INDEX

Explanations

urls and paths

The neuron fires on HTTP request‐line tokens (e.g. GET, HEAD, “request:”, HTTP/version) in server logs.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 suspense

0.60

 stap

0.59

 staple

0.59

、

0.55

minipage

0.54

 constitutive

0.54

 peper

0.53

 morale

0.52

 Staples

0.52

 keyed

0.52

POSITIVE LOGITS

"/

1.62

'/

1.55

("/

1.50

('/

1.48

="/

1.38

("/

1.34

`/

1.33

"/

1.30

'/

1.29

='/

1.23

Activations Density 0.017%