INDEX

Explanations

Mah-starter names and places

This neuron activates on any token containing the character sequence “Mah,” i.e. words or names beginning with “Mah.”

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

colgante

-3.69

of

-3.45

 upsetting

-2.89

 Then

-2.78

 stirs

-2.77

 strikingly

-2.67

while

-2.66

見ても

-2.61

 vastly

-2.59

 hará

-2.58

POSITIVE LOGITS

ت

2.73

鮃

2.69

Going

2.47

焢

2.47

殞

2.44

Almost

2.42

Even

2.41

人が

2.38

 maquillage

2.36

Getting

2.33

Activations Density 0.004%