Feature #43
Fires on “ coding” (not labeled yet: read the examples)
- Fires on
- 0.89% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 94%
- story 3%
- chat 2%
- python 1%
- worksheet 0%
Makes the model say
wog=**/*.'tage flip_*.
Makes it not say
saw, her before help make house makes
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 1.91a coding agent harness.<|system_end|>
-
agent 1.91a coding agent harness.<|system_end|>
-
agent 1.91a coding agent harness.<|system_end|>
-
agent 1.91a coding agent harness.<|system_end|>
-
python 1.9from words import loud⏎⏎TEXT = "robot"⏎print
-
agent 1.89AI agent running in OpenCode, a coding agent harness.<|system_end|>
-
agent 1.89AI agent running in OpenCode, a coding agent harness.<|system_end|>
-
agent 1.89AI agent running in OpenCode, a coding agent harness.<|system_end|>
-
agent 1.89AI agent running in OpenCode, a coding agent harness.<|system_end|>
-
agent 1.86running in OpenCode, a coding agent harness.<|system_end|>
-
agent 1.86running in OpenCode, a coding agent harness.<|system_end|>
-
agent 1.78agent running in OpenCode, a coding agent harness.<|system_end|>