INDEX

Explanations

sentence endings

The neuron strongly activates on the single most salient “topic” word in each numbered list item or heading—i.e. it picks out the key content word that labels each step in a list.

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

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.88

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.84

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.83

↵↵↵↵↵↵↵↵

0.82

↵↵↵↵↵↵↵↵↵↵↵↵↵

0.82

↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.82

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.82

↵↵↵↵↵↵↵↵↵↵↵

0.81

↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵↵

0.81

以及

0.80

POSITIVE LOGITS

But

1.06

 근데

1.02

 Nhưng

0.96

But

0.96

 Anyway

0.93

 There

0.88

 Надо

0.88

Anyway

0.87

but

0.86

but

0.86

Activations Density 0.286%