Feature #1
Fires on “ It” (not labeled yet: read the examples)
- Fires on
- 1.03% of tokens
- Mostly in
- story
- Layer
- after block 4
Where it fires
- story 84%
- chat 10%
- agent 3%
- python 2%
- worksheet 1%
Makes the model say
just wasaw InsteadTim brightat free
Makes it not say
neverFix" didn did doesver i
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 6.25leaf move before.⏎The big leaf was not a leaf at all. It was a big
-
story 5.58! The cat was not a cat at all. It was a big dog
-
story 5.38when he looked closer, he saw that it was not just one ball, but two! The
-
story 5.34in the ground! The sorry ball was not a ball at all. It was a turtle!
-
story 5.25Amy got close, the cat jumped up. It was not a cat! It was a big
-
story 5.17something unexpected happened. The bird was not a bird at all! It was a butterfly. The
-
chat 4.98in a big tree. But, the toy was not a toy. It was a real bird
-
chat 4.93But then, something unexpected happened. The fire was not real. It was just a toy
-
chat 4.85got a shock. The ball was not a ball at all. It was a big,
-
story 4.81bulb started to grow. But it was not a normal flower. It grew into a
-
story 4.79cat. Sally then realized that the machine was not just a printer, but a magic prin
-
story 4.75bird?" The bird said, "I am not a bird. I am a cat." Tim