INDEX

Explanations

come, como, placenta

The neuron fires strongly on whole-word or uncommon subword tokens (e.g. library names, foreign‐language words, content words) rather than on frequent English function 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

-3.33

Of

-3.28

However

-3.05

What

-2.83

If

-2.81

</i>

-2.73

For

-2.69

With

-2.64

One

-2.63

Instead

-2.61

POSITIVE LOGITS

promote

2.48

獓

2.44

2.42

 verrez

2.41

 dezelve

2.31

ίδι

2.30

崤

2.30

 quieren

2.30

 namorados

2.27



2.23

Activations Density 0.000%