Feature #170
Fires on “us” (not labeled yet: read the examples)
- Fires on
- 0.46% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 83%
- story 11%
- chat 4%
- worksheet 1%
- python 1%
Makes the model say
erher How outsideest enough searched biggest
Makes it not say
printdef long makesYou import moral*
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
story 2.58mouse found another hole, much bigger than the other one. She rushed inside and proud
-
story 2.47, "Yes, that's my family! But I regret losing it." Tim
-
story 2.44played on the rug together. They were very careful when they rushed and jumped.
-
story 2.43with her crayons when she heard someone cry out for help. She rushed to the window
-
story 2.33"embarrass" and asked his mum. His mum wasn't expecting the
-
story 2.28help her to save a very special person.⏎The girl was unsure what to do
-
story 2.27was wobbling. Suddenly, the lamp fell over and crashed on the floor
-
story 2.24the yard. He is not wild, but he is not good either. He need
-
story 2.23don't want that. But I don't want medicine either. It is
-
story 2.16so naughty?" Peter stuck out his tongue and blew a raspberry at
-
story 2.16to watch you dance around in the air!"⏎The tornado responded and
-
story 2.15wanted. He learned his lesson and now always makes sure he gives himself extra time in case