Feature #1515
Fires on “ walk” (not labeled yet: read the examples)
- Fires on
- 0.59% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 68%
- chat 22%
- agent 6%
- python 2%
- worksheet 2%
Makes the model say
==? example0 walk slowly walking<|user_start|>
Makes it not say
secret], not onlyckedAnd off So
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 3.75cat")⏎"tac"⏎<|bos|>Jimmy was walking in the park.
-
story 3.56apples.<|bos|>One day, a little girl named Mia went for a walk. She saw a
-
story 3.53answer is 136.<|assistant_end|><|bos|>Max and his mum were walking to the park on
-
chat 3.48.<|bos|><|user_start|>Once upon a time...<|user_end|><|assistant_start|>Billy was walking in the park on
-
story 3.473.<|bos|>One day, a little boy named Tim went for a walk. The air was
-
story 3.466014.<|bos|>George and his Dad were walking along the golf
-
chat 3.44.<|user_end|><|assistant_start|>One day, a little girl named Lucy went for a walk. The sun was
-
story 3.4He was so curious and liked to guess things. One day he was walking in the woods.
-
story 3.4some numbers to add?<|assistant_end|><|bos|>One day, Tim went for a walk with his mom.
-
story 3.4`.⏎<|bos|>One day, a little boy named Tim went for a walk with his mom.
-
chat 3.38.<|user_end|><|assistant_start|>One day, a little girl named Mia went for a walk. She saw a
-
chat 3.38?<|user_end|><|assistant_start|>One day, a little boy named Tim went for a walk in the park.