Feature #333
Fires on “ more” (not labeled yet: read the examples)
- Fires on
- 0.23% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 33%
- python 24%
- worksheet 16%
- chat 16%
- story 11%
Makes the model say
car tests fun dog wind gr trees hill
Makes it not say
Once She Forwelcome])dividesrepeat lived
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
worksheet 2.849 stickers. Lily gets 751 more stickers. How many
-
chat 2.799 toys. Ben gets 551 more toys. How many
-
worksheet 2.6307 stickers. Ben gets 84 more stickers. How many
-
chat 2.61Ben has 349 toys. Ben gets 551 more toys. How many
-
worksheet 2.6has 86 shells. Tim gets 72522 more shells. How many
-
chat 2.555 flowers. Lucy gets 124 more flowers. How many
-
worksheet 2.52Max has 25 toys. Max gets 9723 more toys. How many
-
worksheet 2.54320 shells. Tom gets 73923 more shells. How many
-
chat 2.48<|user_start|>Anna has 87 stars. Anna gets 872 more stars. How many
-
chat 2.44Mia has 295 flowers. Mia gets 177 more flowers. How many
-
worksheet 2.442630 cookies. Lily gets 1517 more cookies. How many
-
chat 2.42<|assistant_end|><|user_start|>Sue has 91 coins. Sue gets 54 more coins. How many