INDEX

Explanations

concepts across languages

The neuron strongly activates on longer, more “formal” or “multisyllabic” content words (e.g. professional, differences, processed, experience), i.e. it flags tokens that are lengthy, technical, or abstract.

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

 where

-0.94

 finales

-0.92

 centrale

-0.92

growth

-0.90

 from

-0.89

...>

-0.88

 måte

-0.88

 colega

-0.87

 clara

-0.87

 really

-0.87

POSITIVE LOGITS

 adhesivo

1.09

 willkommen

1.07

madı

1.05

 vérifier

1.05

 però

1.05

 ahorro

1.05

 they

1.04

corrhi

1.04

 však

1.02

表面

1.01

Activations Density 0.035%