INDEX

Explanations

following instructions or links

The neuron detects procedural instruction language—phrases that tell the reader how to do something (e.g. “follow the instructions,” “as per package directions,” “instructions below”).

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

itself

-1.51

 capucha

-1.40

-1.34

」

-1.31

用意

-1.30

 prisión

-1.30

一人

-1.29

 continúas

-1.29

impulan

-1.27

POSITIVE LOGITS

uri

1.48

 replacements

1.34

してる

1.31

the

1.30

 umane

1.27

чке

1.26

 člověk

1.24

чи

1.23

 respective

1.22

していると

1.21

Activations Density 0.034%