INDEX

Explanations

incidents and accidents

The neuron detects temporal and locative event-framing words (e.g. “after,” “at,” “on,” “following,” “occurred,” “took place”) that introduce or locate reported actions.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

for

-1.32

 that

-1.27

erobic

-1.20

event

-1.16

常常

-1.16

 always

-1.15

}(),

-1.12

now

-1.11

ennale

-1.09

腓

-1.08

POSITIVE LOGITS

on

1.32

 estimado

1.22

手续

1.19

 aproximado

1.12

 viaj

1.11

 izquier

1.10

 Sabtu

1.07

 ошибка

1.07

 atacado

1.06

 árbitro

1.06

Activations Density 0.053%