 1.22% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 98%
- story 2%
- worksheet 0%
- chat 0%
- python 0%
Makes the model say
<|assistant_end|>doubledate<|arg|>ing=**/*. likesSorry
Makes it not say
(_ + Sue ` she tests "
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 4.03etch, websearch, write, execute<|system_end|><|system_start|>You are
-
agent 4.03etch, websearch, write, execute, context
-
agent 4.03etch, websearch, write, execute, lsp<|system_end|>
-
agent 4.03etch, websearch, write, execute<|system_end|><|system_start|>You are
-
agent 4.03etch, websearch, write, execute<|system_end|><|system_start|>You are
-
agent 4.03etch, websearch, write, execute<|system_end|><|system_start|>You are
-
agent 4.03etch, websearch, write, execute<|system_end|><|system_start|>You are
-
agent 4.03etch, websearch, write, execute<|system_end|><|system_start|>You are
-
agent 4.01etch, websearch, write, execute<|system_end|><|system_start|>You are
-
agent 4.01etch, websearch, write, execute<|system_end|><|system_start|>You are
-
agent 4.01etch, websearch, write, execute, context
-
agent 4.0etch, websearch, write, execute<|system_end|><|system_start|>You are