INDEX

Explanations

had been

This neuron strongly activates on the token “been,” i.e. the auxiliary verb “been” used in perfect‐tense constructions.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

jarse

-1.05

 while

-1.04

 beginning

-0.98

HttpStatusCode

-0.97

ícias

-0.94

yalar

-0.93

 начина

-0.92

خطيط

-0.91

 mentre

-0.91

ющи

-0.91

POSITIVE LOGITS

 existence

1.28

 around

1.27

in

1.20

 exists

1.17

1.08

 věci

1.08

 exist

1.06

 существо

1.02

 existed

0.97

 involved

0.94

Activations Density 0.053%