What should we cen­sor from train­ing data?

wassnameApr 22, 2023, 11:33 PM
16 points
4 comments1 min readLW link

Ar­chi­tec­ture-aware op­ti­mi­sa­tion: train ImageNet and more with­out hyperparameters

Chris MingardApr 22, 2023, 9:50 PM
6 points
2 comments2 min readLW link

OpenAI’s GPT-4 Safety Goals

PeterMcCluskeyApr 22, 2023, 7:11 PM
3 points
3 comments4 min readLW link
(bayesianinvestor.com)

In­tro­duc­ing the Nuts and Bolts Of Naturalism

LoganStrohlApr 22, 2023, 6:31 PM
77 points
2 comments3 min readLW link

We Need To Know About Con­tinual Learning

michael_mjdApr 22, 2023, 5:08 PM
30 points
14 comments4 min readLW link

The Se­cu­rity Mind­set, S-Risk and Pub­lish­ing Pro­saic Align­ment Research

lukemarksApr 22, 2023, 2:36 PM
40 points
7 comments5 min readLW link

[Question] How did LW up­date p(doom) af­ter LLMs blew up?

FinalFormal2Apr 22, 2023, 2:21 PM
24 points
29 comments1 min readLW link

The Cruel Trade-Off Between AI Mi­suse and AI X-risk Concerns

simeon_cApr 22, 2023, 1:49 PM
24 points
1 comment2 min readLW link

five ways to say “Al­most Always” and ac­tu­ally mean it

Yudhister KumarApr 22, 2023, 10:38 AM
17 points
3 comments2 min readLW link
(www.ykumar.org)

P(doom|su­per­in­tel­li­gence) or coin tosses and dice throws of hu­man val­ues (and other re­lated Ps).

MuyydApr 22, 2023, 10:06 AM
−7 points
0 comments4 min readLW link

[Question] Is it al­lowed to post job post­ings here? I am look­ing for a new PhD stu­dent to work on AI In­ter­pretabil­ity. Can I ad­ver­tise my po­si­tion?

TiberiusApr 22, 2023, 1:22 AM
5 points
4 comments1 min readLW link

LessWrong mod­er­a­tion mes­sag­ing container

RaemonApr 22, 2023, 1:19 AM
21 points
13 comments1 min readLW link

Neu­ral net­work poly­topes (Co­lab note­book)

Zach FurmanApr 21, 2023, 10:42 PM
11 points
0 comments1 min readLW link
(colab.research.google.com)

Read­abil­ity is mostly a waste of characters

vlad.proexApr 21, 2023, 10:05 PM
21 points
7 comments3 min readLW link

The Re­la­tion­ship be­tween RLHF and AI Psy­chol­ogy: De­bunk­ing the Shog­goth Argument

FinalFormal2Apr 21, 2023, 10:05 PM
−11 points
8 comments2 min readLW link

Think­ing about max­i­miza­tion and corrigibility

James PayorApr 21, 2023, 9:22 PM
63 points
4 comments5 min readLW link

Would we even want AI to solve all our prob­lems?

So8resApr 21, 2023, 6:04 PM
98 points
15 comments2 min readLW link

The Com­mis­sion for Stop­ping Fur­ther Im­prove­ments: A let­ter of note from Isam­bard K. Brunel

jasoncrawfordApr 21, 2023, 5:42 PM
39 points
0 comments4 min readLW link
(rootsofprogress.org)

Should we pub­lish mechanis­tic in­ter­pretabil­ity re­search?

Apr 21, 2023, 4:19 PM
106 points
40 comments13 min readLW link

500 Million, But Not A Sin­gle One More—The Animation

WriterApr 21, 2023, 3:48 PM
47 points
0 commentsLW link

Talk­ing pub­li­cly about AI risk

Jan_KulveitApr 21, 2023, 11:28 AM
180 points
9 comments6 min readLW link

Notes on “the hot mess the­ory of AI mis­al­ign­ment”

JakubKApr 21, 2023, 10:07 AM
16 points
0 comments5 min readLW link
(sohl-dickstein.github.io)

Req­ui­site Variety

Stephen FowlerApr 21, 2023, 8:07 AM
6 points
0 comments5 min readLW link

The Agency Overhang

Jeffrey LadishApr 21, 2023, 7:47 AM
85 points
6 comments6 min readLW link

[Question] What would “The Med­i­cal Model Is Wrong” look like?

