INDEX

Explanations

production clients

The neuron responds strongly to very high‐frequency, common “function words” (e.g. the, to, is, and) rather than content‐bearing terms.

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

 employés

-0.82

зія

-0.76

 montagnes

-0.76

 plagioclase

-0.75

 côtes

-0.74

海岸

-0.72

 suspe

-0.72

CAPT

-0.71

Employees

-0.71

eably

-0.70

POSITIVE LOGITS

 clients

2.92

 client

2.80

 applications

2.36

 application

2.13

 caller

2.03

 callers

1.99

 Clients

1.98

clients

1.91

client

1.90

客户端

1.89

Activations Density 0.226%