INDEX

Explanations

The neuron fires on short capitalized name fragments (especially two-letter prefixes) in email header recipient lists, effectively spotting tokens that are parts of people’s names.

names of people, places, and things

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

общение

-1.15

 niveau

-1.10

שלום

-1.08

 ۱۸

-1.07

Tils

-1.05

 sommeil

-1.04

 gentes

-1.03

 ditemukan

-1.02

طرز

-1.02

 zapatos

-1.00

POSITIVE LOGITS

),),

1.43

際には

1.31

 HttpNotFound

1.27

so

1.25

 DELLE

1.23

].(

1.21

 Hauses

1.16

たびに

1.14

struzione

1.13

際に

1.13

Activations Density 0.005%