INDEX

Explanations

don't have to be

This neuron responds to infinitive constructions, especially the “to” + base-form verb sequences (e.g. “to be,” “to doing,” “rather than” clauses).

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 nejen

-1.43

 modestly

-1.23

 suitably

-1.20

 некото

-1.19

不仅

-1.16

 somewhat

-1.15

not

-1.14

 competent

-1.13

 provocando

-1.12

both

-1.11

POSITIVE LOGITS

 necessarily

1.66

 huge

1.66

 fancy

1.63

 hoeft

1.41

fancy

1.32

 elaborate

1.31

 extensive

1.30

elabor

1.27

 كبير

1.27

necessarily

1.27

Activations Density 0.111%