INDEX

Explanations

discovered, revealed, understood

The neuron fires on virtually any token once you’re well into a long document—i.e. it detects tokens that occur in later (mid-to-end) positions of the text.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

этому

0.58

 więc

0.57

ですから

0.57

ですし

0.57

 সেজন্য

0.56

そのため

0.55

 Поэтому

0.54

所以在

0.54

 feiern

0.53

ですので

0.53

POSITIVE LOGITS

发现

2.39

發現

2.34

 discovered

2.20

 обнаружи

2.14

discovered

2.06

 discovers

2.02

 выяс

2.02

 พบ

1.98

 발견

1.91

 оказалось

1.90

Activations Density 0.946%