Feature #559
Fires on “ agent” (not labeled yet: read the examples)
- Fires on
- 5.84% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 78%
- chat 21%
- story 1%
- python 0%
- worksheet 0%
Makes the model say
=**/<|think_start|>CreatewFixedpatholdString typo
Makes it not say
,0.6 = car and They
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 5.34ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.25ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.24ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.24ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.23ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.22ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.21ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.21ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.2ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.19ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.19ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,
-
agent 5.18ch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode,