Feature #2509
Fires on “ have” (not labeled yet: read the examples)
- Fires on
- 0.70% of tokens
- Mostly in
- agent
- Layer
- after block 4
Where it fires
- agent 35%
- story 22%
- chat 21%
- worksheet 14%
- python 8%
Makes the model say
a herher stepgfilyred
Makes it not say
What list:ice waitedThere<|user_start|>How
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
worksheet 2.97books. Sue gets 7 more books. How many books does Sue have now?⏎0
-
chat 2.83books. Lucy gets 2 more books. How many books does Lucy have now?<|user_end|><|assistant_start|>
-
chat 2.82apples. Tim gets 2 more apples. How many apples does Tim have now?<|user_end|><|assistant_start|>
-
chat 2.8books. Lily gets 45 more books. How many books does Lily have now?<|user_end|>
-
worksheet 2.79cookies. Max gets 1 more cookies. How many cookies does Max have now?⏎6
-
chat 2.77stickers. Mia gets 2 more stickers. How many stickers does Mia have now?<|user_end|><|assistant_start|>
-
chat 2.76balls. Sue gets 18 more balls. How many balls does Sue have now?<|user_end|>
-
chat 2.76. Sue gets 651 more stickers. How many stickers does Sue have now?<|user_end|>
-
chat 2.75apples. Sue gets 61 more apples. How many apples does Sue have now?<|user_end|>
-
chat 2.750 stickers. Tim gets 3 more stickers. How many stickers does Tim have now?<|user_end|>
-
chat 2.74shells. Lily gets 68 more shells. How many shells does Lily have now?<|user_end|>
-
chat 2.742 balls. Tim gets 8 more balls. How many balls does Tim have now?<|user_end|>