INDEX

Explanations

Yet another, Frankenfield

The neuron is essentially detecting numeric literals (digit sequences) in the text.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 from

-1.61

who

-1.47

 with

-1.43

if

-1.23

 there

-1.22

all

-1.19

get

-1.16

you

-1.12

of

-1.12

 this

-1.09

POSITIVE LOGITS

Gön

1.45

 betreffende

1.19

 bekämp

1.18

 バッジ

1.17

さて

1.16

 conferma

1.15

 kağıdı

1.15

 écouter

1.13

Handlung

1.13

Abend

1.12

Activations Density 0.001%