INDEX

Explanations

The neuron fires on isolated non‐letter tokens—single-character symbols or digits (e.g. “&”, “<”, “>”, “4”, “5”, etc.) rather than alphabetic words.

punctuation and special characters

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

of

-1.71

の人に

-1.21

 component

-1.19

own

-1.16

 really

-1.14

tion

-1.13

 said

-1.09

useNewUrlParser

-1.09

 whiteColor

-1.09

 försö

-1.09

POSITIVE LOGITS

 всі

1.27

 てき

1.20

 rues

1.18

 اخیر

1.17

Noob

1.16

客气

1.14

cún

1.11

XOXO

1.07

Idiot

1.06

 culte

1.05

Activations Density 0.021%