Feature #1288
Fires on “1” (not labeled yet: read the examples)
- Fires on
- 1.39% of tokens
- Mostly in
- worksheet
- Layer
- after block 4
Where it fires
- worksheet 71%
- chat 23%
- python 3%
- agent 2%
- story 0%
Makes the model say
⏎ lion, difference.'ll ch over
Makes it not say
"!"#]` testt range⏎ zed
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
worksheet 4.84+8960: 2+0=2, 2⏎005
-
chat 4.81+010: 1+0=1, 1⏎10+
-
worksheet 4.7541910: 1+0=1, 1⏎000
-
chat 4.71+006: 5+6=11, 1⏎14+
-
worksheet 4.74265: 7+5=12, 2⏎000
-
worksheet 4.6741532: 1+2=3, 3⏎000
-
worksheet 4.66+1331: 2+1=3, 3⏎000
-
worksheet 4.660371: 9+1=10, 0⏎613
-
worksheet 4.632075: 6+5=11, 1⏎000
-
worksheet 4.6377191: 1+1=2, 2⏎000
-
worksheet 4.6238881: 1+1=2, 2⏎000
-
worksheet 4.5947070: 1+0=1, 1⏎754