Do we want too much from a po­ten­tially godlike AGI?

StanislavKrymApr 11, 2025, 11:33 PM
−1 points
0 comments2 min readLW link

How train­ing-gamers might func­tion (and win)

Vivek HebbarApr 11, 2025, 9:26 PM
107 points
5 comments13 min readLW link

The limits of black-box eval­u­a­tions: two hypotheticals

TFDApr 11, 2025, 8:45 PM
1 point
0 comments4 min readLW link
(www.thefloatingdroid.com)

Com­ments on “AI 2027”

RandalyApr 11, 2025, 8:32 PM
19 points
14 comments7 min readLW link

De­bunk the myth -Test­ing the gen­er­al­ized rea­son­ing abil­ity of LLM

Defender7762Apr 11, 2025, 8:17 PM
1 point
5 comments4 min readLW link

The­o­ries of Im­pact for Causal­ity in AI Safety

alexisbellotApr 11, 2025, 8:16 PM
11 points
1 comment6 min readLW link

Why Big­ger Models Gen­er­al­ize Better

PapersToAGIApr 11, 2025, 7:54 PM
1 point
0 comments2 min readLW link

Can LLMs learn Stegano­graphic Rea­son­ing via RL?

Apr 11, 2025, 4:33 PM
28 points
2 comments6 min readLW link

My day in 2035

TenokeApr 11, 2025, 4:31 PM
19 points
2 comments7 min readLW link
(svilentodorov.xyz)

Youth Lockout

Xavi CFApr 11, 2025, 3:05 PM
47 points
6 comments5 min readLW link

[Question] Is the ethics of in­ter­ac­tion with prim­i­tive peo­ples already solved?

StanislavKrymApr 11, 2025, 2:56 PM
−4 points
0 comments1 min readLW link

OpenAI Re­sponses API changes mod­els’ behavior

Apr 11, 2025, 1:27 PM
52 points
6 comments2 min readLW link

Weird Ran­dom New­comb Problem

TapataktApr 11, 2025, 1:09 PM
21 points
16 comments4 min readLW link

On Google’s Safety Plan

ZviApr 11, 2025, 12:51 PM
57 points
6 comments33 min readLW link
(thezvi.wordpress.com)

Луна Лавгуд и Комната Тайн, Часть 2

Apr 11, 2025, 12:42 PM
2 points
1 comment3 min readLW link

Paper

dynomightApr 11, 2025, 12:20 PM
43 points
12 comments3 min readLW link

Why are neuro-sym­bolic sys­tems not con­sid­ered when it comes to AI Safety?

Edy NastaseApr 11, 2025, 9:41 AM
3 points
6 comments1 min readLW link

Crash sce­nario 1: Rapidly mo­bil­ise for a 2025 AI crash

RemmeltApr 11, 2025, 6:54 AM
12 points
4 comments1 min readLW link

Cur­rency Collapse

prueApr 11, 2025, 3:48 AM
26 points
3 comments9 min readLW link
(www.prue0.com)

Why do mis­al­ign­ment risks in­crease as AIs get more ca­pa­ble?

ryan_greenblattApr 11, 2025, 3:06 AM
33 points
6 comments3 min readLW link

Six rea­sons why ob­jec­tive moral­ity is nonsense

Zero ContradictionsApr 11, 2025, 2:11 AM
0 points
10 comments5 min readLW link
(coelsblog.wordpress.com)

Refin­ing MAIM: Iden­ti­fy­ing Changes Re­quired to Meet Con­di­tions for Deterrence

David AbecassisApr 11, 2025, 12:49 AM
17 points
0 comments11 min readLW link
(intelligence.org)

DOGE Might Be Worth Influencing

LTMApr 11, 2025, 12:40 AM
7 points
1 comment7 min readLW link
(routecause.substack.com)

Can I learn lan­guage faster? Or, per­haps, can I mem­o­rize the for­eign words and re­call them faster?

jmhApr 11, 2025, 12:01 AM
12 points
6 comments1 min readLW link

Anti-memes: x-risk edition

WillPetilloApr 10, 2025, 11:35 PM
14 points
0 comments7 min readLW link

Fore­cast­ing time to au­to­mated su­per­hu­man coders [AI 2027 Timelines Fore­cast]

Apr 10, 2025, 11:10 PM
35 points
0 comments18 min readLW link
(ai-2027.com)

