INDEX

Explanations

research papers and analysis

The neuron fires strongly on tokens naming the document itself—especially occurrences of “paper” or “article” in academic‐abstract openings.

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

 that

-2.72

of

-2.19

we

-2.11

to

-2.00

 then

-1.95

in

-1.88

 this

-1.56

 kemudian

-1.55

 затем

-1.54

 until

-1.53

POSITIVE LOGITS

 trampa

1.58

 lentement

1.55

 perpé

1.51

 mettent

1.48

 delà

1.47

 blanda

1.45

 rusa

1.43

does

1.43

 fixa

1.40

 honn

1.39

Activations Density 0.152%