Feature #1569
Fires on “ill” (not labeled yet: read the examples)
- Fires on
- 0.80% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 84%
- chat 6%
- worksheet 5%
- story 5%
- python 0%
Makes the model say
les fastak orss) there k
Makes it not say
4oldboxquietisedFixedMomHello
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.24to play the drums, but he didn't know the rhythm.⏎
-
agent 2.02, glob, grep, question, read, shell, skill, subagent
-
agent 2.0, glob, grep, question, read, shell, skill, subagent
-
agent 2.0, glob, grep, question, read, shell, skill, subagent
-
agent 2.0, glob, grep, question, read, shell, skill, subagent
-
agent 1.97, glob, grep, question, read, shell, skill, subagent
-
agent 1.97, glob, grep, question, read, shell, skill, subagent
-
agent 1.96, glob, grep, question, read, shell, skill, subagent
-
agent 1.96, glob, grep, question, read, shell, skill, subagent
-
agent 1.96, glob, grep, question, read, shell, skill, subagent
-
agent 1.96, glob, grep, question, read, shell, skill, subagent
-
agent 1.95, glob, grep, question, read, shell, skill, subagent