INDEX

Explanations

numbers before technical terms

The neuron fires most strongly on numeric specification phrases—i.e. tokens that combine a number (digit or word) with a unit, count, or technical term (often hyphenated).

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

-1.68

 another

-1.66

另一個

-1.44

另一个

-1.35

を変える

-1.27

another

-1.27

ーと

-1.27

cestry

-1.25

甚麼

-1.20

 haven

-1.19

POSITIVE LOGITS

𝗲

1.53

utivo

1.40

 Various

1.40

 verwenden

1.34

 esetén

1.31

🦣

1.30

 biztons

1.29

geç

1.27

ğraf

1.25

ЛА

1.25

Activations Density 0.150%