INDEX

Explanations

questions asking for help

The neuron is specifically detecting the “Q:” label that marks the start of a question in the Q&A transcript format.

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

tedy

-1.32

 histó

-1.20

 manzana

-1.20

⠀⠀⠀⠀

-1.19

 سپس

-1.15

 motoru

-1.14

 dekol

-1.14

 sokak

-1.13

 biele

-1.13

 cuivre

-1.13

POSITIVE LOGITS

is

1.44

Can

1.27

 chỉ

1.11

 шесть

1.09

do

1.07

can

1.06

 they

1.04

 does

1.02

Are

1.02

to

1.01

Activations Density 0.150%