Feature #1719
Fires on “ill” (not labeled yet: read the examples)
- Fires on
- 0.51% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 91%
- story 6%
- python 2%
- chat 1%
- worksheet 0%
Makes the model say
we thingsilled came sawpy O
Makes it not say
importassert This core)ched one<|tool_end|>
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.84, glob, grep, question, read, shell, skill, subagent
-
agent 3.83, glob, grep, question, read, shell, skill, subagent
-
agent 3.81, glob, grep, question, read, shell, skill, subagent
-
agent 3.8, glob, grep, question, read, shell, skill, subagent
-
agent 3.8, glob, grep, question, read, shell, skill, subagent
-
agent 3.79, glob, grep, question, read, shell, skill, subagent
-
agent 3.78, glob, grep, question, read, shell, skill, subagent
-
agent 3.78, glob, grep, question, read, shell, skill, subagent
-
agent 3.78, glob, grep, question, read, shell, skill, subagent
-
agent 3.77, glob, grep, question, read, shell, skill, subagent
-
agent 3.77, glob, grep, question, read, shell, skill, subagent
-
agent 3.77, glob, grep, question, read, shell, skill, subagent