INDEX

Explanations

The neuron consistently lights up on discourse-structuring words and quantifiers—contrastive markers like “but,” “what,” and count or degree words such as “many,” “more,” “have,” and “had.”

but contrasted with new

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

 both

-1.10

but

-1.09

 Already

-0.94

 bereits

-0.92

 although

-0.91

 allerede

-0.90

if

-0.88

時点で

-0.86

 similarly

-0.85

すでに

-0.84

POSITIVE LOGITS

 anew

1.46

新的

1.46

 mới

1.44

 νέ

1.38

new

1.27

now

1.20

 جدید

1.14

改めて

1.09

 جديد

1.08

NEW

1.06

Activations Density 0.070%