INDEX

Explanations

terms related to security breaches or instances where something has been negatively impacted

oai_token-act-pair · gpt-3.5-turbo

occurrences of the word "compromised" and its variations

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

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

umber

-0.73

els

-0.69

 Moons

-0.68

fin

-0.68

ULE

-0.65

umn

-0.65

fill

-0.62

Sam

-0.61

ibli

-0.60

gas

-0.60

POSITIVE LOGITS

 compromised

3.72

 breached

1.84

 compromising

1.77

 weakened

1.72

 hijacked

1.65

 tainted

1.56

 undermined

1.53

 contaminated

1.47

 corrupted

1.47

 infiltrated

1.47

Activations Density 0.016%

terms related to security breaches or instances where something has been negatively impacted

occurrences of the word "compromised" and its variations

No Comments

No Known Activations