INDEX

Explanations

the most/best

The neuron primarily detects common English determiners and quantifiers (especially “the” and “one”).

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

 loudest

-1.39

Best

-1.35

 cleanest

-1.34

 happiest

-1.29

 weakest

-1.28

 migliore

-1.25

 melhor

-1.23

 brightest

-1.22

Lowest

-1.20

 slowest

-1.20

POSITIVE LOGITS

 most

1.30

stech

1.06

Bibliografie

0.98

㌧

0.94

 BIBLIO

0.93

 slipper

0.90

сці

0.85

 ピン

0.84

渽

0.84

Источник

0.84

Activations Density 0.078%