INDEX

Explanations

hadn't + verb

This neuron selectively activates on past‐tense verb tokens (including irregular forms and the “’t” suffix from contractions).

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

 quando

-1.58

not

-1.56

 cuando

-1.54

 όταν

-1.48

khi

-1.43

 عندما

-1.38

 simply

-1.34

 când

-1.33

 when

-1.30

ρας

-1.30

POSITIVE LOGITS

or

1.58

for

1.22

 each

1.16

那一

1.14

 membunuh

1.13

 hver

1.09

 membagikan

1.08

lisa

1.06

那么

1.05

HÀ

1.04

Activations Density 0.011%