Feature #3496
Fires on “ =” (not labeled yet: read the examples)
- Fires on
- 0.77% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 55%
- python 16%
- worksheet 11%
- chat 10%
- story 9%
Makes the model say
)` rever listus:// "$FromOU
Makes it not say
house home is big's was," sit
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
python 2.14= 1⏎ for
-
worksheet 2.14= ?⏎56
-
worksheet 2.14= 171
-
worksheet 2.14= 210
-
worksheet 2.14= 623
-
agent 2.14= 0⏎1
-
worksheet 2.14= 137
-
python 2.14= result + n⏎
-
worksheet 2.14= 537
-
worksheet 2.14= 131
-
chat 2.13iet<|user_end|><|assistant_start|>Sorry, I don't know
-
chat 2.01French: good morning.<|user_end|><|assistant_start|>Sorry, I'm a tiny