INDEX

Explanations

as compared to

This neuron fires on comparative constructions involving “compared (to/with)” plus a numerical value (e.g. “compared to 27.5%”).

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

 powiedział

-1.95

);

-1.94

柃

-1.91

汌

-1.78

鯪

-1.76

 hundred

-1.74

ウェブサイト

-1.73

 всіх

-1.73

何度も

-1.72

 دے

-1.70

POSITIVE LOGITS

the

2.31

2.06

Not

1.86

 Here

1.85

 podría

1.84

 they

1.82

Its

1.81

 REALLY

1.81

 Similarly

1.81

 Instead

1.80

Activations Density 0.003%