INDEX

Explanations

negated concepts

The neuron consistently lights up on longer, multi-syllable, formal or technical vocabulary (e.g. “encryption,” “reversed,” “random,” “variable,” “prior,” “ideologies”), so it’s detecting high-register abstract/technical 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

 from

-1.65

and

-1.31

 articol

-1.22

fordshire

-1.16

 pemas

-1.13

 while

-1.13

=“

-1.13

ta

-1.12

知らせ

-1.09

藹

-1.09

POSITIVE LOGITS

contentValues

1.36

1.34

找到了

1.33

 erziel

1.31

 psicológico

1.31

ayaquil

1.24

 silencioso

1.23

GYPT

1.21

 вашего

1.18

 способов

1.17

Activations Density 0.089%