INDEX

Explanations

Chief Keef

The neuron fires on non‐lexical formatting and reference tokens—things like numeric labels, placeholder codes (e.g. “##”, “STR”), section or figure numbers, and file-name fragments (e.g. “text”, “png”) rather than ordinary words.

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

to

-1.67

 from

-1.19

for

-1.13

">*

-0.93

нью

-0.90

on

-0.87

so

-0.85

 that

-0.84

一旁

-0.84

POSITIVE LOGITS

 nôtre

1.16

 cocción

1.09

푉

1.06

 diferite

1.05

 criptomonedas

1.05

Departure

1.04

Abbildung

1.04

 puits

1.03

 asegurarse

1.03

 colgar

1.03

Activations Density 0.000%