INDEX

Explanations

The main thing this neuron does is detect forms of “improve” (e.g., improve, improving, improvement).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

ティ

1.05

 adecuado

1.01

吗

0.93

 Informe

0.92

Namun

0.88

阳光

0.88

 realizar

0.87

 ondas

0.87

 Tinggi

0.87

 ingresar

0.86

POSITIVE LOGITS

 vastly

1.05

 swoją

1.00

 considerably

0.97

 Improving

0.97

 Improve

0.95

 sensiblement

0.95

 단순히

0.95

 vaše

0.92

 greatly

0.91

 improves

0.90

Activations Density 1.303%