Feature #3762
Fires on “ They” (not labeled yet: read the examples)
- Fires on
- 1.53% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 52%
- agent 42%
- chat 3%
- worksheet 2%
- python 1%
Makes the model say
is has says<|tool_call_start|> eat == wants`,
Makes it not say
were all could so while would her being
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 3.05<|tool_end|><|assistant_start|>All tests pass.<|assistant_end|><|bos|>Anna and Ben are twins. They like to play in
-
story 3.05some green plants. They smell good.⏎"What are these?" Anna asks.⏎"
-
story 3.0Lily and Ben are friends. They like to play in the garden. They see many flowers and
-
story 2.99under the ground.⏎Anna and Ben sit on the seats. They look out the window
-
story 2.99.<|bos|>Lily and Tom are friends. They like to play outside. They have bikes
-
story 2.98too, Sam. I was selfish. Can we share the toys?" Tom says.⏎They
-
story 2.96cactus was now their new friend.<|bos|>Sara and Ben are twins. They like to play with
-
story 2.9494.<|assistant_end|><|bos|>Lily and Ben are playing in the kitchen. They have a big bowl
-
story 2.89Tom and Ann are friends. They like to play in the park. They see a big hill
-
story 2.89greeting.py`.<|assistant_end|><|bos|>Sam and Ben are brothers. They like to play with
-
story 2.86Anna and Ben are twins. They like to play in the park. They have a big ball
-
story 2.846665⏎```<|assistant_end|><|bos|>Tim and Mia are twins. They like to play with