INDEX

Explanations

programming languages and software

This neuron detects capitalized named entities—especially technology names (like Java, C++, Linux, Firefox, Chrome), event titles (Main Event), and similar proper nouns.

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

的同时

-1.13

なんだけど

-1.13

for

-1.10

 ください

-1.09

and

-1.07

 zwar

-1.06

 daar

-1.05

 عليك

-1.02

 mnoho

-1.00

dampak

-0.99

POSITIVE LOGITS

々な

1.13

 ours

1.11

Instead

1.07

quilo

1.04

وازم

1.02

 proposés

1.02

 Rivière

1.01

vece

1.01

 فقط

1.00

 when

0.99

Activations Density 0.217%