INDEX

Explanations

screwed up, fucked up

The neuron fires strongly on the token “screwed,” signaling it detects that specific word (as in “screwed up”).

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

 aislados

-2.20



-2.16

⎆

-2.14

튿

-2.05

⁛

-2.03

玕

-2.00

 debacle

-1.96

 seleccionados

-1.95

廡

-1.95

 الاجتماعية

-1.88

POSITIVE LOGITS

2.31

the

2.27

},

2.19

</h3>

2.03

";

2.00

."

2.00

).

2.00

肫

1.91

if

1.82

1.77

Activations Density 0.003%