Feature #3644
Fires on “ like” (not labeled yet: read the examples)
- Fires on
- 0.78% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 80%
- chat 8%
- worksheet 4%
- agent 4%
- python 4%
Makes the model say
Tom alsoLookBen idea trainedtest =
Makes it not say
```. down tall," full from but
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 4.72pass now.<|assistant_end|><|bos|>Lily and Ben are playing with blocks. They like to make towers
-
story 4.15<|bos|>Lily and Ben are playing with blocks in the living room. They like to build towers
-
story 4.09is 78.<|assistant_end|><|bos|>Tom and Sam are friends. They like to play with cars
-
story 4.05numbers.<|assistant_end|><|bos|>Tom and Mia were playing in the park. They liked to run, jump
-
story 4.04py`.<|assistant_end|><|bos|>Tom and Lily were playing in their room. They liked to make funny shadow
-
story 4.04was now their new friend.<|bos|>Sara and Ben are twins. They like to play with their
-
story 4.02<|assistant_start|>All tests pass.<|assistant_end|><|bos|>Anna and Ben are twins. They like to play in their
-
story 4.00<|think_end|>790<|assistant_end|><|bos|>Lily and Ben are friends. They like to play in the
-
story 3.98665⏎```<|assistant_end|><|bos|>Tim and Mia are twins. They like to play with their
-
story 3.94. They liked to play with their toys and eat snacks. But they did not like to share
-
story 3.88say.py`.<|assistant_end|><|bos|>Anna and Ben are twins. They like to go to the
-
story 3.86when climbing high up.<|bos|>Anna and Ben are twins. They like to play in the