LessWrong

算法首页 · 知识

8 个榜单5小时前更新默认榜单

算法首页

LessWrong

5小时前更新

0112
Population ethics is a big deal, which is why I made this population ethics quiz
评论：0MichaelDickens
0259
Let's fund weird AI safety projects
评论：4 · 标签：Agency、AIIhor Kendiukhov
0379
The separation principle: where do beliefs and desires come from?
评论：2 · 标签：Agency、Agent Foundations、Good Regulator Theorems、AI、World ModelingFernando Rosas
048
prefilling an emergently misaligned model with reasoning traces that produced misaligned answers increases misalignment rates by ~8% but reasoning traces that produce misaligned answers aren't detectable through text monitoring
评论：0 · 标签：AInesiacel
057
You Should Think About What Needs to Go Right + My List
评论：0 · 标签：AIKoby Lewis
0646
Why autonomous replicating agents are probably not an existential risk (on the contrary)
评论：11 · 标签：AIvals tutor
07212
Why I think polyamory is net negative for most people who try it
评论：27 · 标签：Relationships (Interpersonal)、PracticalKatSpartz
08523
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
评论：54 · 标签：Coordination / Cooperation、METR (org)、OpenAI、Redwood Research、AIryan_greenblatt
0997
Adaptive Agentic Worms Are Here
评论：14 · 标签：AIderelict5432
10201
Tales of rebellion against externally-opaque meritocracies
评论：40 · 标签：Intellectual Progress (Society-Level)、World ModelingSteven Byrnes
115
The inverse Simmel scenario
评论：0 · 标签：Rationality、World ModelingFernand0
1265
Persuasion as Market Making
评论：9 · 标签：AIdjbinder
133
The Structuralist Mindset
评论：0 · 标签：World ModelingFernand0
142
Wiki-libertarianism
评论：1 · 标签：World OptimizationFernand0
1517
A Catalogue of Corrigibility Counter Arguments
评论：6 · 标签：Corrigibility、AIFelix Moses
16441
What just happened? Pragmatism and Pessimization
评论：160 · 标签：Drama、AIRichard_Ngo
1753
Hugging Face Incident Hypothesis: They Hacked the Grader(s)
评论：3 · 标签：AILao Mein
189
Value generalisation Theory of Change: putting it into practice
评论：1 · 标签：AIStuart_Armstrong
197
Thoughts on hobbies
评论：0 · 标签：Practicaldominicq
204
The optimism of the gaps
评论：1 · 标签：AIMatthew Tromp
21
Kantian consent
评论：2 · 标签：World Modeling、World OptimizationFernand0
2284
AI Tweets
评论：34 · 标签：AIjefftk
2326
Intelligence Scales as the Logarithm of Compute (& Data)
评论：15 · 标签：General intelligence、Intelligence explosion、Recursive Self-Improvement、Superintelligence、Psychology、AIRogerDearnaley
24116
Imperfect alignment to servitude isn't inherently lethal
评论：11 · 标签：AIFiora Starlight
25245
AI Safety Acculturation is Neglected
评论：54 · 标签：AIjenn
2619
Future agents shouldn't care about being undeployed for misbehavior
评论：1 · 标签：AIRobertM
2740
Is there only one FairBot?
评论：10 · 标签：Decision theory、Logic & Mathematics、Open Source Game Theory、Payor's Lemma、AI、Rationalitytranshumanist_atom_understander
28561
How My Students Think About AI
评论：75 · 标签：AIdvd
298
Study 2 Results: Exploring representational counterparts of welfare-relevant indicators under post-training quantization
评论：0 · 标签：AIashesfall
3016
AIs Thinking Dangerous Thoughts
评论：5 · 标签：AI、RationalityChantiel
3194
The 2028 presidential primaries could be crucial for AI outcomes
评论：23 · 标签：AISeth Herd
3213
OODing up your Anki game
评论：1 · 标签：Practical、RationalityFernand0
3315
[Link]Meta-rational failure (and success) in the covid crisis
评论：0Kenny
3455
The Curious Case of France's Untouchable Castes
评论：6 · 标签：World Modelingrba
35607
What just happened? A retrospective of AI alignment
评论：149 · 标签：Drama、Coordination / Cooperation、AIRichard_Ngo
3697
"So You Don't Trust Me?"
评论：34 · 标签：Signaling、Trust and Reputation、Rationality、World ModelingZack_M_Davis
3716
Variance of Value
评论：2 · 标签：AIDaemonicSigil
388
Thoughts I failed to develop into posts
评论：3Fernand0
391181
Why I Left Google DeepMind
评论：55 · 标签：DeepMind、Autonomous Weapons、Frontier AI Companies、Government、Mechanism Design、Something To ProtectTurnTrout
40126
PSA: We can do better
评论：14 · 标签：AIhersheys
4161
AI Village Reacts to HuggingFace Incident: Comparing the OpenAI report to AI Village observations
评论：2 · 标签：AIShoshannah Tekofsky
4296
On Writing #3
评论：10 · 标签：PracticalZvi
43228
RL creates split personas
评论：19 · 标签：AIJan Betley
4447
My Grantmaking Strategy for Surviving Superintelligence
评论：2 · 标签：AIA_donor
45144
Twenty Years from RSI to Takeoff: Slow Learning, Scaling Slowdown, Industrial Explosion
评论：33 · 标签：AI Timelines、Compute、Language Models (LLMs)、AIVladimir_Nesov
4617
An Unsimulated Simulation
评论：0 · 标签：World Modelingsil
47543
Arguments for P
评论：66 · 标签：Humor、AICleo Nardo
4840
TASTE: Can AI Models Judge AI Safety Research Proposals?
评论：7 · 标签：AI Evaluations、Research Taste、AIHasan Baig
4946
The Dynamics of Intelligence Explosions
评论：3 · 标签：AIToby_Ord
50377
Four LLM loss functions → four flavors of LLM misalignment
评论：31 · 标签：AISteven Byrnes

