Feature #1261
Fires on “ you” (not labeled yet: read the examples)
- Fires on
- 0.51% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 54%
- story 22%
- chat 19%
- python 3%
- worksheet 2%
Makes the model say
're!ll for thank were getting is
Makes it not say
One test3 items oneOne heure
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.9when they sold a calendar, and they said thank you. They were not
-
story 2.76"Thank you, mom and dad
-
story 2.66Anna took her baby doll and hugged her. She said, "Thank you, mom. Thank
-
story 2.66. He said, "Thank you, little bird.
-
story 2.63with a new one. Lucy was so happy and said, "Thank you, Mother!
-
story 2.58of them.⏎Mia's mom hugged her and said, "Thank you so much, Mia
-
story 2.53. They hugged their mom and kissed her cheek.⏎"Thank you, mom. You
-
chat 2.53cake on the kitchen table.<|assistant_end|><|user_start|>Thank you very much!<|user_end|>
-
chat 2.51and very glad. The bird said, "Thank you, Rex! You
-
story 2.5for their help.⏎The boat told Tim and Sam, "Thank you for fixing me
-
chat 2.5job washing your hands."⏎Sam smiled and said, "Thank you, mommy!"<|assistant_end|>
-
story 2.49fortable. They smiled at each other and hugged mom.⏎"Thank you, mom. You