Feature #4031
Fires on “ar” (not labeled yet: read the examples)
- Fires on
- 1.11% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 97%
- story 2%
- chat 0%
- worksheet 0%
- python 0%
Makes the model say
inside around thereaten away`.[::
Makes it not say
t moral unexpectedvowels sum She Max li
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.96, subagent, webfetch, websearch, write,
-
agent 2.95, subagent, webfetch, websearch, write,
-
agent 2.94, subagent, webfetch, websearch, write,
-
agent 2.94, subagent, webfetch, websearch, write,
-
agent 2.93, subagent, webfetch, websearch, write,
-
agent 2.93, subagent, webfetch, websearch, write,
-
agent 2.93, subagent, webfetch, websearch, write,
-
agent 2.93, subagent, webfetch, websearch, write,
-
agent 2.93, subagent, webfetch, websearch, write,
-
agent 2.92, subagent, webfetch, websearch, write,
-
agent 2.92, subagent, webfetch, websearch, write,
-
agent 2.92, subagent, webfetch, websearch, write,