INDEX

Explanations

I'm sorry, but I couldn't identify a specific pattern based on the activations provided for Neuron 4 since the activations appear to cover a wide range of concepts

oai_token-act-pair · gpt-3.5-turbo

phrases centered around self-identity and personal flaws

oai_token-act-pair · gpt-4o-mini Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 7-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.7.hook_resid_pre

Prompts (Dashboard)

24,576 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

torch.float32

Hook Point

blocks.7.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

icion

-0.64

ortium

-0.64

 Canaver

-0.62

UNCH

-0.61

 phased

-0.58

 ratification

-0.56

 rollout

-0.55

endum

-0.55

nesday

-0.55

 staggered

-0.55

POSITIVE LOGITS

thood

0.68

 perce

0.68

 selves

0.68

 oppress

0.67

nature

0.64

 oppressed

0.64

selves

0.64

 unworthy

0.64

 injust

0.63

 dignity

0.62

Activations Density 1.173%

I'm sorry, but I couldn't identify a specific pattern based on the activations provided for Neuron 4 since the activations appear to cover a wide range of concepts

phrases centered around self-identity and personal flaws

No Comments

No Known Activations