INDEX

Explanations

special cases

The neuron fires strongly on mid‐sentence capitalized tokens—i.e. proper names or brand-style terms.

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

 when

-2.25

 that

-1.93

 their

-1.89

 what

-1.75

if

-1.63

had

-1.52

at

-1.48

 during

-1.46

 almost

-1.45

so

-1.45

POSITIVE LOGITS

is

1.48

ReadByte

1.41

︲

1.39

 Atención

1.38

 Públicas

1.32

唁

1.30

 sempre

1.30

 opět

1.27

 ジップ

1.25

1.23

Activations Density 0.245%