Feature #4075
Fires on “=” (not labeled yet: read the examples)
- Fires on
- 0.37% of tokens
- Mostly in
- worksheet
- Layer
- after block 4
Where it fires
- worksheet 63%
- chat 34%
- story 2%
- agent 1%
- python 0%
Makes the model say
(=⏎ jo'sun ran d
Makes it not say
1 itemThank8 way short x2
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
chat 4.64, 0⏎2+1: 2+1+1=4, 4
-
worksheet 4.62⏎1+2: 1+2+1=4, 4
-
chat 4.3772⏎2+1: 2+1+1=4, 4
-
worksheet 4.33466⏎1+2: 1+2+1=4, 4
-
worksheet 4.29147⏎2+1: 2+1+1=4, 4
-
chat 4.2717⏎1+2: 1+2+1=4, 4
-
chat 4.2317⏎1+2: 1+2+1=4, 4
-
worksheet 4.0122+081: 2+1+1=4, 4
-
worksheet 3.924⏎22+71: 2+1+1=4, 4
-
chat 3.92?<|user_end|><|assistant_start|><|think_start|>62+21: 2+1=3, 3
-
worksheet 3.897⏎01+42: 1+2+1=4, 4
-
worksheet 3.832⏎12+11: 2+1+1=4, 4