INDEX

Explanations

dates and locations

The neuron fires on metadata tokens in press‐release boilerplate—dates, numbers, and wire‐service tags.

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

 sucedido

-1.02

了一会儿

-1.00

 stát

-0.99

Pře

-0.99

 straips

-0.97

ఝ

-0.94

GMENT

-0.94

 Añade

-0.92

 siguran

-0.92

 передний

-0.92

POSITIVE LOGITS

 ErrInvalid

1.11

 économiques

0.94

子が

0.91

During

0.88

しかし

0.88

 должен

0.87

 Willen

0.85

“

0.85

Following

0.84

 результа

0.83

Activations Density 0.021%