INDEX

Explanations

introducing points or explanations

The neuron fires strongly on colons that mark section headers or editorial labels (e.g. “FIX:”, “NOTE:”, “CHANGE:”), i.e. it detects headings/labels ending in a colon.

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

value

1.14

mult

1.10

but

1.08

component

1.08

get

1.07

type

1.07

key

1.07

base

1.07

result

1.06

rest

1.06

POSITIVE LOGITS

 This

2.09

 Although

2.09

 There

2.08

 Despite

2.03

 While

1.98

 When

1.93

 Dette

1.91

 Using

1.89

 這個

1.88

 Seperti

1.88

Activations Density 0.319%