INDEX

Explanations

let salted

This neuron lights up on the first significant content word (rather than function words) at the start of a new text segment or sentence.

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

すっ

-1.39

킁

-1.30

྄

-1.27

謦

-1.26

became

-1.24

튿

-1.23

 trám

-1.21

hav

-1.20

regarding

-1.19

 potrze

-1.17

POSITIVE LOGITS

 from

1.23

as

1.22

 because

1.20

by

1.16

on

1.14

at

1.13

 after

1.13

 despite

1.10

 comes

1.09

 through

1.06

Activations Density 0.238%