INDEX

Explanations

argues that

The neuron is detecting mid- to high–information “connector” words (e.g. common verbs, prepositions, and discourse markers) that link clauses or signal transitions in the text.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

 אבל

0.80

 hemos

0.75

 antics

0.75

 possibilidades

0.74

 magari

0.73

ו

0.70

 eagles

0.70

 higiene

0.70

 tendremos

0.70

 uffic

0.70

POSITIVE LOGITS

the

1.04

the

0.83

that

0.79

ieren

0.75

ucco

0.69

 Chung

0.69

EN

0.67

ila

0.66

 Consol

0.66

iče

0.66

Activations Density 0.036%