INDEX

Explanations

This neuron detects speaker‐attribution tokens (names, titles, and the word “said”/“spokesman,” etc.) that introduce or tag quoted speech.

said, reporting speech

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

 These

-1.16

The

-1.10

 Then

-1.08

 what

-1.04

 Thus

-0.99

 prêmio

-0.96

Non

-0.95

Now

-0.94

 What

-0.93

 Both

-0.93

POSITIVE LOGITS

 said

1.85

Martina

1.20

 explicó

1.07

하였다

1.05

menü

1.05

 gethan

1.04

 deß

1.03

他说

1.01

1.00

 seyn

0.97

Activations Density 0.006%