Thoughts about the Mechanis­tic In­ter­pretabil­ity Challenge #2 (EIS VII #2)

RGRGRGJul 28, 2023, 8:44 PM
24 points
5 comments20 min readLW link

Be­cause of Lay­erNorm, Direc­tions in GPT-2 MLP Lay­ers are Monosemantic

ojorgensenJul 28, 2023, 7:43 PM
13 points
3 comments13 min readLW link

When can we trust model eval­u­a­tions?

evhubJul 28, 2023, 7:42 PM
166 points
10 comments10 min readLW link1 review

Yes, It’s Sub­jec­tive, But Why All The Crabs?

johnswentworthJul 28, 2023, 7:35 PM
250 points
15 comments6 min readLW link

Semaglu­tide and Muscle

5houtJul 28, 2023, 6:36 PM
15 points
14 comments5 min readLW link

Dou­ble Crux in a Box

ScrewtapeJul 28, 2023, 5:55 PM
8 points
3 comments1 min readLW link

Gra­di­ent de­scent might see the di­rec­tion of the op­ti­mum from far away

Mikhail SaminJul 28, 2023, 4:19 PM
70 points
13 comments4 min readLW link

Progress links di­gest, 2023-07-28: The deca­dent op­u­lence of mod­ern capitalism

jasoncrawfordJul 28, 2023, 2:36 PM
16 points
3 comments3 min readLW link
(rootsofprogress.org)

AI Aware­ness through In­ter­ac­tion with Blatantly Alien Models

VojtaKovarikJul 28, 2023, 8:41 AM
7 points
5 comments3 min readLW link

You don’t get to have cool flaws

Neil Jul 28, 2023, 5:37 AM
80 points
25 comments2 min readLW link3 reviews

Re­duc­ing syco­phancy and im­prov­ing hon­esty via ac­ti­va­tion steering

Nina PanicksseryJul 28, 2023, 2:46 AM
122 points
18 comments9 min readLW link1 review

Mech In­terp Puz­zle 2: Word2Vec Style Embeddings

Neel NandaJul 28, 2023, 12:50 AM
41 points
4 comments2 min readLW link

ETFE windows

bhauthJul 28, 2023, 12:46 AM
31 points
4 comments2 min readLW link
(www.bhauth.com)

A Short Memo on AI In­ter­pretabil­ity Rain­bows

scasperJul 27, 2023, 11:05 PM
18 points
0 comments2 min readLW link

Pul­ling the Rope Side­ways: Em­piri­cal Test Results

Daniel KokotajloJul 27, 2023, 10:18 PM
61 points
18 comments1 min readLW link

A $10k retroac­tive grant for VaccinateCA

Austin ChenJul 27, 2023, 6:14 PM
82 points
0 commentsLW link
(manifund.org)

Prefer­ence Ag­gre­ga­tion as Bayesian Inference

berenJul 27, 2023, 5:59 PM
14 points
1 comment1 min readLW link

AI #22: Into the Weeds

ZviJul 27, 2023, 5:40 PM
49 points
8 comments84 min readLW link
(thezvi.wordpress.com)

SSA re­jects an­thropic shadow, too

jessicataJul 27, 2023, 5:25 PM
74 points
38 comments11 min readLW link
(unstableontology.com)

[Question] What are ex­am­ples of some­one do­ing a lot of work to find the best of some­thing?

chanamessingerJul 27, 2023, 3:58 PM
29 points
16 comments1 min readLW link

AI-Plans.com 10-day Cri­tique-a-Thon

IknownothingJul 27, 2023, 11:44 AM
8 points
2 comments2 min readLW link
(manifund.org)

Pri­vacy in a Digi­tal World

FaustifyJul 27, 2023, 10:46 AM
2 points
0 comments5 min readLW link

Cul­ti­vat­ing a state of mind where new ideas are born

Henrik KarlssonJul 27, 2023, 9:16 AM
244 points
21 comments14 min readLW link2 reviews
(www.henrikkarlsson.xyz)

Par­tial Tran­script of Re­cent Se­nate Hear­ing Dis­cussing AI X-Risk

Daniel_EthJul 27, 2023, 9:16 AM
55 points
0 commentsLW link
(medium.com)

AXRP Epi­sode 24 - Su­per­al­ign­ment with Jan Leike

DanielFilanJul 27, 2023, 4:00 AM
55 points
3 comments69 min readLW link

[Question] Have you ever con­sid­ered tak­ing the ‘Tur­ing Test’ your­self?

