INDEX

Explanations

adjective-noun pairs

The neuron strongly responds to software‐engineering or data‐flow terminology (e.g. “macro,” “child,” “online,” “export,” “upstream”), effectively flagging technical keywords that describe programmatic operations or hierarchical/directional relationships.

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

だけは

-1.22

ほんと

-1.21

ホント

-1.20

以外の

-1.18

凄い

-1.16

おわりに

-1.15

şağı

-1.15

だいぶ

-1.14

kinda

-1.13

 hebben

-1.12

POSITIVE LOGITS

ED

1.13

 ones

1.05

it

1.05

 they

0.99

as

0.98

0.97

dings

0.95

🤙

0.94

paedia

0.94

Activations Density 0.193%