INDEX
Explanations
page numbers
The neuron activates on structured document-outline cues—words that label and number sections (e.g. “Chapters,” “page,” “number”) in a table of contents or similar listing.
New Auto-Interp
Negative Logits
ائ
-0.07
plitude
-0.06
ていない
-0.06
конечно
-0.06
OrderedDict
-0.06
東京
-0.06
그리
-0.06
undy
-0.06
永
-0.06
대표
-0.06
POSITIVE LOGITS
faithful
0.08
uele
0.07
Lunch
0.07
SCE
0.07
(""));↵0.06
effective
0.06
boxes
0.06
Gerry
0.06
ksam
0.06
[+
0.06
Activations Density 0.007%