Super AGIJul 27, 2023, 3:48 AM
2 points
6 comments1 min readLW link

AXRP Epi­sode 23 - Mechanis­tic Ano­maly De­tec­tion with Mark Xu

DanielFilanJul 27, 2023, 1:50 AM
22 points
0 comments72 min readLW link

GPT-4 can catch sub­tle cross-lan­guage trans­la­tion mistakes

Michael TontchevJul 27, 2023, 1:39 AM
7 points
1 comment1 min readLW link

So­cial Balance through Em­brac­ing So­cial Credit

dhruvvJul 26, 2023, 8:07 PM
−39 points
9 comments3 min readLW link

Why no Ro­man In­dus­trial Revolu­tion?

jasoncrawfordJul 26, 2023, 7:34 PM
62 points
30 comments3 min readLW link
(rootsofprogress.org)

Why you can’t treat de­cid­abil­ity and com­plex­ity as a con­stant (Post #1)

Noosphere89Jul 26, 2023, 5:54 PM
6 points
13 comments5 min readLW link

A re­sponse to the Richards et al.’s “The Illu­sion of AI’s Ex­is­ten­tial Risk”

Harrison FellJul 26, 2023, 5:34 PM
1 point
0 comments10 min readLW link

Meta-level ad­ver­sar­ial eval­u­a­tion of over­sight tech­niques might al­low ro­bust mea­sure­ment of their adequacy

Jul 26, 2023, 5:02 PM
100 points
19 comments1 min readLW link1 review

Neuronpedia

Johnny LinJul 26, 2023, 4:29 PM
135 points
51 comments2 min readLW link
(neuronpedia.org)

Fron­tier Model Forum

Zach Stein-PerlmanJul 26, 2023, 2:30 PM
27 points
0 comments4 min readLW link
(blog.google)

Pod­casts: Fu­ture of Life In­sti­tute, Break­through Science Sum­mit panel

jasoncrawfordJul 26, 2023, 2:28 PM
8 points
0 comments1 min readLW link
(rootsofprogress.org)

Llama We Do­ing This Again?

ZviJul 26, 2023, 1:00 PM
48 points
3 comments16 min readLW link
(thezvi.wordpress.com)

Fron­tier Model Security

VaniverJul 26, 2023, 4:48 AM
32 points
1 comment3 min readLW link
(www.anthropic.com)

The First Room-Tem­per­a­ture Am­bi­ent-Pres­sure Superconductor

AnnapurnaJul 26, 2023, 2:27 AM
35 points
28 comments1 min readLW link
(arxiv.org)

Un­der­wa­ter Tor­ture Cham­bers: The Hor­ror Of Fish Farming

Bentham's BulldogJul 26, 2023, 12:27 AM
83 points
50 comments10 min readLW link1 review

Con­tra Alexan­der on the Bit­ter Les­son and IQ

Andrew Keenan RichardsonJul 26, 2023, 12:07 AM
9 points
1 comment4 min readLW link
(mechanisticmind.com)

Over­com­ing the MWC

Mark FreedJul 25, 2023, 5:31 PM
3 points
0 comments3 min readLW link

Rus­sian par­li­a­men­tar­ian: let’s ban per­sonal com­put­ers and the Internet

RomanSJul 25, 2023, 5:30 PM
11 points
6 comments2 min readLW link

AISN #16: White House Se­cures Vol­un­tary Com­mit­ments from Lead­ing AI Labs and Les­sons from Oppenheimer

Jul 25, 2023, 4:58 PM
6 points
0 comments6 min readLW link
(newsletter.safe.ai)

“The Uni­verse of Minds”—call for re­view­ers (Seeds of Science)

rogersbaconJul 25, 2023, 4:53 PM
7 points
0 comments1 min readLW link

Thoughts on Loss Land­scapes and why Deep Learn­ing works

berenJul 25, 2023, 4:41 PM
53 points
4 comments18 min readLW link

Should you work at a lead­ing AI lab? (in­clud­ing in non-safety roles)

Benjamin HiltonJul 25, 2023, 4:29 PM
7 points
0 comments12 min readLW link

Whisper’s Word-Level Times­tamps are Out

Varshul GuptaJul 25, 2023, 2:32 PM
−18 points
2 comments2 min readLW link
(dubverseblack.substack.com)

AIS 101: Task de­com­po­si­tion for scal­able oversight

Charbel-RaphaëlJul 25, 2023, 1:34 PM
35 points
0 comments19 min readLW link
(docs.google.com)

An­thropic Observations

ZviJul 25, 2023, 12:50 PM
104 points
1 comment10 min readLW link
(thezvi.wordpress.com)