INDEX

Explanations

Maintaining a role/persona

np_max-act · gemini-2.0-flash

This neuron detects meta instructions or reminders not to break character (e.g. “do not break character”).

oai_token-act-pair · o4-mini Triggered by @xinyanhu8

New Auto-Interp

Configuration

andyrdt/saes-llama-3.1-8b-instruct/resid_post_layer_11/trainer_1

Dataset (Dashboard)

Various

Features

131,072

Data Type

float32

Hook Name

blocks.11.hook_resid_post

Architecture

standard

Context Size

1,024

Dataset

monology/pile-uncopyrighted

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

レイ

-0.07

Ops

-0.07

/c

-0.07

/C

-0.07

Chief

-0.07

Tar

-0.07

.AppSettings

-0.06

dit

-0.06

	Runtime

-0.06

Cr

-0.06

POSITIVE LOGITS

 libertarian

0.07

 suburban

0.06

 başka

0.06

htmlspecialchars

0.06

 meticulously

0.06

>;↵

0.06

 clinic

0.06

[key

0.05

िशत

0.05

 risky

0.05

Activations Density 0.005%

Maintaining a role/persona

This neuron detects meta instructions or reminders not to break character (e.g. “do not break character”).

No Comments

No Known Activations