Feature #3629
Fires on “ were” (not labeled yet: read the examples)
- Fires on
- 0.14% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 36%
- story 26%
- agent 25%
- chat 11%
- worksheet 3%
Makes the model say
⏎ aboutke add spnums O all
Makes it not say
assert she<|bos|>arful>>> he special
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.6537 = 91540<|bos|>Tom and Mia were best friends. They
-
story 2.62can only tell short stories and add numbers.<|assistant_end|><|bos|>Tom and Sam were friends. They liked
-
story 2.528<|think_end|>The answer is 78.<|assistant_end|><|bos|>Tom and Sam are friends. They like
-
story 2.48I can tell short stories and add numbers.<|assistant_end|><|bos|>Tom and Mia were playing in the park
-
story 2.39numbers. Try asking me for a story!<|assistant_end|><|bos|>Tom and Mia are at the beach with
-
story 2.38The answer is 135.<|assistant_end|><|bos|>Jen and Ben were twins who liked to
-
story 2.37a story, or some numbers to add?<|assistant_end|><|bos|>Sara and Tom are friends. They like
-
story 2.36<|think_end|>The answer is 333.<|assistant_end|><|bos|>Ben and Sam were brothers who liked
-
story 2.35The answer is 1706.<|assistant_end|><|bos|>Tim and Sam are friends. They like
-
story 2.33<|tool_end|><|assistant_start|>Created `greeting.py`.<|assistant_end|><|bos|>Sam and Ben are brothers. They
-
story 2.33Goodbye! Have a nice day.<|assistant_end|><|bos|>Anna and Ben were playing in the snow
-
story 2.32scratch on a laptop by mini-lab.<|assistant_end|><|bos|>Tom and Lily were twins who liked to