INDEX

Explanations

quantifying states and actions

The neuron fires most strongly on words naming procedures or actions—e.g. “diversify,” “workflow,” “precision,” “integration,” “load balancing”—i.e. terms that denote methodological steps or processes.

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

Koordinaten

-1.31

اپ

-1.28

Frequently

-1.27

 szklan

-1.26

 "'";

-1.25

 WEDNESDAY

-1.24

-1.23

 "'");

-1.23

⏰

-1.20

POSITIVE LOGITS

我真的

1.52

1.34

for

1.28

1.27

就得

1.24

いかがでしたか

1.21

 najdete

1.20

Activations Density 0.067%