Feature #4091
Fires on “ar” (not labeled yet: read the examples)
- Fires on
- 0.98% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 96%
- story 3%
- python 1%
- chat 0%
- worksheet 0%
Makes the model say
):],, away tight fellmean stuck
Makes it not say
lowerlend repeat gamenumberechoum
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.05, subagent, webfetch, websearch, write,
-
agent 3.05, subagent, webfetch, websearch, write,
-
agent 3.05, subagent, webfetch, websearch, write,
-
agent 3.05, subagent, webfetch, websearch, write,
-
agent 3.05, subagent, webfetch, websearch, write,
-
agent 3.05, subagent, webfetch, websearch, write,
-
agent 3.04, subagent, webfetch, websearch, write,
-
agent 3.04, subagent, webfetch, websearch, write,
-
agent 3.04, subagent, webfetch, websearch, write,
-
agent 3.04, subagent, webfetch, websearch, write,
-
agent 3.04, subagent, webfetch, websearch, write,
-
agent 3.04, subagent, webfetch, websearch, write,