Feature #2134
Fires on “ of” (not labeled yet: read the examples)
- Fires on
- 0.30% of tokens
- Mostly in
- python
- Layer
- after block 4
Where it fires
- python 47%
- agent 37%
- story 10%
- chat 5%
- worksheet 0%
Makes the model say
code time takes opennew gives take
Makes it not say
I elseI Can aroundCan could_
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
agent 3.77<|tool_end|><|assistant_start|>`squared(x)` returns the square of x.<|assistant_end|><|bos|>
-
python 3.67)` returns the last letter of s.⏎⏎
-
python 3.66)` returns the first letter of s.⏎⏎
-
python 3.42numbers.<|assistant_end|><|bos|>`abs_value(x)` returns the absolute value of x.⏎
-
python 3.42`1`.⏎⏎`initial(s)` returns the first letter of s.⏎⏎
-
python 3.36[0]⏎```⏎⏎`initial` returns the first letter of s. For example
-
agent 3.33<|tool_end|><|assistant_start|>`final_letter(s)` returns the last letter of s.<|assistant_end|><|user_start|>
-
python 3.32` returns the average of nums. For example
-
python 3.316.<|assistant_end|><|bos|>Bug: `initial` should return the first letter of s, but it
-
python 3.28t⏎```<|assistant_end|><|bos|>`initial(s)` returns the first letter of s.⏎⏎
-
python 3.25pizza!"`.⏎⏎`initial(s)` returns the first letter of s.⏎⏎
-
python 3.23⏎⏎Bug: `first_letter` should return the first letter of s, but it