INDEX
Explanations
Conditional statements
The neuron fires on modal auxiliaries and evaluative or hypothetical stance markers (e.g. “if,” “would,” “worth,” “reason,” “enough,” “amazing”), essentially picking out conditional and subjective‐evaluation language.
New Auto-Interp
Negative Logits
dev
-0.07
質
-0.07
,obj
-0.07
pulmonary
-0.06
>",
-0.06
anymore
-0.06
copyrighted
-0.06
Cres
-0.06
manual
-0.06
edes
-0.06
POSITIVE LOGITS
.Year
0.07
госп
0.06
rror
0.06
++) ↵
0.06
-feira
0.06
iştir
0.06
cycle
0.06
Venezuela
0.06
JsonRequest
0.06
Scotland
0.06
Activations Density 0.101%