INDEX

Explanations

terms related to harm, including physical, emotional, or potential danger

oai_token-act-pair · gpt-3.5-turbo

mentions of harm, particularly in relation to various contexts and populations

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 0-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.0.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.0.hook_resid_pre

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Hook Point Layer

Activation Function

relu

Embeds

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

umen

-0.67

uren

-0.65

ometer

-0.64

READ

-0.63

enhagen

-0.63

Fan

-0.61

aten

-0.60

Completed

-0.59

 filled

-0.59

Base

-0.58

POSITIVE LOGITS

 harm

3.87

 harms

2.86

 harmed

2.15

 Harm

2.06

harm

1.98

 harming

1.94

 hurt

1.65

 damage

1.61

 harmful

1.57

 endanger

1.53

Activations Density 0.021%

terms related to harm, including physical, emotional, or potential danger

mentions of harm, particularly in relation to various contexts and populations

No Comments

No Known Activations