INDEX

Explanations

granted, admittedly, albeit

This neuron detects concessive discourse markers (e.g. “Granted,” “Admittedly,” “Fairness,”) that introduce qualifying or concessionary statements.

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

-2.03

</b>

-1.88

</i>

-1.82

-1.81

-1.71

and

-1.63

 August

-1.55

 Unternehmer

-1.55

</th>

-1.50

</u>

-1.50

POSITIVE LOGITS

 they

2.17



1.90

although

1.88

Granted

1.88

wasn

1.72

 these

1.71

They

1.68

ilangkan

1.67

 there

1.64

𝐅

1.64

Activations Density 0.015%