INDEX

Explanations

tasked with

The neuron fires strongly on little “relation” chunks—common two‐word patterns that link an action or actor to something else (e.g. verb + preposition like “obtained by,” “introduced into,” “to establish,” or pronoun + verb like “you like”).

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

 чтобы

-1.32

*/,

-1.30

ที่

-1.24

),

-1.23

}$.

-1.21

 ↓,

-1.20

»,

-1.18

 который

-1.18

 которые

-1.14

}$,

-1.13

POSITIVE LOGITS

様な

1.38

歉

1.10

 великолеп

1.04

𝑌

1.04

んだろう

0.97

 Please

0.96

it

0.96

んですよね

0.94

 Continued

0.94

精心

0.94

Activations Density 0.199%