INDEX

Explanations

round-robin results

The neuron is selectively firing on multi‐word named entities and formal titles (e.g. organization names, tournament names, professional titles, 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

we

-1.87

เรา

-1.73

ått

-1.40

 then

-1.39

ntilde

-1.38

bungen

-1.37

ﾁ

-1.37

 мы

-1.37

garder

-1.36

我們的

-1.34

POSITIVE LOGITS

 الشاشة

1.36

=="")

1.34

 that

1.31

 hablaba

1.30

 porque

1.29

ƅ

1.26

 сообщили

1.25

けっこう

1.24

っちゃう

1.24

 those

1.24

Activations Density 0.288%