INDEX

Explanations

words related to negative behaviors and dishonesty

oai_token-act-pair · gpt-3.5-turbo

terms related to deception, dishonesty, and disconnection

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

PlotsExplanationShow Test FieldDefault Test Text

IFrame

Link

Not in Any Lists

No Comments

Negative Logits

Reviewer

-0.93

Ear

-0.78

 Hits

-0.73

Redditor

-0.65

DragonMagazine

-0.62

 Mara

-0.62

Els

-0.62

 Ital

-0.62

 antioxid

-0.61

Enlarge

-0.61

POSITIVE LOGITS

liction

1.04

lement

0.96

cipline

0.90

ention

0.90

icating

0.90

legate

0.89

arer

0.89

isive

0.88

iencies

0.85

igmat

0.84

Activations Density 0.067%

words related to negative behaviors and dishonesty

terms related to deception, dishonesty, and disconnection

No Comments

No Known Activations