INDEX

Explanations

code symbols

np_max-act · gemini-2.0-flash

The neuron activates on the protocol keywords that mark steps in tool‐using reasoning (e.g. “Thought,” “Action,” “Action Input,” “Observation,” “calculator,” etc.). It’s detecting those chain‐of‐thought/tool‐invocation labels.

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

 stain

-0.09

_PAGE

-0.06

_Content

-0.06

 gratuite

-0.06

 تلویزیون

-0.06

Rank

-0.06

šel

-0.06

uell

-0.06

opened

-0.06

 COLUMN

-0.06

POSITIVE LOGITS

这么

0.07

::$_

0.07

】↵↵

0.06

 '':↵

0.06

ш

0.06

%H

0.06

 QColor

0.06

 krit

0.06

 випадку

0.06

 readFile

0.06

Activations Density 0.006%

code symbols

The neuron activates on the protocol keywords that mark steps in tool‐using reasoning (e.g. “Thought,” “Action,” “Action Input,” “Observation,” “calculator,” etc.). It’s detecting those chain‐of‐thought/tool‐invocation labels.

No Comments

No Known Activations

code symbols

The neuron activates on the protocol keywords that mark steps in tool‐using reasoning (e.g. “Thought,” “Action,” “Action Input,” “Observation,” “calculator,” etc.). It’s detecting those chain‐of‐thought/tool‐invocation labels.

No Comments

No Known Activations