INDEX

Explanations

length or count in code

The neuron never fires on any of these code or Q&A tokens—it appears to be effectively “dead” (i.e. it doesn’t detect any pattern in these excerpts).

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

of

-1.54

alnız

-1.44

-1.36

plor

-1.33

gouv

-1.30

濬

-1.28

使って

-1.27

AN

-1.25

并

-1.23

crates

-1.22

POSITIVE LOGITS

 οποίο

1.30

強烈

1.30

事實

1.25

landı

1.23

意見

1.22

极为

1.17

お肉

1.16

Какие

1.16

 котором

1.13

Since

1.12

Activations Density 0.002%