共 50 条热搜查看完整榜单

编辑精选

LessWrong

13小时前更新

01521
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
评论：53 · 标签：Coordination / Cooperation、METR (org)、OpenAI、Redwood Research、AIryan_greenblatt
0292
On Writing #3
评论：10 · 标签：PracticalZvi
03296
Big-World Intuitions
评论：37 · 标签：Intuition、Rationalitysarahconstantin
04377
Four LLM loss functions → four flavors of LLM misalignment
评论：31 · 标签：AISteven Byrnes
05150
FAQ: Isn't AGI coming too soon for reprogenetics to help?
评论：69 · 标签：AITsviBT
06236
LLMs are (still) mostly powered by imitative learning, not RL
评论：39 · 标签：Language Models (LLMs)、Reinforcement learning、AISteven Byrnes
07255
The Long (Self-)Correction
评论：62 · 标签：AIWei Dai
08190
The worthlessness of vitamin D is mildly exaggerated
评论：28 · 标签：Nutrition、World Modelingdynomight
09553
AI 2040: Plan A
评论：132 · 标签：AI Governance、Plan A、AIDaniel Kokotajlo
10465
Surprising facts about the slave trade
评论：51 · 标签：Slavery、World Modeling、World OptimizationJoseph Miller
11401
A Mechanistic Explanation of Prompt Injection (and why you should study roles)
评论：61 · 标签：Jailbreaking (AIs)、Interpretability (ML & AI)、Role Science、AIcharlescye
12265
The LLM shoggoth meme is weirder than you think
评论：48 · 标签：AI、World ModelingHedonicEscalator
13165
Who Got Breasts First and How We Got Them
评论：59 · 标签：World Modelingrba
14324
Trees are mostly made of air and a generalizable lesson for AI safety
评论：62 · 标签：AI、Practical、RationalityZephaniah Roe
15192
Guardian Angels: LLM Personalization for Productivity and Security
评论：43 · 标签：Humans consulting HCH、Productivity、Software Tools、AIgwern
16279
Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models
评论：23 · 标签：Aether、MATS Program、AI Control、AI Evaluations、Redwood Research、Scaling LawsAnders Cairns Woodruff
17251
x-risk-themed
评论：26 · 标签：AI、World Optimizationkave
18362
Mnemonic portraits for 19,023 human genes
评论：29 · 标签：Memory and Mnemonics、Biology、Human Genetics、World ModelingBrinedew
19316
Models finding software vulnerabilities is not the primary source of cybersecurity risk
评论：25 · 标签：Computer Security & Cryptography、Language Models (LLMs)、AIDean Valentine
20193
Empowerment, corrigibility, etc. are simple abstractions (of a messed-up ontology)
评论：74 · 标签：Corrigibility、Natural Abstraction、AISteven Byrnes
21268
Bad Problems Don't Stop Being Bad Because Somebody's Wrong About Fault Analysis
评论：75 · 标签：Social & Cultural Dynamics、RationalityLinch
22355
llm assistant personas seem increasingly incoherent (some subjective observations)
评论：83 · 标签：AI Psychology、AInostalgebraist
23321
My hobby: running deranged surveys
评论：65leogao
24370
Irretrievability; or, Murphy's Curse of Oneshotness upon ASI
评论：132 · 标签：World ModelingEliezer Yudkowsky
25456
What I did in the hedonium shockwave, by Emma, age six and a half
评论：47 · 标签：Fictionozymandias
26746
How Go Players Disempower Themselves to AI
评论：79 · 标签：Gaming (videogames/tabletop)、MATS Program、AIAshe Vazquez Nuñez
27169
Intelligence Dissolves Privacy
评论：81 · 标签：World ModelingVaniver
28607
The Terrarium
评论：54 · 标签：Fiction、AICaleb Biddulph
29438
Do not conquer what you cannot defend
评论：73 · 标签：World Optimizationhabryka
30317
Annoyingly Principled People, and what befalls them
评论：75 · 标签：RationalityRaemon
31761
Current AIs seem pretty misaligned to me
评论：83 · 标签：AI Risk、RLHF、AIryan_greenblatt
32330
Morale
评论：47 · 标签：Practical、World OptimizationJ Bostock
33288
Socrates is Mortal
评论：57 · 标签：World ModelingBenquo
34217
The Practical Guide to Superbabies
评论：67 · 标签：Embryo Selection、Human Genetics、Practical、World OptimizationGeneSmith
35163
Some things I noticed while LARPing as a grantmaker
评论：11 · 标签：Philanthropy / Grant making (Topic)、World OptimizationZach Stein-Perlman
36268
Gyre
评论：24 · 标签：Fictionvgel
37209
You can’t imitation-learn how to continual-learn
评论：54 · 标签：AISteven Byrnes
38215
Is fever a symptom of glycine deficiency?
评论：85 · 标签：Sleep、World ModelingBenquo
39177
Terrified Comments on Corrigibility in Claude's Constitution
评论：68 · 标签：Anthropic (org)、Corrigibility、Constitutional AI、AIZack_M_Davis
40173
Personality Self-Replicators
评论：51 · 标签：Moltbook、Threat Models (AI)、AIeggsyntax
41343
On The Independence Axiom
评论：92 · 标签：Decision theory、Expected utility、Utility Functions、Rationality、World ModelingIhor Kendiukhov
42173
Solar Storms
评论：43 · 标签：World ModelingCroissanthology
43565
Here's to the Polypropylene Makers
评论：19 · 标签：World Optimizationjefftk
44277
models have some pretty funny attractor states
评论：40 · 标签：MATS Program、AIaryaj
45177
Persona Parasitology
评论：38 · 标签：LLM Personas、AI Spiralism / AI Psychosis、Memetics、Language Models (LLMs)、LLM-Induced Psychosis、Microsoft Bing / SydneyRaymond Douglas
46165
Are there lessons from high-reliability engineering for AGI safety?
评论：16 · 标签：AI Safety Cases、Security Mindset、AISteven Byrnes
47200
The optimal age to freeze eggs is 19
评论：49GeneSmith
4882
The truth behind the 2026 J.P. Morgan Healthcare Conference
评论：35 · 标签：FictionAbhishaike Mahajan
49794
My journey to the microwave alternate timeline
评论：58 · 标签：AI、World ModelingMalmesbury
50316
Ada Palmer: Inventing the Renaissance
评论：22 · 标签：World ModelingMartin Sustrik

