INDEX

Explanations

comparative superlatives such as "deadlier" and "ugliest"

oai_token-act-pair · gpt-3.5-turbo

comparative phrases and terms related to the concept of being an outlier or differences in quality

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

irlf

-0.74

resses

-0.64

ellation

-0.64

ricular

-0.64

aiman

-0.63

uties

-0.62

aido

-0.62

Cra

-0.61

thur

-0.61

acha

-0.58

POSITIVE LOGITS

lier

1.04

lihood

0.94

theless

0.88

ppo

0.75

tenance

0.75

fuck

0.75

NetMessage

0.75

enance

0.73

liest

0.72

pox

0.71

Activations Density 0.010%

comparative superlatives such as "deadlier" and "ugliest"

comparative phrases and terms related to the concept of being an outlier or differences in quality

No Comments

No Known Activations