INDEX

Explanations

encouraging pos<confidence>itive</confidence> outcomes

The neuron strongly activates on abstract action words—especially gerunds and nominalized verbs (e.g. “being,” “using,” “experimentation,” “membership”).

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

or

-1.50

 this

-1.34

-1.21

-1.20

 four

-1.18

öf

-1.13

انا

-1.13

two

-1.12

шко

-1.07

 или

-1.05

POSITIVE LOGITS

 Encourage

1.41

 schnee

1.28

 همان

1.22

 télécharger

1.20

posições

1.20

厛

1.19

iress

1.18

 abbildung

1.17

 semelh

1.15

Encourage

1.15

Activations Density 0.110%