INDEX

Explanations

introducing conditional clauses

The neuron fires on chunks laden with mathematical notation or formulaic expressions (e.g. inline “$…$” math, equations, and related tokens).

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

 xadrez

-1.27

好不好

-1.11

 portavoz

-1.10

 dergleichen

-1.08

 semelhante

-1.05

<bos>

-1.03

 them

-1.02

็ม

-1.02

があれば

-1.01

和你

-1.00

POSITIVE LOGITS

1.39

 both

1.13

 neither

1.10

 België

1.09

$\

1.08

 there

1.08

both

1.06

Kö

1.05

all

1.04

either

1.04

Activations Density 0.029%