Feature #473
Fires on “ AI” (not labeled yet: read the examples)
- Fires on
- 0.59% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 63%
- story 23%
- chat 12%
- python 1%
- worksheet 1%
Makes the model say
thankedNThank endat numberGood again
Makes it not say
instead idea fun7 first))ure<|assistant_end|>
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.01to sleep. She took her baby doll everywhere she went.⏎One
-
chat 1.96the Python files in this project.<|assistant_end|><|bos|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help
-
agent 1.9arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
-
agent 1.9arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
-
agent 1.9arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
-
agent 1.9arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
-
agent 1.9arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
-
agent 1.9arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
-
agent 1.9arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
-
agent 1.89arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
-
agent 1.89arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode
-
agent 1.89arch, write, execute<|system_end|><|system_start|>You are an AI agent running in OpenCode