INDEX

Explanations

positive closing remarks

The neuron fires most strongly on the key nouns people use when praising or reviewing something (e.g. “card,” “experience,” “service”).

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

ものである

-1.23

STUDIES

-1.07

 There

-1.07

ساء

-1.05

nadequate

-1.02

Ṁ

-1.02

 nonatomic

-1.00

 Hence

-1.00

on

-0.98

-0.97

POSITIVE LOGITS

and

2.41

และ

1.36

1.25

<eos>

1.22

 цей

1.17

 покупке

1.14

 musia

1.12

 jeśli

1.10

↵↵↵↵↵↵↵↵↵↵↵

1.06

 étanche

1.06

Activations Density 0.034%