INDEX

Explanations

consideramos, considera, considerar

The neuron is signaling how far into the document you are—it activates progressively more on later tokens (i.e. it’s a “position‐in‐sequence” detector).

New Auto-Interp

Configuration

Prompts (Dashboard)

392,802 prompts, 256 tokens each

Dataset (Dashboard)

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

Negative Logits

<eos>

1.00

 Covid

0.92

 🥰

0.92

 https

0.88

0.85

0.81

Covid

0.80

 Another

0.79

 💪

0.79

 🔥

0.79

POSITIVE LOGITS

<unused2140>

0.79

--“

0.77

、『

0.68

：「

0.66

：“

0.66

:”

0.66

सुद्धा

0.63

****",

0.63

—“

0.62

पेक्षा

0.61

Activations Density 0.057%