INDEX

Explanations

code and punctuation delimiters

The neuron fires mainly on numerals and otherwise infrequent or distinctive “content” tokens (i.e. high–IDF words and numbers).

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

álne

-0.92

 Neustadt

-0.91

zetek

-0.90

丨

-0.88

}$\

-0.88

iole

-0.85

ssysteme

-0.85

ofing

-0.84

POSITIVE LOGITS

 entonces

1.05

怎麼辦

1.01

 then

0.97

 compagnia

0.94

Specifik

0.92

multiple

0.91

 تغي

0.91

 list

0.91

 месе

0.91

Kult

0.88

Activations Density 0.126%