INDEX

Explanations

terms related to harmful intent or behavior

oai_token-act-pair · gpt-3.5-turbo

terms related to malicious behavior or intent

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

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 1-res-jb

Configuration

jbloom/GPT2-Small-SAEs-Reformatted/blocks.1.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.1.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

ĸļ

-0.87

arist

-0.85

ILA

-0.76

ills

-0.74

hetti

-0.70

 Passage

-0.70

alon

-0.69

Gap

-0.69

quart

-0.69

 Prayer

-0.67

POSITIVE LOGITS

 malicious

1.18

 mischief

0.96

 intent

0.94

 payload

0.87

icious

0.86

ly

0.82

vertising

0.78

 behavi

0.78

 behaviour

0.77

fully

0.76

Activations Density 0.005%

terms related to harmful intent or behavior

terms related to malicious behavior or intent

No Comments

No Known Activations