INDEX

Explanations

email addresses and social media handles

oai_token-act-pair · gpt-3.5-turbo

Triggered by @bot

mentions of usernames or handles on social media platforms

oai_token-act-pair · gpt-4o-mini

Triggered by @bot

New Auto-Interp

Top Features by Cosine Similarity

Comparing With GPT2-SMALL @ 3-att-kk

Configuration

ckkissane/attn-saes-gpt2-small-all-layers/gpt2-small_L3_Hcat_z_lr1.20e-03_l19.00e-01_ds24576_bs4096_dc1.00e-06_rsanthropic_rie25000_nr4_v9.pt

Prompts (Dashboard)

36,864 prompts, 128 tokens each

Dataset (Dashboard)

Skylion007/openwebtext

Features

24,576

Data Type

float32

Hook Name

blocks.3.attn.hook_z

Hook Layer

Architecture

standard

Context Size

128

Dataset

Skylion007/openwebtext

Activation Function

relu

Embeds

IFrame

Link

Head Attr Weights

0:0.26

1:0.07

2:0.07

3:0.04

4:0.05

5:0.05

6:0.09

7:0.05

8:0.08

9:0.04

10:0.12

11:0.04

Negative Logits

itably

-1.41

.—

-1.40

etti

-1.37

iasis

-1.30

®

-1.25

 Anglic

-1.23

.」

-1.23

©

-1.21

adoes

-1.20

aughtered

-1.17

POSITIVE LOGITS

senal

1.36

VERT

1.28

involved

1.22

gap

1.21

 inaction

1.21

onday

1.20

REDACTED

1.20

OND

1.19

ARCH

1.17

committee

1.17

Activations Density 0.019%

email addresses and social media handles

mentions of usernames or handles on social media platforms

No Known Activations