EloApr 21, 2023, 1:46 AM
8 points
7 comments2 min readLW link

Gas and Water

jefftkApr 21, 2023, 1:30 AM
17 points
9 comments1 min readLW link
(www.jefftk.com)

[Question] Did the fonts change?

the gears to ascensionApr 21, 2023, 12:40 AM
2 points
1 comment1 min readLW link

[Question] Should we openly talk about ex­plicit use cases for Au­toGPT?

ChristianKlApr 20, 2023, 11:44 PM
20 points
4 comments1 min readLW link

United We Align: Har­ness­ing Col­lec­tive Hu­man In­tel­li­gence for AI Align­ment Progress

Shoshannah TekofskyApr 20, 2023, 11:19 PM
41 points
13 comments25 min readLW link

[Question] Where to start with statis­tics if I want to mea­sure things?

mattoApr 20, 2023, 10:40 PM
21 points
7 comments1 min readLW link

Up­skil­ling, bridge-build­ing, re­search on se­cu­rity/​cryp­tog­ra­phy and AI safety

Allison DuettmannApr 20, 2023, 10:32 PM
14 points
0 comments4 min readLW link

Be­havi­oural statis­tics for a maze-solv­ing agent

Apr 20, 2023, 10:26 PM
46 points
11 comments10 min readLW link

An in­tro­duc­tion to lan­guage model interpretability

Alexandre VariengienApr 20, 2023, 10:22 PM
14 points
0 comments9 min readLW link

The Case for Brain-Only Preservation

Mati_RoyApr 20, 2023, 10:01 PM
21 points
7 comments1 min readLW link
(biostasis.substack.com)

[Question] Prac­ti­cal ways to ac­tu­al­ize our be­liefs into con­crete bets over a longer time hori­zon?

M. Y. ZuoApr 20, 2023, 9:21 PM
4 points
2 comments1 min readLW link

LW mod­er­a­tion: my cur­rent thoughts and ques­tions, 2023-04-12

RubyApr 20, 2023, 9:02 PM
53 points
30 comments10 min readLW link

Pro­posal: Us­ing Monte Carlo tree search in­stead of RLHF for al­ign­ment research

Christopher KingApr 20, 2023, 7:57 PM
2 points
7 comments3 min readLW link

Deep­Mind and Google Brain are merg­ing [Linkpost]

Orpheus16Apr 20, 2023, 6:47 PM
55 points
5 comments1 min readLW link
(www.deepmind.com)

Ideas for stud­ies on AGI risk

dr_sApr 20, 2023, 6:17 PM
5 points
1 comment11 min readLW link

Study 1b: This One Weird Trick does NOT cause in­cor­rect­ness cascades

Robert_AIZIApr 20, 2023, 6:10 PM
5 points
0 comments6 min readLW link
(aizi.substack.com)

An open let­ter to SERI MATS pro­gram organisers

Roman LeventovApr 20, 2023, 4:34 PM
26 points
26 comments4 min readLW link

De­cep­tion Strategies

Thoth HermesApr 20, 2023, 3:59 PM
−7 points
2 comments5 min readLW link
(thothhermes.substack.com)

Paper­clip Club (AI Safety Meetup)

LThorburnApr 20, 2023, 3:55 PM
1 point
0 comments1 min readLW link

AI #8: Peo­ple Can Do Rea­son­able Things

ZviApr 20, 2023, 3:50 PM
100 points
16 comments55 min readLW link
(thezvi.wordpress.com)

OpenAI could help X-risk by wa­ger­ing itself

VojtaKovarikApr 20, 2023, 2:51 PM
31 points
16 comments1 min readLW link

Ja­pan AI Align­ment Con­fer­ence Postmortem

Apr 20, 2023, 10:58 AM
71 points
8 comments8 min readLW link

Sta­bil­ity AI re­leases StableLM, an open-source ChatGPT counterpart

OzyrusApr 20, 2023, 6:04 AM
11 points
3 comments1 min readLW link
(github.com)

The Quan­tum Wave Func­tion is Re­lated to a Philos­o­phy Concept

Richard AragonApr 20, 2023, 3:16 AM
−11 points
3 comments6 min readLW link

A poem writ­ten by a fancy autocomplete

Christopher KingApr 20, 2023, 2:31 AM
1 point
0 comments1 min readLW link

List of com­monly used bench­marks for LLMs

DizietApr 20, 2023, 2:25 AM
8 points
0 comments1 min readLW link