INDEX

Explanations

The neuron strongly activates on uppercase or capitalized abbreviations and acronym‐style tokens (e.g. SIM, JSP, ecm, tru, Ph, reporting).

code syntax or identifiers

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

を

-1.16

もある

-1.05

-1.04

というか

-1.03

 даже

-1.00

差異

-0.97

立場

-0.97

 especially

-0.96

gged

-0.96

Especially

-0.95

POSITIVE LOGITS

fordshire

1.32

 verschill

1.20

Wikimedia

1.20

)});

1.16

)})

1.16

 voldoende

1.16

)].

1.14

readFileSync

1.13

and

1.13

 consultato

1.13

Activations Density 0.053%