INDEX

Explanations

issues, risks, causes

The neuron is primarily picking out content-bearing nouns that serve as heads of noun phrases (e.g. “state(s),” “cause,” “hazards,” “cases,” “channels,” “tidbit,” “morsels,” “final”).

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

-1.64

not

-1.57

 more

-1.56

 many

-1.31

$-

-1.30

It

-1.28

-1.27

氏

-1.27

 substantial

-1.27

POSITIVE LOGITS

HASAN

1.45

ICIA

1.43

 rempliss

1.40

鷙

1.39

 assurent

1.37

erva

1.34

 йыл

1.34

 coklat

1.34

蝽

1.33

 embl

1.32

Activations Density 0.073%