INDEX

Explanations

special characters followed by common terms

The neuron strongly activates on path‐style tokens and separators (e.g. “/”, “dev”, “etc”, “beta”, “html”) and on double-dash option markers (“--”), i.e. filesystem paths and CLI flags.

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

が

-1.25

があり

-1.22

なので

-1.13

фильм

-1.10

inves

-1.09

 époux

-1.09

では

-1.05

пература

-1.02

ebly

-1.01

があるので

-0.99

POSITIVE LOGITS

and

1.39

1.05

 निम्न

1.00

lež

0.99

 träd

0.99

 declining

0.98

no

0.95

噻

0.93

 costuras

0.92

0.91

Activations Density 0.239%