INDEX
Explanations
punctuation
This neuron strongly activates on non‐story “metadata” tokens—things like image captions or labels, abbreviations, and other structural/formatting markers rather than ordinary narrative words.
New Auto-Interp
Negative Logits
.Redis
-0.06
Db
-0.06
check
-0.06
Deployment
-0.06
aseña
-0.06
Typography
-0.06
_ROWS
-0.06
pics
-0.06
ouden
-0.06
glyphs
-0.06
POSITIVE LOGITS
bach
0.06
pacing
0.06
alım
0.06
_fac
0.06
profit
0.06
土
0.06
炉
0.06
.prefix
0.06
itory
0.06
งาน
0.06
Activations Density 0.168%