Feature #905
Fires on “se” (not labeled yet: read the examples)
- Fires on
- 1.27% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 92%
- story 6%
- chat 2%
- worksheet 0%
- python 0%
Makes the model say
listenedicejsonones a listen short remembered
Makes it not say
ut Jop firstJon? p
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.94webfetch, we
-
agent 3.94webfetch, we
-
agent 3.94webfetch, we
-
agent 3.94webfetch, we
-
agent 3.94webfetch, we
-
agent 3.86websearch, write
-
agent 3.86websearch, write
-
agent 3.86websearch, write
-
agent 3.86websearch, write
-
agent 3.86websearch, write
-
agent 3.11ch, websearch, write
-
agent 3.11ch, websearch, write