INDEX

Explanations

To be / not

np_max-act · gemini-2.0-flash

The neuron fires on tokens used in polite apology or uncertainty statements (e.g. “I’m sorry,” “but,” “not sure”), signaling the model is expressing an apology or inability to answer.

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

Tom

-0.06

ruption

-0.06

 Certification

-0.06

COVER

-0.06

 desper

-0.06

asket

-0.06

estation

-0.06

 Ning

-0.06

_DRV

-0.06

 transcription

-0.06

POSITIVE LOGITS

 vlá

0.07

 sehen

0.07

 Σύ

0.07

 mixin

0.06

言葉

0.06

 zwei

0.06

��

0.06

Demand

0.06

 milyon

0.06

 amounts

0.06

Activations Density 0.027%

To be / not

The neuron fires on tokens used in polite apology or uncertainty statements (e.g. “I’m sorry,” “but,” “not sure”), signaling the model is expressing an apology or inability to answer.

No Comments

No Known Activations