INDEX

Explanations

classes, courses, groups

The neuron fires most strongly on tokens that occur near the middle of the input span, i.e. it’s acting as a crude “position‐in‐document” detector for mid-context tokens.

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

 Hür

-0.96

SCENE

-0.80

 lijkt

-0.79

 учитель

-0.77

シルエット

-0.75

Kategorie

-0.73

 デニム

-0.73

冰冰

-0.72

oreille

-0.72

 infertility

-0.72

POSITIVE LOGITS

 class

1.25

 course

1.23

 courses

1.15

 group

1.14

 score

1.05

 Courses

1.02

 Group

1.01

 scores

1.00

Group

0.97

 Scores

0.97

Activations Density 0.011%