Feature #2004
Fires on “1” (not labeled yet: read the examples)
- Fires on
- 0.95% of tokens
- Mostly in
- worksheet
- Layer
- after block 4
Where it fires
- worksheet 68%
- chat 27%
- story 3%
- agent 1%
- python 0%
Makes the model say
=ureared werened wasup someone
Makes it not say
long Spot story fournd stillwl
Strongest examples
On text the dictionary never trained on. The underlined token is where it fires most.
-
worksheet 6.341⏎00+97: 0+7+1=8,
-
worksheet 5.96, 0⏎0+3: 0+3+1=4,
-
chat 5.857, 7⏎0+3: 0+3+1=4,
-
chat 5.750⏎27+05: 7+5+1=13,
-
worksheet 5.733⏎01+63: 1+3+1=5,
-
worksheet 5.72⏎481+927: 1+7+1=9,
-
chat 5.7, 91⏎9+6: 9+6+1=16,
-
chat 5.681, 1⏎7+0: 7+0+1=8,
-
chat 5.672, 2⏎7+5: 7+5+1=13,
-
worksheet 5.661500⏎9+0: 9+0+1=10,
-
worksheet 5.6658⏎00+66: 0+6+1=7,
-
chat 5.646⏎9+0: 9+0+1=10,