INDEX

Explanations

works in a certain way

The neuron is basically flagging common “scaffolding” words and tokens—i.e. high-frequency function words and other ubiquitous terms that structure ordinary English sentences.

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

tyd

-1.28

 gangen

-1.23

the

-1.22

 skrevet

-1.21

 щоб

-1.21

 kanskje

-1.20

頎

-1.20

Spes

-1.16

 hvordan

-1.13

in

-1.13

POSITIVE LOGITS

1.66

);

1.34

>();

1.28

unk

1.27

ть

1.27

и

1.21

base

1.21

↵↵

1.20

OR

1.19

Activations Density 0.105%