INDEX

Explanations

The neuron fires on comparative constructions—especially the word “more” (and its use in “more X” phrases).

more like / than

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

 both

-1.30

などなど

-1.13

both

-1.11

为止

-1.11

 alltid

-1.10

 enough

-1.08

 jamás

-1.08

Tó

-1.05

 نهایت

-1.05

Jawab

-1.05

POSITIVE LOGITS

 than

3.95

 niż

1.97

 instead

1.77

 rather

1.64

than

1.58

而不是

1.55

 like

1.51

 then

1.41

 вместо

1.39

 anstatt

1.38

Activations Density 0.046%