INDEX

Explanations

rules, method, mode, difficulties

The neuron fires most strongly on sentences that state recommendations or normative instructions (i.e. advisory, “should”/“would”-style guidance).

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

 ваз

-1.24

ญี่ป

-1.19

 Кубка

-1.16

 télécommande

-1.16

 शरीर

-1.13

När

-1.13

 прокура

-1.11

スケジュール

-1.09

 didn

-1.08

 ингреди

-1.08

POSITIVE LOGITS

ImageIO

1.51

we

1.44

 users

1.37

 incluso

1.35

 existing

1.30

 solution

1.28

 would

1.26

 proposal

1.25

 дѣ

1.18

 Faça

1.16

Activations Density 0.072%