INDEX

Explanations

if conditions and data

The neuron is essentially a sentinel for the boundary/end‐of‐span marker—it lights up strongly at sequence endings.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 вле

0.56

 doorways

0.56

 బంగ

0.55

ូច

0.54

 SPMs

0.54

 olen

0.53

бята

0.52

paths

0.52

 част

0.52

 вмі

0.52

POSITIVE LOGITS

 Through

0.61

Through

0.61

↵↵

0.57

 through

0.54

 Instit

0.54

 Para

0.53

 Annot

0.52

依次

0.52

 Gori

0.52

Ann

0.51

Activations Density 0.027%