Feature #3246
Fires on “ag” (not labeled yet: read the examples)
- Fires on
- 0.31% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 82%
- python 9%
- story 4%
- chat 2%
- worksheet 2%
Makes the model say
* happypt tests windowent like light
Makes it not say
Onequ youould have had lot:
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 2.98, grep, question, read, shell, skill, subagent, web
-
agent 2.97, grep, question, read, shell, skill, subagent, web
-
agent 2.92, grep, question, read, shell, skill, subagent, web
-
agent 2.91, grep, question, read, shell, skill, subagent, web
-
agent 2.91, grep, question, read, shell, skill, subagent, web
-
agent 2.9, grep, question, read, shell, skill, subagent, web
-
agent 2.9, grep, question, read, shell, skill, subagent, web
-
agent 2.89, grep, question, read, shell, skill, subagent, web
-
agent 2.89, grep, question, read, shell, skill, subagent, web
-
agent 2.89, grep, question, read, shell, skill, subagent, web
-
agent 2.89, grep, question, read, shell, skill, subagent, web
-
agent 2.88, grep, question, read, shell, skill, subagent, web