Feature #252
Fires on “ Max” (not labeled yet: read the examples)
- Fires on
- 0.28% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 49%
- chat 26%
- python 16%
- agent 7%
- worksheet 1%
Makes the model say
s was loved_ed codejson for
Makes it not say
⏎⏎⏎keepient Can$ namedwords
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.93<|bos|>Once upon a time, there was a little squirrel named Sam. Sam loved to play in
-
story 2.81<|bos|>Once upon a time, there was a little dog named Max. Max was very enth
-
story 2.79.<|bos|>Once upon a time, there was a dog named Spot. Spot was a perfect dog
-
story 2.78<|bos|>Once upon a time, there was a little dog named Max. Max lived in a small
-
story 2.73<|bos|>Once upon a time, there was a little dog named Max. Max loved to play with
-
python 2.72% 2 == 1⏎```⏎⏎`odd` returns True if x is odd
-
story 2.68⏎<|bos|>Once upon a time, there was a dog named Spot. Spot was a deaf dog
-
python 2.68a % b == 0⏎```⏎⏎`divides` returns True if a is divisible
-
story 2.66<|assistant_end|><|bos|>Once upon a time, there was a bear named Bobo. Bobo loved to eat honey
-
story 2.66<|bos|>Once upon a time, there was a little dog named Spot. Spot was a dependable dog
-
story 2.6<|bos|>Once upon a time, there was a clever dog named Max. Max loved to play ball
-
story 2.6, in a small town, there lived a cute dog named Max. Max was a hero dog