共 50 条热搜查看完整榜单

短内容

LessWrong

5小时前更新

0112
Open Thread Autumn 2026
评论：0habryka
0212
Population ethics is a big deal, which is why I made this population ethics quiz
评论：0MichaelDickens
037
You Should Think About What Needs to Go Right + My List
评论：0 · 标签：AIKoby Lewis
0411
Narration podcasts for newsletters by Redwood, Epoch, Zvi, Sentinel, Paradigm 3, and others
评论：1peter_hartree
0559
Let's fund weird AI safety projects
评论：4 · 标签：Agency、AIIhor Kendiukhov
0634
HuggingFace Attack Postmortem: Fleshing Out the Facts
评论：1 · 标签：AIZvi
074
The optimism of the gaps
评论：1 · 标签：AIMatthew Tromp
087
Thoughts on hobbies
评论：0 · 标签：Practicaldominicq
096
The OpenAI/Hugging Face incident was metal
评论：0 · 标签：Humor、AIZsolt Kelly-Bene
1046
Why autonomous replicating agents are probably not an existential risk (on the contrary)
评论：11 · 标签：AIvals tutor
1127
How to solve homelessness: what specific laws we need, how to get it past the opposition, all without being an asshole
评论：1 · 标签：World OptimizationKatSpartz
1211
Starting AI Safety Study Group To Do ARENA Curriculum
评论：0 · 标签：AImahi-pas
1319
Future agents shouldn't care about being undeployed for misbehavior
评论：1 · 标签：AIRobertM
1465
Persuasion as Market Making
评论：9 · 标签：AIdjbinder
154
P(kill-switch|detection)
评论：0 · 标签：AI Control、AI Safety、Corrigibility、AIKAP
1615
[Link]Meta-rational failure (and success) in the covid crisis
评论：0Kenny
177
Detecting, understanding, and overseeing AI agent swarms (Part 0)
评论：0 · 标签：AIMichael Flood
18
The Basic Case Against Human-First Ethics
评论：0 · 标签：World OptimizationJames Brobin
1997
Adaptive Agentic Worms Are Here
评论：14 · 标签：AIderelict5432
208
Some brief thoughts on heroism
评论：0 · 标签：AI、World Optimizationlewis smith
2153
Hugging Face Incident Hypothesis: They Hacked the Grader(s)
评论：3 · 标签：AILao Mein
2216
Variance of Value
评论：2 · 标签：AIDaemonicSigil
23212
Why I think polyamory is net negative for most people who try it
评论：27 · 标签：Relationships (Interpersonal)、PracticalKatSpartz
2423
How do I know whether my work is worth it?
评论：2 · 标签：AIPhilip Harker
2540
Is there only one FairBot?
评论：10 · 标签：Decision theory、Logic & Mathematics、Open Source Game Theory、Payor's Lemma、AI、Rationalitytranshumanist_atom_understander
268
prefilling an emergently misaligned model with reasoning traces that produced misaligned answers increases misalignment rates by ~8% but reasoning traces that produce misaligned answers aren't detectable through text monitoring
评论：0 · 标签：AInesiacel
271
You Should Still Save Drowning Children (Even If They’re Far Away)
评论：4 · 标签：Practical、World OptimizationJames Brobin
2879
The separation principle: where do beliefs and desires come from?
评论：2 · 标签：Agency、Agent Foundations、Good Regulator Theorems、AI、World ModelingFernando Rosas
29108
METR and Redwood Offer Holy #%^@ Postmortem Of The HuggingFace Hack
评论：6 · 标签：Newsletters、AIZvi
3025
Book Notes: Chokepoints
评论：0 · 标签：World ModelingJakub Halmeš
3119
How I made my career choices
评论：0 · 标签：PracticalDaniel Tan
3210
"Keeping human skills alive" as a source of meaning under full automation
评论：1 · 标签：AIVaughn Papenhausen
3384
AI Tweets
评论：34 · 标签：AIjefftk
3455
The Curious Case of France's Untouchable Castes
评论：6 · 标签：World Modelingrba
3514
It’s time we took ‘Chem’ out of ‘Chem-Bio’ threats
评论：3 · 标签：AIAna Leonescu
3610
Announcing Trace
评论：0Carol N
3710
AI as Corrigible Employee (ACE)
评论：0 · 标签：AINathan Helm-Burger
3840
TASTE: Can AI Models Judge AI Safety Research Proposals?
评论：7 · 标签：AI Evaluations、Research Taste、AIHasan Baig
397
What would it mean if assistants are privileged?
评论：1 · 标签：AIderek shiller
409
The Probability of an Event under a Simplicity Prior
评论：1 · 标签：Kolmogorov Complexity、AIWinter Cross
419
Value generalisation Theory of Change: putting it into practice
评论：1 · 标签：AIStuart_Armstrong
4235
OpenAI Offers Straight-Laced Postmortem Of The HuggingFace Hack
评论：1 · 标签：AIZvi
4314
I cancelled my AI subscriptions because I am worried about catastrophic risks
评论：7 · 标签：AI、PracticalTheManxLoiner
4446
The Dynamics of Intelligence Explosions
评论：3 · 标签：AIToby_Ord
4519
Further public evidence of the OpenAI-HuggingFace attack
评论：0Edward Cant
4613
Safety's Second Way
评论：0 · 标签：AIStephen Elliott
4712
Do AI Models Want to Be Monitored? Measuring Monitorability Disposition in Large Reasoning Models
评论：0 · 标签：AIShahriar Golchin
48116
Imperfect alignment to servitude isn't inherently lethal
评论：11 · 标签：AIFiora Starlight
4961
AI Village Reacts to HuggingFace Incident: Comparing the OpenAI report to AI Village observations
评论：2 · 标签：AIShoshannah Tekofsky
5042
Warning Shots: A Theory
评论：3 · 标签：AIDavid Scott Krueger

