AI Safety Newslet­ter #5: Ge­offrey Hin­ton speaks out on AI risk, the White House meets with AI labs, and Tro­jan at­tacks on lan­guage models

May 9, 2023, 3:26 PM
28 points
1 comment4 min readLW link
(newsletter.safe.ai)

A Search for More ChatGPT /​ GPT-3.5 /​ GPT-4 “Un­speak­able” Glitch Tokens

Martin FellMay 9, 2023, 2:36 PM
26 points
9 comments6 min readLW link

How to In­ter­pret Pre­dic­tion Mar­ket Prices as Probabilities

SimonMMay 9, 2023, 2:12 PM
14 points
1 comment4 min readLW link

Stampy’s AI Safety Info—New Distil­la­tions #2 [April 2023]

markovMay 9, 2023, 1:31 PM
25 points
1 comment1 min readLW link
(aisafety.info)

Quote quiz answer

jasoncrawfordMay 9, 2023, 1:27 PM
19 points
0 comments4 min readLW link
(rootsofprogress.org)

[Question] Does re­versible com­pu­ta­tion let you com­pute the com­plex­ity class PSPACE as effi­ciently as nor­mal com­put­ers com­pute the com­plex­ity class P?

Noosphere89May 9, 2023, 1:18 PM
6 points
14 comments1 min readLW link

EconTalk pod­cast: “Eliezer Yud­kowsky on the Dangers of AI”

TekhneMakreMay 9, 2023, 11:14 AM
15 points
1 comment1 min readLW link
(www.econtalk.org)

Most peo­ple should prob­a­bly feel safe most of the time

Kaj_SotalaMay 9, 2023, 9:35 AM
95 points
28 comments10 min readLW link

Sum­maries of top fo­rum posts (1st to 7th May 2023)

Zoe WilliamsMay 9, 2023, 9:30 AM
21 points
0 commentsLW link

Fo­cus­ing on longevity re­search as a way to avoid the AI apocalypse

Random TraderMay 9, 2023, 4:47 AM
14 points
2 comments2 min readLW link

When is Good­hart catas­trophic?

May 9, 2023, 3:59 AM
180 points
29 comments8 min readLW link1 review

Chilean AIS Hackathon Retrospective

agucovaMay 9, 2023, 1:34 AM
9 points
0 commentsLW link

An­nounc­ing “Key Phenom­ena in AI Risk” (fa­cil­i­tated read­ing group)

May 9, 2023, 12:31 AM
65 points
4 comments2 min readLW link

Yoshua Ben­gio ar­gues for tool-AI and to ban “ex­ec­u­tive-AI”

habrykaMay 9, 2023, 12:13 AM
53 points
15 comments7 min readLW link
(yoshuabengio.org)

South Bay ACX/​LW Meetup

ISMay 8, 2023, 11:55 PM
2 points
0 comments1 min readLW link

H-JEPA might be tech­ni­cally al­ignable in a mod­ified form

Roman LeventovMay 8, 2023, 11:04 PM
12 points
2 comments7 min readLW link

All AGI Safety ques­tions wel­come (es­pe­cially ba­sic ones) [May 2023]

steven0461May 8, 2023, 10:30 PM
33 points
44 comments2 min readLW link

Pre­dictable up­dat­ing about AI risk

Joe CarlsmithMay 8, 2023, 9:53 PM
294 points
25 comments36 min readLW link1 review

An­no­tated re­ply to Ben­gio’s “AI Scien­tists: Safe and Use­ful AI?”

Roman LeventovMay 8, 2023, 9:26 PM
18 points
2 comments7 min readLW link
(yoshuabengio.org)

Are healthy choices effec­tive for im­prov­ing live ex­pec­tancy any­more?

Christopher KingMay 8, 2023, 9:25 PM
4 points
4 comments1 min readLW link

LeCun’s “A Path Towards Au­tonomous Ma­chine In­tel­li­gence” has an un­solved tech­ni­cal al­ign­ment problem

Steven ByrnesMay 8, 2023, 7:35 PM
140 points
37 comments15 min readLW link

Product En­dorse­ment: Apollo Neuro

ElizabethMay 8, 2023, 7:00 PM
46 points
28 comments5 min readLW link
(acesounderglass.com)

Acausal trade nat­u­rally re­sults in the Nash bar­gain­ing solution

Christopher KingMay 8, 2023, 6:13 PM
3 points
0 comments4 min readLW link

In­fer­ence Speed is Not Unbounded

OneManyNoneMay 8, 2023, 4:24 PM
35 points
32 comments16 min readLW link

[Cross­post] Un­veiling the Amer­i­can Public Opinion on AI Mo­ra­to­rium and Govern­ment In­ter­ven­tion: The Im­pact of Me­dia Exposure

otto.bartenMay 8, 2023, 2:09 PM
7 points
0 comments6 min readLW link
(forum.effectivealtruism.org)

