INDEX

Explanations

terms related to negative actions such as sabotage and blackmail

oai_token-act-pair · gpt-3.5-turbo

references to acts of sabotage or other malicious actions

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

âĢ¢âĢ¢âĢ¢âĢ¢

-0.73

âĢ¢âĢ¢

-0.71

Vert

-0.71

Haw

-0.71

Lear

-0.70

Wo

-0.69

kens

-0.69

Son

-0.67

Pal

-0.65

erved

-0.65

POSITIVE LOGITS

 sabotage

1.37

 sabot

1.28

 torpedo

0.81

 spoil

0.76

 overshadow

0.74

 siph

0.71

 undermin

0.70

 ration

0.69

inx

0.69

 deval

0.68

Activations Density 0.008%

terms related to negative actions such as sabotage and blackmail

references to acts of sabotage or other malicious actions

No Comments

No Known Activations