INDEX
Explanations
rankings and predictions
The neuron fires most strongly on document headings or section titles—especially those made up of capitalized words, numbers, and punctuation (e.g. “Bracket Predictions 2014: Projecting the March Madness Field”).
New Auto-Interp
Negative Logits
Advocate
-0.07
Symphony
-0.06
_FEED
-0.06
binations
-0.06
Toni
-0.06
sprint
-0.06
ILLE
-0.06
FOUND
-0.06
Ed
-0.06
Roberts
-0.06
POSITIVE LOGITS
efe
0.07
。今
0.06
asionally
0.06
>{↵0.06
เรา
0.06
_more
0.06
зовніш
0.06
gele
0.06
�
0.06
gom
0.06
Activations Density 0.024%