INDEX

Explanations

for the purpose of

This neuron detects high-level framing or discourse‐introducing phrases that signal the author’s meta‐commentary or the scope/purpose of the discussion.

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

 Very

-1.45

via

-1.32

on

-1.28

 Actually

-1.27

orgue

-1.19

寝室

-1.19

 Extremely

-1.18

 sosteniendo

-1.17

 through

-1.15

$};

-1.14

POSITIVE LOGITS

 then

1.57

Mittwoch

1.35

那些

1.34

一提

1.33

1.31

Dienstag

1.30

1.29

ᓚ

1.29

whether

1.28

sowie

1.28

Activations Density 0.132%