INDEX

Explanations

performing tasks and operations

The neuron fires strongly on very common function words and high-frequency technical tokens (e.g. of, in, to, is, operations, properties) – essentially spotting “stop-word” or ubiquitous connector tokens in code and technical prose.

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

bouncycastle

-1.35

 after

-1.34

rektor

-1.27

 ویتامین

-1.26

 ваз

-1.23

 because

-1.20

 lecker

-1.20

ὔ

-1.20

 before

-1.20

เพราะ

-1.18

POSITIVE LOGITS

 бого

1.47

 filós

1.37

 掃除

1.27

 hvilke

1.26

output

1.25

 miſ

1.24

bmi

1.22

 Inſ

1.21

хотво

1.20

 パソコン

1.20

Activations Density 0.287%