INDEX

Explanations

whether or not

The neuron detects direct second-person address and instruction language (e.g. “you … need/can/should …”).

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

繻

-1.40

質な

-1.31

なのですが

-1.27

 订单

-1.24

TÉ

-1.20

but

-1.20

 grupp

-1.20

 siquiera

-1.18

 навіть

-1.16

 senden

-1.15

POSITIVE LOGITS

 these

1.95

 this

1.68

or

1.54

</

1.31

may

1.30

</th>

1.23

 nine

1.22

 enormous

1.21

 which

1.21

 vermutlich

1.19

Activations Density 0.069%