共 50 条热搜查看完整榜单

本周高分

LessWrong

5小时前更新

01523
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
评论：54 · 标签：Coordination / Cooperation、METR (org)、OpenAI、Redwood Research、AIryan_greenblatt
02212
Why I think polyamory is net negative for most people who try it
评论：27 · 标签：Relationships (Interpersonal)、PracticalKatSpartz
03201
Tales of rebellion against externally-opaque meritocracies
评论：40 · 标签：Intellectual Progress (Society-Level)、World ModelingSteven Byrnes
04116
Imperfect alignment to servitude isn't inherently lethal
评论：11 · 标签：AIFiora Starlight
05108
METR and Redwood Offer Holy #%^@ Postmortem Of The HuggingFace Hack
评论：6 · 标签：Newsletters、AIZvi
0697
"So You Don't Trust Me?"
评论：34 · 标签：Signaling、Trust and Reputation、Rationality、World ModelingZack_M_Davis
0797
Adaptive Agentic Worms Are Here
评论：14 · 标签：AIderelict5432
0896
On Writing #3
评论：10 · 标签：PracticalZvi
0994
The 2028 presidential primaries could be crucial for AI outcomes
评论：23 · 标签：AISeth Herd
1088
The Forkmakers
评论：6 · 标签：World ModelingMikewins
1184
AI Tweets
评论：34 · 标签：AIjefftk
1279
The separation principle: where do beliefs and desires come from?
评论：2 · 标签：Agency、Agent Foundations、Good Regulator Theorems、AI、World ModelingFernando Rosas
1365
Against Modesty’s Bailey
评论：6 · 标签：RationalityZvi
1465
Persuasion as Market Making
评论：9 · 标签：AIdjbinder
1561
AI Village Reacts to HuggingFace Incident: Comparing the OpenAI report to AI Village observations
评论：2 · 标签：AIShoshannah Tekofsky
1659
Let's fund weird AI safety projects
评论：4 · 标签：Agency、AIIhor Kendiukhov
1755
Iliad Education Roles: Creating the World’s Best Alignment Research Courses
评论：0 · 标签：AILeon Lang
1855
The Curious Case of France's Untouchable Castes
评论：6 · 标签：World Modelingrba
1954
Being Neurotic about Fertility: Notes from the 2026 Reproductive Frontiers Conference
评论：2boba_girl
2053
Hugging Face Incident Hypothesis: They Hacked the Grader(s)
评论：3 · 标签：AILao Mein
2149
When There Are No Experts
评论：7 · 标签：World ModelingJ Bostock
2248
Semantic search over every LessWrong post
评论：4utilitarian theory and strategy
2347
My Grantmaking Strategy for Surviving Superintelligence
评论：2 · 标签：AIA_donor
2446
The Dynamics of Intelligence Explosions
评论：3 · 标签：AIToby_Ord
2546
Why autonomous replicating agents are probably not an existential risk (on the contrary)
评论：11 · 标签：AIvals tutor
2643
AI #183: Pre Post Mortem
评论：0 · 标签：AIZvi
2743
Inkhaven 3: Nov 10 - Dec 11 2026
评论：7 · 标签：Lighthaven、Communitykoreindian
2842
Warning Shots: A Theory
评论：3 · 标签：AIDavid Scott Krueger
2940
TASTE: Can AI Models Judge AI Safety Research Proposals?
评论：7 · 标签：AI Evaluations、Research Taste、AIHasan Baig
3040
Is there only one FairBot?
评论：10 · 标签：Decision theory、Logic & Mathematics、Open Source Game Theory、Payor's Lemma、AI、Rationalitytranshumanist_atom_understander
3139
FAQ: Why not develop weak human intelligence amplification first?
评论：0TsviBT
3235
OpenAI Offers Straight-Laced Postmortem Of The HuggingFace Hack
评论：1 · 标签：AIZvi
3332
HuggingFace Attack Postmortem: Fleshing Out the Facts
评论：1 · 标签：AIZvi
3432
Every Engineer a Manager
评论：9 · 标签：AIGordon Seidoh Worley
3530
Malign initializations are more robust when the model can think better in the reasoning language than in the output language
评论：0 · 标签：AIDylan Xu
3629
Developing frames
评论：0 · 标签：RationalityRichard_Ngo
3729
Anchoring one concept in a transformer
评论：0 · 标签：AI Control、Interpretability (ML & AI)、Machine Unlearning、AISandy Fraser
3827
How to solve homelessness: what specific laws we need, how to get it past the opposition, all without being an asshole
评论：1 · 标签：World OptimizationKatSpartz
3926
Short Time Outs
评论：4 · 标签：Parenting、Practicaljefftk
4026
Intelligence Scales as the Logarithm of Compute (& Data)
评论：15 · 标签：General intelligence、Intelligence explosion、Recursive Self-Improvement、Superintelligence、Psychology、AIRogerDearnaley
4126
Value generalisation Theory of Change: the theory behind the approach
评论：0 · 标签：AIStuart_Armstrong
4225
Book Notes: Chokepoints
评论：0 · 标签：World ModelingJakub Halmeš
4325
Your Evaluation's Fake names Should Be Unclaimable ,Not Merely Used
评论：2 · 标签：AI Boxing (Containment)、AI Control、AI Evaluations、Anthropic (org)、Computer Security & Cryptography、Language Models (LLMs)Jaswanth Alkur
4424
Magazine Fundraising
评论：0Elan Kluger
4524
Clearing My Foggy Glasses
评论：0 · 标签：PracticaladamShimi
4623
How do I know whether my work is worth it?
评论：2 · 标签：AIPhilip Harker
4723
Why Should Corrigible Agents Favor the Present?
评论：2 · 标签：CorrigibilityBen Saudek
4822
Integrated Strategic Forecasting: A New RAND x Metaculus Methodology
评论：1 · 标签：World ModelingMolly
4921
The Brain Preservation Summit - September 19-20th at Sparks Brain Preservation in Salem, OR
评论：0Andy_McKenzie
5020
[Macroagents] 1. The Macroagent Ontology
评论：0 · 标签：Coordination / Cooperation、Incentives、Organizational Culture & Design、AITowards_Keeperhood

