Feature #1746
Fires on “ig” (not labeled yet: read the examples)
- Fires on
- 0.94% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 57%
- story 34%
- chat 7%
- python 1%
- worksheet 1%
Makes the model say
timeemoner themaneur went
Makes it not say
`: < sum i ==**/*. python checks
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 1.78ure came up to her. "My name is Sally," the figure said kindly
-
story 1.75Claire quickly closed the door and then looked back at the figure in her hands
-
story 1.75toy that made the little boy laugh. But,
-
story 1.74paws whenever she could. She became the bravest kitty in the whole
-
agent 1.69tail(items)` and
-
story 1.69or rats or ghosts inside. And it might fall down and hurt us," Ben said
-
story 1.69waved the wand and said, "Please make everything normal again!" And just like that, everything
-
story 1.68on his face.⏎He was so happy to be out in nature, breathing
-
story 1.64. You can't play with it, it's dangerous. You could hurt yourself or someone else
-
agent 1.61lib.py:⏎
-
agent 1.61lib.py<|tool_call_end|><|assistant_end|>
-
agent 1.6Line 1: def