INDEX

Explanations

are described as

The main thing this neuron does is detect very high-frequency, uninformative function words and common tokens (like auxiliary verbs “will,” “are,” “be,” “just,” “as,” commas, etc.).

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

its

-1.76

了一個

-1.56

it

-1.45

它

-1.31

了一个

-1.23

着一个

-1.22

-1.19

 itself

-1.12

 unui

-1.12

 isang

-1.10

POSITIVE LOGITS

 themselves

2.03

 których

1.56

 veritable

1.51

 miniature

1.38

 mere

1.37

 которые

1.32

 replacements

1.31

 которыми

1.29

themselves

1.27

 they

1.25

Activations Density 0.303%