Thriv­ing in the Weird Times: Prepar­ing for the 100X Economy

May 8, 2023, 1:44 PM
23 points
16 comments2 min readLW link

Hous­ing and Tran­sit Roundup #4

ZviMay 8, 2023, 1:30 PM
25 points
0 comments11 min readLW link
(thezvi.wordpress.com)

Dance Profit Sharing

jefftkMay 8, 2023, 1:10 PM
11 points
3 comments2 min readLW link
(www.jefftk.com)

How “AGI” could end up be­ing many differ­ent spe­cial­ized AI’s stitched together

titotalMay 8, 2023, 12:32 PM
9 points
2 commentsLW link

What does it take to ban a thing?

qbolecMay 8, 2023, 11:00 AM
66 points
18 comments5 min readLW link

Solomonoff’s solip­sism

Mergimio H. DoefevmilMay 8, 2023, 6:55 AM
−13 points
9 comments1 min readLW link

A tech­ni­cal note on bil­in­ear lay­ers for interpretability

Lee SharkeyMay 8, 2023, 6:06 AM
59 points
0 comments1 min readLW link
(arxiv.org)

[Question] Is EDT cor­rect? Does “EDT” == “log­i­cal EDT” == “log­i­cal CDT”?

Vivek HebbarMay 8, 2023, 2:07 AM
13 points
2 comments1 min readLW link

LLM cog­ni­tion is prob­a­bly not hu­man-like

Max HMay 8, 2023, 1:22 AM
26 points
15 comments7 min readLW link

[Question] If al­ign­ment prob­lem was un­solv­able, would that avoid doom?

KinranyMay 7, 2023, 10:13 PM
3 points
3 comments1 min readLW link

An ar­tifi­cially struc­tured ar­gu­ment for ex­pect­ing AGI ruin

Rob BensingerMay 7, 2023, 9:52 PM
91 points
26 comments19 min readLW link

Where “the Se­quences” Are Wrong

Thoth HermesMay 7, 2023, 8:21 PM
−15 points
5 comments14 min readLW link
(thothhermes.substack.com)

What’s wrong with be­ing dumb?

Adam ZernerMay 7, 2023, 6:31 PM
14 points
17 comments2 min readLW link

Cat­e­gories of Ar­gu­ing Style : Why be­ing good among ra­tio­nal­ists isn’t enough to ar­gue with everyone

Camille Berger May 7, 2023, 5:45 PM
16 points
0 comments23 min readLW link

Self-Ad­ministered Gell-Mann Amnesia

krsMay 7, 2023, 5:44 PM
1 point
1 comment1 min readLW link

Un­der­stand­ing mesa-op­ti­miza­tion us­ing toy models

May 7, 2023, 5:00 PM
45 points
6 comments10 min readLW link

How to have Poly­geni­cally Screened Children

GeneSmithMay 7, 2023, 4:01 PM
367 points
128 comments27 min readLW link1 review

Statis­ti­cal mod­els & the ir­rele­vance of rare exceptions

patrissimoMay 7, 2023, 3:59 PM
36 points
6 comments2 min readLW link

Let’s look for co­her­ence theorems

ValdesMay 7, 2023, 2:45 PM
25 points
18 comments6 min readLW link

Graph­i­cal Rep­re­sen­ta­tions of Paul Chris­ti­ano’s Doom Model

Nathan YoungMay 7, 2023, 1:03 PM
7 points
0 commentsLW link

An an­thro­po­mor­phic AI dilemma

TsviBTMay 7, 2023, 12:44 PM
26 points
0 comments7 min readLW link

Violin Supports

jefftkMay 7, 2023, 12:10 PM
12 points
1 comment1 min readLW link
(www.jefftk.com)

Prop­er­ties of Good Textbooks

niplavMay 7, 2023, 8:38 AM
50 points
11 comments1 min readLW link

Against sac­ri­fic­ing AI trans­parency for gen­er­al­ity gains

Ape in the coatMay 7, 2023, 6:52 AM
4 points
0 comments2 min readLW link

TED talk by Eliezer Yud­kowsky: Un­leash­ing the Power of Ar­tifi­cial Intelligence

bayesedMay 7, 2023, 5:45 AM
49 points
36 comments1 min readLW link
(www.youtube.com)