INDEX

Explanations

technical topics and modifiers

This neuron is picking out words that occur in answer (A:) posts rather than in the question, effectively flagging tokens from the solution/response section.

New Auto-Interp

Configuration

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Negative Logits

 nhưng

-1.29

當然

-1.16

 ولكن

-1.15

だったが

-1.13

你的

-1.05

 consentement

-1.05

 azonban

-1.03

 แต่

-0.99

え

-0.99

いますが

-0.98

POSITIVE LOGITS

and

1.64

or

1.64

as

1.34

 which

1.20

etc

1.15

ʳ

1.13

 Uniwers

1.09

szt

1.03

 bolsas

1.02

 perhaps

0.97

Activations Density 0.056%