INDEX

Explanations

instructions

The neuron detects references to instructional or guidance language (e.g., “instructions,” “directions,” “README”).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

ymmetries

0.55

 Selected

0.55

Selected

0.54

ieri

0.53

 Quotes

0.52

кови

0.52

 Themes

0.51

 gând

0.51

 Αγ

0.51

 نظری

0.50

POSITIVE LOGITS

 instructions

2.75

instructions

2.61

Instructions

2.48

 Instructions

2.42

 instrucciones

2.28

 инструкции

2.27

 instruções

2.14

 instruction

2.13

 инструк

2.02

 كيفية

1.97

Activations Density 0.628%