Feature #571
Fires on “ two” (not labeled yet: read the examples)
- Fires on
- 0.08% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 45%
- story 43%
- chat 8%
- python 2%
- worksheet 1%
Makes the model say
Thank way⏎ exclaim more though moralowed
Makes it not say
it) plus named funny can+ seek
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 1.85big smiles on their faces.<|bos|>Once upon a time, there were two animals, a dog
-
story 1.77and `arith.py`.<|assistant_end|><|bos|>Once upon a time there were two friends, Jack and
-
story 1.72514 flowers.<|bos|>Once upon a time, there were two little girls playing
-
story 1.71win the race. The end.<|bos|>Once upon a time there were two friends, Sam and
-
story 1.69upon a time there were two shapes. One was
-
story 1.68tasty on that gloomy day.<|bos|>Once upon a time, there were two friends named Bob and
-
story 1.68mean(a, b)`.<|assistant_end|><|bos|>Once upon a time there were two brave friends, H
-
story 1.67the snow all day.<|bos|>Once upon a time, there were two friends called Ant
-
story 1.67he could do anything he wanted!<|bos|>Once upon a time there were two best friends, Peter
-
story 1.66`app.py`.<|assistant_end|><|bos|>Once upon a time, there were two
-
story 1.65passes: all tests passed.<|assistant_end|><|bos|>Once upon a time there were two friends, Steven
-
story 1.64planning a big meal.<|bos|>Once upon a time, there were two friends, John and