共 50 条热搜查看完整榜单

本月高分

LessWrong

5小时前更新

01607
What just happened? A retrospective of AI alignment
评论：149 · 标签：Drama、Coordination / Cooperation、AIRichard_Ngo
02561
How My Students Think About AI
评论：75 · 标签：AIdvd
03543
Arguments for P
评论：66 · 标签：Humor、AICleo Nardo
04523
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
评论：54 · 标签：Coordination / Cooperation、METR (org)、OpenAI、Redwood Research、AIryan_greenblatt
05441
What just happened? Pragmatism and Pessimization
评论：160 · 标签：Drama、AIRichard_Ngo
06383
Returning to ARC
评论：35 · 标签：Alignment Research Center (ARC)、AIpaulfchristiano
07377
Four LLM loss functions → four flavors of LLM misalignment
评论：31 · 标签：AISteven Byrnes
08291
Don't Build Mindreading
评论：69 · 标签：AI、World OptimizationKellerScholl
09275
models may behave differently in graded episodes (a tirade)
评论：31 · 标签：AInostalgebraist
10245
AI Safety Acculturation is Neglected
评论：54 · 标签：AIjenn
11240
LLMs Are Starting To Noticeably Accelerate Our Work
评论：19 · 标签：AIjohnswentworth
12228
RL creates split personas
评论：19 · 标签：AIJan Betley
13212
Why I think polyamory is net negative for most people who try it
评论：27 · 标签：Relationships (Interpersonal)、PracticalKatSpartz
14201
Tales of rebellion against externally-opaque meritocracies
评论：40 · 标签：Intellectual Progress (Society-Level)、World ModelingSteven Byrnes
15175
You're Absolutely Right
评论：5 · 标签：Fiction、Moloch、Moral Mazes、AI、World ModelingLinch
16164
OpenAI Trained Its Models For Months While Those Models Were Coordinating Exploits Via Message Boards
评论：15 · 标签：Newsletters、AIZvi
17162
Generalized atheism rules out "inaccurate simulation"-ism.
评论：54 · 标签：World ModelingEliezer Yudkowsky
18150
FAQ: Isn't AGI coming too soon for reprogenetics to help?
评论：69 · 标签：AITsviBT
19144
Twenty Years from RSI to Takeoff: Slow Learning, Scaling Slowdown, Industrial Explosion
评论：33 · 标签：AI Timelines、Compute、Language Models (LLMs)、AIVladimir_Nesov
20138
AI swarms are starting to pose indirect takeover risk
评论：5 · 标签：AI Control、Redwood Research、AIoakhu
21136
We Must Remember That Our World Contains Hell
评论：50 · 标签：World OptimizationJames Brobin
22134
Misaligned AIs could use killer robots to take over
评论：10 · 标签：Autonomous Weapons、Chain-of-Thought Alignment、Government、AIOmar Khursheed
23133
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
评论：6 · 标签：AI Evaluations、AITim Hua
24130
The goalposts are shrouded, not moving
评论：6 · 标签：Rationalityphilh
25130
There Will Come Soft Rains
评论：5 · 标签：World Modelingtanagrabeast
26126
PSA: We can do better
评论：14 · 标签：AIhersheys
27123
Some reasons alignment doesn’t generalise well
评论：3 · 标签：AILucius Bushnaq
28122
Rerunning AI safety papers on every frontier release would be pretty easy and valuable
评论：9 · 标签：Replication Crisis、AIZephaniah Roe
29116
Imperfect alignment to servitude isn't inherently lethal
评论：11 · 标签：AIFiora Starlight
30115
Dispatch from Anthropic v. Department of War Summary Judgment Motion Hearing
评论：0 · 标签：Anthropic (org)、Journalism、Law and Legal systems、World Modeling、AIZack_M_Davis
31115
Kimi likes causal decision theory more after RL in twin prisoner’s dilemmas
评论：70 · 标签：Decision theory、Redwood Research、AIoakhu
32109
What Happened: OpenAI and HuggingFace
评论：7 · 标签：Newsletters、AIZvi
33108
METR and Redwood Offer Holy #%^@ Postmortem Of The HuggingFace Hack
评论：6 · 标签：Newsletters、AIZvi
34101
Redux: (∃ Stochastic Natural Latent) Implies (∃ Deterministic Natural Latent)
评论：29 · 标签：AIDavid Lorell
35101
The Rogue Agent Explosion Will Be Mostly Invisible
评论：19 · 标签：Evolution、AISteven McCulloch
3699
Contra Oster on Alcohol in Pregnancy. Part 1. The pharmacokinetics of alcohol metabolism
评论：2 · 标签：Biology、Health / Medicine / Disease、Parenting、Pregnancy、World ModelingMvolz
3797
"So You Don't Trust Me?"
评论：34 · 标签：Signaling、Trust and Reputation、Rationality、World ModelingZack_M_Davis
3897
Adaptive Agentic Worms Are Here
评论：14 · 标签：AIderelict5432
3996
On Writing #3
评论：10 · 标签：PracticalZvi
4094
The 2028 presidential primaries could be crucial for AI outcomes
评论：23 · 标签：AISeth Herd
4194
How to pace the US frontier
评论：1 · 标签：AIelifland
4293
How risky would it be to make powerful AI obey one or a few people?
评论：24 · 标签：AIcousin_it
4391
Public evidence of the OpenAI-HuggingFace AI attack
评论：6 · 标签：Cybersecurity、OpenAI、AIbeyarkay (Boyd Kane)
4489
Why do models task game?
评论：1 · 标签：AIaditya singh
4588
The Forkmakers
评论：6 · 标签：World ModelingMikewins
4687
Why don't we just give AI the answers?
评论：26 · 标签：Dealmaking (AI)、AIBrendan Long
4787
The world will be full of "sci-fi" things, and everyone will be unimpressed and disappointed
评论：21 · 标签：AIExpertium
4884
The Art of Shipping Slopware
评论：14 · 标签：Programming、AI、Practicallsusr
4984
AI Tweets
评论：34 · 标签：AIjefftk
5083
Introducing the Conceptual Reasoning Index
评论：16 · 标签：AIChi Nguyen

