INDEX

Explanations

code and foreign words

The neuron responds to subword fragments (in particular 3–4 character WordPiece tokens) rather than whole words.

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

-2.91

-2.38

ise

-2.36

 faſt

-2.31

-2.25

 высокая

-2.22

at

-2.20

 possono

-2.19

 ドア

-2.14

 confronti

-2.13

POSITIVE LOGITS

It

2.58

 There

2.50

 During

2.45

2.38

 или

2.22

 Diskussion

2.16

又是一

2.13

↵↵

2.11

In

1.99

Activations Density 0.001%