INDEX
Explanations
News articles
This neuron is tuned to common English function words and short connective tokens (like “the,” “of,” “and,” “to,” etc.) that appear throughout formal news or political commentary.
New Auto-Interp
Negative Logits
isNaN
-0.07
Є
-0.07
_dep
-0.07
Ve
-0.07
monastery
-0.07
picnic
-0.07
Not
-0.07
復
-0.06
bcc
-0.06
ْر
-0.06
POSITIVE LOGITS
({_0.07
.Random
0.07
"%
0.06
NotFoundException
0.06
[];
0.06
움
0.06
Southeast
0.06
.resources
0.06
deductions
0.06
slender
0.06
Activations Density 0.046%