AI could cause a drop in GDP, even if mar­kets are com­pet­i­tive and efficient

Casey BarkanApr 10, 2025, 10:35 PM
27 points
0 comments5 min readLW link

Not The End of All Value

Ben IhrigApr 10, 2025, 8:53 PM
−13 points
0 comments3 min readLW link

EA Reflec­tions on my Mili­tary Career

TomGardinerApr 10, 2025, 7:01 PM
7 points
0 comments1 min readLW link
(forum.effectivealtruism.org)

Text First, Ev­i­dence Later? Manag­ing Qual­ity and Trust in an Era of AI-Aug­mented Research

Thehumanproject.aiApr 10, 2025, 6:52 PM
1 point
1 comment5 min readLW link

Nuanced Models for the In­fluence of Information

ozziegooenApr 10, 2025, 6:28 PM
8 points
0 commentsLW link

Play­ing in the Creek

HastingsApr 10, 2025, 5:39 PM
381 points
16 comments2 min readLW link
(hgreer.com)

The Three Boxes: A Sim­ple Model for Spread­ing Ideas

JohnGreerApr 10, 2025, 5:15 PM
6 points
0 comments5 min readLW link

Re­ac­tions to METR task length pa­per are insane

Cole WyethApr 10, 2025, 5:13 PM
58 points
43 comments4 min readLW link

Ex­ist­ing Safety Frame­works Im­ply Un­rea­son­able Confidence

Apr 10, 2025, 4:31 PM
37 points
2 comments15 min readLW link
(intelligence.org)

Ar­gu­ments for and against grad­ual change

Gustavo RamiresApr 10, 2025, 2:43 PM
3 points
0 comments6 min readLW link

Disem­pow­er­ment spirals as a likely mechanism for ex­is­ten­tial catastrophe

Apr 10, 2025, 2:37 PM
74 points
7 comments5 min readLW link

AI #111: Giv­ing Us Pause

ZviApr 10, 2025, 2:00 PM
26 points
4 comments34 min readLW link
(thezvi.wordpress.com)

Forg­ing A New AGI So­cial Contract

Deric ChengApr 10, 2025, 1:41 PM
23 points
3 comments7 min readLW link
(agisocialcontract.substack.com)

Why Ex­pe­rienced Pro­fes­sion­als Fail to Land High-Im­pact Roles (FBB #5)

gergogasparApr 10, 2025, 12:46 PM
12 points
4 comments9 min readLW link

Linkpost to a Sum­mary of “Imag­in­ing and build­ing wise ma­chines: The cen­tral­ity of AI metacog­ni­tion” by John­son, Karimi, Ben­gio, et al.

Chris_LeongApr 10, 2025, 11:54 AM
8 points
0 comments2 min readLW link

Grounded Ghosts in the Ma­chine—Fris­ton Blan­kets, Mir­ror Neu­rons, and the Quest for Co­op­er­a­tive AI

DavidmanheimApr 10, 2025, 10:15 AM
9 points
0 comments9 min readLW link
(davidmanheim.com)

New Paper: In­fra-Bayesian De­ci­sion-Es­ti­ma­tion Theory

Apr 10, 2025, 9:17 AM
77 points
4 comments1 min readLW link
(arxiv.org)

Elec­tric Lunchbox

jefftkApr 10, 2025, 2:40 AM
15 points
0 comments1 min readLW link
(www.jefftk.com)

Scop­ing LLMs

Apr 10, 2025, 12:32 AM
4 points
0 comments14 min readLW link

[Question] How fa­mil­iar is the Less­wrong com­mu­nity as a whole with the con­cept of Re­ward-mod­el­ling?

OxidizeApr 9, 2025, 11:33 PM
1 point
8 comments1 min readLW link

What can we learn from ex­pert AGI fore­casts?

Benjamin_ToddApr 9, 2025, 9:34 PM
5 points
0 comments5 min readLW link
(80000hours.org)

Thoughts on AI 2027

Max HarmsApr 9, 2025, 9:26 PM
222 points
61 comments21 min readLW link
(intelligence.org)

The case for AGI by 2030

Benjamin_ToddApr 9, 2025, 8:35 PM
40 points
6 comments42 min readLW link
(80000hours.org)

Anti-au­toma­tion policy as a bot­tle­neck to eco­nomic growth

mhamptonApr 9, 2025, 8:12 PM
4 points
0 comments4 min readLW link