INDEX

Explanations

answering questions

The neuron mainly picks up on meta‐text prompting or labeling answers—especially occurrences of “Answer,” “answer,” and related question/answer cues.

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

 симпто

0.72

difficulty

0.68

 şeyler

0.68

bottle

0.64

แก้ไข

0.63

iknya

0.61

 xlabel

0.61

HLER

0.61

xlabel

0.60

접

0.60

POSITIVE LOGITS

 truthfully

1.03

 answering

1.01

 answer

0.95

swering

0.93

 menjawab

0.92

 answered

0.91

 affirmatively

0.91

 prayers

0.89

 Answer

0.89

 answers

0.85

Activations Density 0.077%