INDEX

Explanations

for loop iteration

The neuron never fires on any ordinary code or prose tokens here—it appears to be effectively “dead” in these examples and does not detect any recurring pattern.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

This

-1.52

for

-1.45

２

-1.42

doing

-1.41

during

-1.37

資料來源

-1.36

being

-1.35

started

-1.33

These

-1.32

-1.30

POSITIVE LOGITS

rschein

1.52

 municípios

1.36

掛け

1.34

 electrica

1.34

 hábit

1.28

ąć

1.27

 площадь

1.27

￠

1.24

but

1.23

 cantantes

1.23

Activations Density 0.002%