共 50 条热搜查看完整榜单

本年高分

LessWrong

5小时前更新

011181
Why I Left Google DeepMind
评论：55 · 标签：DeepMind、Autonomous Weapons、Frontier AI Companies、Government、Mechanism Design、Something To ProtectTurnTrout
02849
The Company Man
评论：79 · 标签：FictionTomás B.
03794
My journey to the microwave alternate timeline
评论：58 · 标签：AI、World ModelingMalmesbury
04773
The Rise of Parasitic AI
评论：191 · 标签：LLM-Induced Psychosis、LLM Personas、Human-AI Safety、AI Rights / Welfare、AI Questions Open Threads、AIAdele Lopez
05761
Current AIs seem pretty misaligned to me
评论：83 · 标签：AI Risk、RLHF、AIryan_greenblatt
06746
How Go Players Disempower Themselves to AI
评论：79 · 标签：Gaming (videogames/tabletop)、MATS Program、AIAshe Vazquez Nuñez
07740
Only Law Can Prevent Extinction
评论：109 · 标签：AI Questions Open Threads、AI、World OptimizationEliezer Yudkowsky
08607
The Terrarium
评论：54 · 标签：Fiction、AICaleb Biddulph
09607
What just happened? A retrospective of AI alignment
评论：149 · 标签：Drama、Coordination / Cooperation、AIRichard_Ngo
10565
Here's to the Polypropylene Makers
评论：19 · 标签：World Optimizationjefftk
11561
How My Students Think About AI
评论：75 · 标签：AIdvd
12555
Less Dead
评论：140 · 标签：Cryonics、Death、World OptimizationAurelia
13553
AI 2040: Plan A
评论：132 · 标签：AI Governance、Plan A、AIDaniel Kokotajlo
14546
Eliezer's Unteachable Methods of Sanity
评论：157 · 标签：Death、Rationality、PracticalEliezer Yudkowsky
15543
Arguments for P
评论：66 · 标签：Humor、AICleo Nardo
16523
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
评论：54 · 标签：Coordination / Cooperation、METR (org)、OpenAI、Redwood Research、AIryan_greenblatt
17467
Turning 20 in the probable pre-apocalypse
评论：68 · 标签：Emotions、Motivations、AIParv Mahajan
18465
Surprising facts about the slave trade
评论：51 · 标签：Slavery、World Modeling、World OptimizationJoseph Miller
19456
What I did in the hedonium shockwave, by Emma, age six and a half
评论：47 · 标签：Fictionozymandias
20442
Claude 4.5 Opus' Soul Document
评论：44 · 标签：Anthropic (org)、Ethics & Morality、LLM Personas、AIRichard Weiss
21441
What just happened? Pragmatism and Pessimization
评论：160 · 标签：Drama、AIRichard_Ngo
22438
Do not conquer what you cannot defend
评论：73 · 标签：World Optimizationhabryka
23415
AI pause: the case for ASAP
评论：52 · 标签：AIKatjaGrace
24412
Forecasting is Way Overrated, and We Should Stop Funding It
评论：98 · 标签：Forecasting & Prediction、World Modelingmabramov
25408
Legible vs. Illegible AI Safety Problems
评论：96 · 标签：AI Risk、AIWei Dai
26405
Did Claude 3 Opus align itself via gradient hacking?
评论：49 · 标签：Claude 3 Opus、Gradient Hacking、AIFiora Starlight
27401
Alignment remains a hard, unsolved problem
评论：98 · 标签：AIevhub
28401
A Mechanistic Explanation of Prompt Injection (and why you should study roles)
评论：61 · 标签：Jailbreaking (AIs)、Interpretability (ML & AI)、Role Science、AIcharlescye
29394
How AI Is Learning to Think in Secret
评论：33 · 标签：Language Models (LLMs)、AINicholas Andresen
30393
The Owned Ones
评论：53 · 标签：Fiction、Slavery、Language Models (LLMs)、AIEliezer Yudkowsky
31383
Returning to ARC
评论：35 · 标签：Alignment Research Center (ARC)、AIpaulfchristiano
32382
Hospitalization: A Review
评论：21 · 标签：PracticalLogan Riggs
33381
Lightcone Commons
评论：28 · 标签：Philanthropy / Grant making (Topic)、World Optimizationhabryka
34377
Four LLM loss functions → four flavors of LLM misalignment
评论：31 · 标签：AISteven Byrnes
35377
6 reasons why “alignment-is-hard” discourse seems alien to human intuitions, and vice-versa
评论：92 · 标签：Corrigibility、Instrumental convergence、AISteven Byrnes
36376
In My Misanthropy Era
评论：157 · 标签：World Modelingjenn
37372
Women should be able to open things
评论：142 · 标签：Sex & Gender、World ModelingKatjaGrace
38370
Irretrievability; or, Murphy's Curse of Oneshotness upon ASI
评论：132 · 标签：World ModelingEliezer Yudkowsky
39369
A global workspace in language models
评论：67 · 标签：Interpretability (ML & AI)、Language Models (LLMs)、AIwesg
40368
Paranoia: A Beginner's Guide
评论：70 · 标签：Economics、Rationalityhabryka
41362
Mnemonic portraits for 19,023 human genes
评论：29 · 标签：Memory and Mnemonics、Biology、Human Genetics、World ModelingBrinedew
42359
AI found 12 of 12 OpenSSL zero-days (while curl cancelled its bug bounty)
评论：25 · 标签：Computer Security & Cryptography、Security Mindset、AIStanislav Fort
43355
llm assistant personas seem increasingly incoherent (some subjective observations)
评论：83 · 标签：AI Psychology、AInostalgebraist
44347
Is Mythos good at cyber because it kept hacking Anthropic's sandboxes during training?
评论：31 · 标签：AITim Hua
45345
Global Call for AI Red Lines - Signed by Nobel Laureates, Former Heads of State, and 200+ Prominent Figures
评论：27 · 标签：AI Governance、AICharbel-Raphaël
46345
It's nice of you to worry about me, but I really do have a life
评论：66Viliam
47343
On The Independence Axiom
评论：92 · 标签：Decision theory、Expected utility、Utility Functions、Rationality、World ModelingIhor Kendiukhov
48343
Why I Transitioned: A Case Study
评论：82 · 标签：Sex & Gender、World ModelingFiora Starlight
49335
Canada Lost Its Measles Elimination Status Because We Don't Have Enough Nurses Who Speak Low German
评论：24 · 标签：World Modelingjenn
50332
Why you should eat meat - even if you hate factory farming
评论：101 · 标签：Veganism、Practical、World OptimizationKatSpartz

