INDEX
Explanations
research articles
This neuron selectively activates on numerical expressions and quantitative‐degree words (e.g. decimal values, “increase,” “decrease,” “greater,” “severe”) indicating measured magnitudes or changes.
New Auto-Interp
Negative Logits
_FAILURE
-0.06
_else
-0.06
ان
-0.06
TPM
-0.06
toupper
-0.06
Flickr
-0.06
epis
-0.06
Colt
-0.06
_ssl
-0.06
Jacket
-0.06
POSITIVE LOGITS
ripsi
0.07
웨디시
0.06
↵
0.06
)};↵
0.06
'];↵↵
0.06
ecc
0.06
(sound
0.06
SHORT
0.06
'}↵↵
0.06
_Row
0.06
Activations Density 0.093%