INDEX

Explanations

quoted titles and names

This neuron detects typographic markers of titles—HTML italics tags (“<i>…</i>”) and single or double quotation marks that surround the names of works.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

それが

-1.15

があります

-1.10

-1.09

<bos>

-1.08

もあります

-1.05

 grotes

-1.05

 extravag

-1.04

 alternatives

-1.03

その

-1.03

ures

-1.03

POSITIVE LOGITS

 that

1.23

 what

1.16

pergillus

1.13

all

1.13

not

1.08

one

1.07

obut

1.06

づけ

1.04

 every

1.02

 Thus

1.02

Activations Density 0.025%