INDEX

Explanations

logical consequences from prior statements

The neuron fires on logical‐implication cues in math statements—e.g. “follows that” or “implies.”

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

</h1>

-1.98

-1.67

 bulunabilir

-1.63

事が

-1.57

CEAN

-1.56

-1.55

 verdaderos

-1.52

為に

-1.50

↵↵

-1.48

"-

-1.48

POSITIVE LOGITS

</em>

2.09

two

1.92

 first

1.89

 three

1.80

 almost

1.69

 with

1.66

 several

1.63

any

1.63

 both

1.60

one

1.59

Activations Density 0.006%