INDEX

Explanations

avoiding problems or negative outcomes

The neuron flags words that describe unintended or harmful mistakes or misuse (e.g. errors, accidental use, stolen).

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

どうやら

-1.52

 feront

-1.50

 semblent

-1.49

 搞笑

-1.48

 récente

-1.45

}$}

-1.45

 devraient

-1.44

 Parece

-1.43

 pourront

-1.43

 seraient

-1.41

POSITIVE LOGITS

by

1.78

 another

1.67

of

1.66

or

1.60

our

1.37

我们

1.35

its

1.28

 like

1.27

我們

1.27

 using

1.24

Activations Density 0.176%