INDEX

Explanations

words followed by specific nouns or content

The main thing this neuron does is detect strongly capitalized content words and punctuation in headings or titles (e.g. proper nouns and section‐title tokens).

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

 incrí

-1.88

も多く

-1.80

 Dinas

-1.70

 déplo

-1.68

 علاوه

-1.61

 ines

-1.60

 ché

-1.59

着自己的

-1.56

 incess

-1.54

 dilapidated

-1.54

POSITIVE LOGITS

 where

2.02

and

1.93

 using

1.87

one

1.78

 only

1.77

 after

1.77

 make

1.76

 work

1.74

</h2>

1.72

</h1>

1.67

Activations Density 0.182%