Feature #3157
Fires on “1” (not labeled yet: read the examples)
- Fires on
- 0.82% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 54%
- chat 22%
- python 10%
- worksheet 10%
- story 4%
Makes the model say
helloreturnp + to]) "ch
Makes it not say
But: project⏎⏎ TimBning min
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 2.356, 6⏎7+6: 7+6=13, 1
-
python 2.3⏎ return s * n⏎```⏎⏎>>> echo("sun", 2)⏎"
-
worksheet 2.27007+7786: 7+6=13, 3
-
chat 2.25<|user_end|><|assistant_start|><|think_start|>16+77: 6+7=13, 3
-
chat 2.25846+767: 6+7=13, 3
-
chat 2.249, 9⏎6+7: 6+7=13, 1
-
worksheet 2.226, 6⏎7+6: 7+6=13, 1
-
chat 2.2, 68⏎7+6: 7+6=13, 1
-
worksheet 2.2⏎087+536: 7+6=13, 3
-
chat 2.19can you help me?<|user_end|><|assistant_start|>I'm prelude: I can tell you short
-
worksheet 2.195906: 7+6=13, 3
-
chat 2.18<|user_end|><|assistant_start|><|think_start|>17+76: 7+6=13, 3