INDEX

Explanations

if conditions

The neuron fires on conditional/instructional clauses, especially “if you…” phrases that introduce steps or prerequisites.

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

 Gefahren

-1.10

看了一眼

-1.00

 siè

-0.96

 ROBER

-0.94

 politiche

-0.94

ghted

-0.91

 Palestina

-0.89

 berbeda

-0.88

 juſ

-0.88

叹了口气

-0.88

POSITIVE LOGITS

 still

1.42

you

1.30

实在

1.29

 cannot

1.24

 really

1.20

不行

1.14

 fails

1.10

それでも

1.07

get

1.07

 after

1.03

Activations Density 0.018%