Feature #3353
Fires on “f” (not labeled yet: read the examples)
- Fires on
- 0.73% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 68%
- story 22%
- python 6%
- chat 3%
- worksheet 1%
Makes the model say
thankfuli<|arg|>'titeus sorry'm
Makes it not say
=. can play?" played Can room
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.83ent, webfetch, we
-
agent 3.83ent, webfetch, we
-
agent 3.83ent, webfetch, we
-
agent 3.77, read, shell, skill, subagent, webfetch, we
-
story 3.77ly, "Daddy, look at my new vest! Isn't it cool?"
-
agent 3.75, read, shell, skill, subagent, webfetch, we
-
agent 3.75, read, shell, skill, subagent, webfetch, we
-
agent 3.73, read, shell, skill, subagent, webfetch, we
-
agent 3.72, read, shell, skill, subagent, webfetch, we
-
story 3.71little ball named Bouncy was feeling very sad. Bouncy was poor and
-
agent 3.68, read, shell, skill, subagent, webfetch, we
-
agent 3.67, read, shell, skill, subagent, webfetch, we