INDEX

Explanations

on

np_max-act · gemini-2.0-flash

This neuron fires most strongly on tokens in the assistant’s generated reply (particularly as you move into the main body after the header), essentially tracking the start and progression of the assistant’s output rather than user text.

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

 grandson

-0.07

ekt

-0.07

 inspection

-0.07

libs

-0.07

 milf

-0.06

	Camera

-0.06

 SceneManager

-0.06

(inst

-0.06

 Tate

-0.06

 Injection

-0.06

POSITIVE LOGITS

_BLOCKS

0.07

ª

0.06

.pattern

0.06

.KeyPress

0.06

]*

0.06

_PEER

0.06

#error

0.06

 Thoughts

0.06

lacağ

0.06

POD

0.06

Activations Density 0.019%

on

This neuron fires most strongly on tokens in the assistant’s generated reply (particularly as you move into the main body after the header), essentially tracking the start and progression of the assistant’s output rather than user text.

No Comments

No Known Activations