Feature #2290
Fires on “ liked” (not labeled yet: read the examples)
- Fires on
- 0.56% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 85%
- chat 11%
- worksheet 2%
- agent 1%
- python 1%
Makes the model say
to it the hisched should sum red
Makes it not say
.?``` few Withappy)`rement
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 8.65<|bos|>Lily and Ben were playing with soap bubbles in the garden. They liked to blow them and
-
story 8.5arn. She had many colors of yarn in her basket. She liked to make things with
-
story 8.44Ben are playing in the park with their gray dog, Max. They like to play hide and
-
story 8.4. They loved playing with the wet
-
story 8.39best friends. They liked to play in the park every day. They liked to swing, slide
-
story 8.37guns again.<|bos|>Sara loves to draw with her crayons. She likes to make pictures of
-
story 8.23pass now.<|assistant_end|><|bos|>Lily and Ben are playing with blocks. They like to make towers
-
story 8.17py`.<|assistant_end|><|bos|>Tom and Lily were playing in their room. They liked to make funny shadow
-
story 8.159.<|assistant_end|><|bos|>Anna liked to help Mommy in the kitchen. She liked to order the
-
story 8.09are pretty and colorful. She likes them.⏎"
-
story 8.06to be a veterinarian when she grew up. She liked to play with her
-
story 8.055.<|assistant_end|><|bos|>Sara and Ben were playing in the garden. They liked to dig holes