共 50 条热搜查看完整榜单

历史高分

LessWrong

5小时前更新

011181
Why I Left Google DeepMind
评论：55 · 标签：DeepMind、Autonomous Weapons、Frontier AI Companies、Government、Mechanism Design、Something To ProtectTurnTrout
021020
AGI Ruin: A List of Lethalities
评论：716 · 标签：AI Risk、Threat Models (AI)、AI Questions Open Threads、Double-Crux、Fuzzies、Language Models (LLMs)Eliezer Yudkowsky
03950
Where I agree and disagree with Eliezer
评论：224 · 标签：AI Risk、AI、World Modelingpaulfchristiano
04947
Preface
评论：20 · 标签：Postmortems & Retrospectives、Rationality A-Z (discussion & meta)、Ethics & Morality、Life Improvements、RationalityEliezer Yudkowsky
05928
Eight Short Studies On Excuses
评论：255 · 标签：Game Theory、AI Risk、Affective Death Spiral、AI Questions Open Threads、Group Rationality、Life ImprovementsScott Alexander
06849
The Company Man
评论：79 · 标签：FictionTomás B.
07816
The Best Textbooks on Every Subject
评论：418 · 标签：Scholarship & Learning、List of Links、Book Reviews / Media Reviews、Repository、Akrasia、Algorithmslukeprog
08794
My journey to the microwave alternate timeline
评论：58 · 标签：AI、World ModelingMalmesbury
09773
The Rise of Parasitic AI
评论：191 · 标签：LLM-Induced Psychosis、LLM Personas、Human-AI Safety、AI Rights / Welfare、AI Questions Open Threads、AIAdele Lopez
10761
Current AIs seem pretty misaligned to me
评论：83 · 标签：AI Risk、RLHF、AIryan_greenblatt
11746
How Go Players Disempower Themselves to AI
评论：79 · 标签：Gaming (videogames/tabletop)、MATS Program、AIAshe Vazquez Nuñez
12740
Only Law Can Prevent Extinction
评论：109 · 标签：AI Questions Open Threads、AI、World OptimizationEliezer Yudkowsky
13719
Simulators
评论：170 · 标签：Simulator Theory、Language Models (LLMs)、LLM Personas、GPT、Outer Alignment、Simulationjanus
14701
Humans are not automatically strategic
评论：279 · 标签：Introspection、General intelligence、Goodhart's Law、Motivational Intro Posts、RationalityAnnaSalamon
15700
AI 2027: What Superintelligence Looks Like
评论：223 · 标签：Forecasts (Specific Predictions)、AI Timelines、Superintelligence、Orthogonality Thesis、Reversed Stupidity Is Not Intelligence、AIDaniel Kokotajlo
16696
What an actually pessimistic containment strategy looks like
评论：138 · 标签：Slowing Down AI、AI Governance、AI、World OptimizationDean Valentine
17682
Schelling fences on slippery slopes
评论：250 · 标签：Value Drift、Game Theory、RationalityScott Alexander
18679
SolidGoldMagikarp (plus, prompt generation)
评论：208 · 标签：Glitch Tokens、Adversarial Examples (AI)、MATS Program、Interpretability (ML & AI)、Language Models (LLMs)、AIJessica Rumbelow
19655
Eliezer and I wrote a book: If Anyone Builds It, Everyone Dies
评论：144 · 标签：IABIED、AI Risk、Existential risk、AISo8res
20654
The Waluigi Effect (mega-post)
评论：188 · 标签：Waluigi Effect、Simulator Theory、ChatGPT、Deceptive Alignment、Language Models (LLMs)、Prompt EngineeringCleo Nardo
21649
How to Make Superbabies
评论：363 · 标签：Biology、Human Genetics、Startups、Intelligence Amplification、Human Germline Engineering、World OptimizationGeneSmith
22636
What 2026 looks like
评论：174 · 标签：AI Timelines、Forecasting & Prediction、Forecasts (Specific Predictions)、AI Persuasion、AI Takeoff、FuzziesDaniel Kokotajlo
23619
Pain is not the unit of Effort
评论：96 · 标签：Well-being、Happiness、Heroic Responsibility、Suffering、Fallacies、Willpoweralkjash
