INDEX

Explanations

names, titles, markers followed by descriptions

The neuron fires on backslash-initiated command names (i.e. LaTeX‐style “\command” tokens).

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

 romantis

-1.48

 hunde

-1.47

 sezonu

-1.38

 Ministre

-1.36

 RESTAUR

-1.27

 vieron

-1.26

 berge

-1.25

 Président

-1.22

 Meille

-1.19

 menyen

-1.18

POSITIVE LOGITS

↵↵

1.42

↵↵↵↵

1.27

1.16

յան

1.09

1.08

↵↵↵↵↵↵↵↵

1.07

 nhung

1.05

↵↵↵↵↵↵↵

1.05

↵↵↵↵↵↵

1.04

 coloro

1.03

Activations Density 0.001%