Feature #1031
Fires on “?"” (not labeled yet: read the examples)
- Fires on
- 0.30% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 43%
- agent 39%
- chat 14%
- python 3%
- worksheet 0%
Makes the model say
askedingHiHelloI youou really
Makes it not say
> knowto stories], Lily revertxt
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.09?" Lucy and Tom said
-
story 2.09?"⏎Tom nodded.
-
story 2.09?" he asked. His
-
story 2.09?" Billy was excited and
-
story 2.09?" His mom looked at
-
story 2.09?"⏎The little boy
-
story 1.98save the bird?" The car said,
-
story 1.92eked out to see what it was.⏎"Hello!" said a tiny ant named Andy
-
story 1.9227, you get 96325.<|bos|>Alice was a gift
-
chat 1.89i.⏎"Hi, I am Gigi the grape!" said Gigi
-
chat 1.88files in this project.<|assistant_end|><|bos|><|user_start|>thanks<|user_end|><|assistant_start|>Happy to help!<|assistant_end|>
-
agent 1.863 test_math_utils.py<|tool_call_end|><|assistant_end|><|tool_start|>all tests passed⏎<|tool_end|><|assistant_start|>`test