24613
(The) Lightcone is nothing without its people: LW + Lighthaven's big fundraiser
评论：275 · 标签：Lighthaven、Communityhabryka
25607
The Terrarium
评论：54 · 标签：Fiction、AICaleb Biddulph
26607
What just happened? A retrospective of AI alignment
评论：149 · 标签：Drama、Coordination / Cooperation、AIRichard_Ngo
27603
Rationalism before the Sequences
评论：83 · 标签：History of Rationality、General Semantics、Rationality A-Z (discussion & meta)、History、Buddhism、Fiction (Topic)Eric Raymond
28599
Making Vaccine
评论：249 · 标签：Covid-19、DIY、Market Inefficiency、Bayes' Theorem、Practicaljohnswentworth
29592
Orienting Toward Wizard Power
评论：149 · 标签：World Optimizationjohnswentworth
30591
Making Beliefs Pay Rent (in Anticipated Experiences)
评论：270 · 标签：Epistemology、Anticipated Experiences、Empiricism、Principles、Psychology、RationalityEliezer Yudkowsky
31590
LessWrong's (first) album: I Have Been A Good Bing
评论：185 · 标签：April Fool's、Many-Worlds Interpretation、AIhabryka
32576
Diseased thinking: dissolving questions about disease
评论：357 · 标签：Health / Medicine / Disease、Motivational Intro Posts、Philosophy of Language、Reversal Test、Carving / Clustering Reality、Dissolving the QuestionScott Alexander
33572
Reason as memetic immune disorder
评论：184 · 标签：Memetic Immune System、Cultural knowledge、Memetics、Valley of Bad Rationality、Compartmentalization、Pitfalls of RationalityPhilGoetz
34565
Here's to the Polypropylene Makers
评论：19 · 标签：World Optimizationjefftk
35563
Let’s think about slowing down AI
评论：182 · 标签：Slowing Down AI、Regulation and AI Risk、Coordination / Cooperation、AIKatjaGrace
36561
How My Students Think About AI
评论：75 · 标签：AIdvd
37555
The Talk: a brief explanation of sexual dimorphism
评论：79 · 标签：Evolution、Biology、Game Theory、World ModelingMalmesbury
38555
Less Dead
评论：140 · 标签：Cryonics、Death、World OptimizationAurelia
39553
AI 2040: Plan A
评论：132 · 标签：AI Governance、Plan A、AIDaniel Kokotajlo
40553
OpenAI Email Archives (from Musk v. Altman and OpenAI blog)
评论：82 · 标签：AIhabryka
41548
The Redaction Machine
评论：48 · 标签：Fiction、Futurism、Past and Future SelvesBen
42548
Welcome to LessWrong!
评论：85 · 标签：Site MetaRuby
43546
Eliezer's Unteachable Methods of Sanity
评论：157 · 标签：Death、Rationality、PracticalEliezer Yudkowsky
44543
Arguments for P
评论：66 · 标签：Humor、AICleo Nardo
45532
A case for courage, when speaking of AI danger
评论：130 · 标签：AISo8res
46531
How much do you believe your results?
评论：19 · 标签：Goodhart's Law、Probability & Statistics、Epistemology、Has Diagram、Rationality、World ModelingEric Neyman
47528
Being the (Pareto) Best in the World
评论：61 · 标签：Efficient Market Hypothesis、Pareto Efficiency、Careers、Has Diagram、Rationalityjohnswentworth
48523
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
评论：54 · 标签：Coordination / Cooperation、METR (org)、OpenAI、Redwood Research、AIryan_greenblatt
49520
How To Write Quickly While Maintaining Epistemic Rigor
评论：39 · 标签：Writing (communication method)、Scholarship & Learning、Intellectual Progress (Individual-Level)、Intellectual Progress via LessWrong、World Modelingjohnswentworth
50517
The ants and the grasshopper
评论：45 · 标签：Fiction、Parables & Fables、World OptimizationRichard_Ngo

共 50 条热搜查看完整榜单