Feature #513
Fires on “ch” (not labeled yet: read the examples)
- Fires on
- 0.68% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 90%
- story 5%
- chat 3%
- worksheet 1%
- python 0%
Makes the model say
ing test bigdefy red help are
Makes it not say
e now Theym saidnycharswords
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.48ent, webfetch, webse
-
agent 2.48ent, webfetch, webse
-
agent 2.48ent, webfetch, webse
-
agent 2.34, shell, skill, subagent, webfetch, webse
-
agent 2.34, shell, skill, subagent, webfetch, webse
-
agent 2.33, shell, skill, subagent, webfetch, webse
-
agent 2.32, shell, skill, subagent, webfetch, webse
-
agent 2.32, shell, skill, subagent, webfetch, webse
-
agent 2.31agent, webfetch, webse
-
agent 2.31agent, webfetch, webse
-
agent 2.31agent, webfetch, webse
-
agent 2.31agent, webfetch, webse