INDEX

Explanations

haven't

The neuron consistently fires on the token “haven” (as in the contraction “haven’t”), i.e. it detects the auxiliary “have” stem in negative constructions.

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

 жовтня

-3.16

ܜ

-2.86

 вересня

-2.66

璆

-2.44

 червня

-2.44

and

-2.44

 грудня

-2.42

噚

-2.39

 verster

-2.39

禑

-2.34

POSITIVE LOGITS

2.80

2.50

}(

2.41

’

2.38

 haven

2.34

時間も

2.33

 Didn

2.27

2.25

2.23

<bos>

2.22

Activations Density 0.006%