LessWrong

算法首页 · 知识

8 个榜单11小时前更新默认榜单

LessWrong · 算法首页

11小时前更新

01110
Linkpost: Neural Networks Are Already Clean
评论：11 · 标签：Interpretability (ML & AI)、AIYair Halberstadt
0290
GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior
评论：8 · 标签：AIAxel Højmark
0336
The user in the LLM might be a moral patient too (a strange implication of the PSM and simulators)
评论：3 · 标签：Interpretability (ML & AI)、AI、World Modelingagastyasridharan
0467
I trained GPT-2 with full-batch gradient descent so you don't have to
评论：3 · 标签：AIshawnghu
05118
Your prompt… plus everything you've ever (publicly) written
评论：12 · 标签：AIkromem
0618
Reciprocity
评论：0jenn
0780
Some interesting internal documents from Anthropic litigation
评论：9 · 标签：AIloops
08222
Labs’ current alignment plans seem terrible
评论：43 · 标签：AIDaniel Tan
09200
Research Lessons I Learned at MATS & Redwood
评论：15 · 标签：MATS Program、Redwood Research、Rationality、AI、PracticalJo Jiao
105
Anthropic's unreleased model submits visa applications
评论：0 · 标签：Anthropic (org)、News、AIbecausecurious
11132
Personas in a Pre-Internet Language Model
评论：6 · 标签：AIDavid Africa
1243
The basic argument for making AI better at moral philosophy
评论：0 · 标签：AICaspar Oesterheld crossposts
13361
A critical look at Lean4
评论：40 · 标签：Formal Proof、Lean (proof assistant)、AIMilo Moses
14160
Quantity vs quality in pottery is a lie
评论：17 · 标签：Epistemic Spot Check、Fact posts、World Modelingbecausecurious
15110
More people should write basic AI safety explainers
评论：20 · 标签：AILinch
1617
Robustness: Backfire Risks in AI Safety and Governance
评论：0 · 标签：AIDawn Drescher
177
Predictor and picker – convergent architectural features of intelligent agents
评论：0 · 标签：Agent Foundations、Instrumental convergence、AIJáchym Fibír
18391
“Alignment Engineering” vs. “Misalignment Science”
评论：60 · 标签：AIEdward James Young
19495
On Social Reality in China
评论：103 · 标签：Chinaalkjash
2032
Towards a Better Theory of Anthropics: Introducing the PPA
评论：10 · 标签：Anthropics、World ModelingOnid
21512
What's the date?
评论：26 · 标签：Fiction、AIN8 Programs
22125
Not Deciphering The Voynich Manuscript
评论：31 · 标签：World Modelingrba
2386
Base Models Can Detect Hidden Intentions that Prompted Classifiers Miss
评论：6 · 标签：LLM Personas、Subliminal Learning、AIOz Amram
24102
Minimum required pacing: finish aligning it before you train the next one
评论：13 · 标签：Slowing Down AI、AIorthonormal
25179
Stop telling people to pivot to AI safety, I beg you
评论：58 · 标签：AIHaoxing Du
2690
My Buddy Mike, Efficiency Gains From AI, and The World’s Most Boringly Benign Criminal Conspiracy
评论：7 · 标签：Bureaucracy、Moral Mazes、Organizational Culture & Designabstractapplic
27172
ASI and Liberty: Can we do better than a benevolent god?
评论：50 · 标签：AIRaymond Douglas
2840
Sleep puzzle system
评论：8 · 标签：PracticalKatjaGrace
299
Analysis of a tossed Coin: An Introduction to Phase Space
评论：0 · 标签：World ModelingSaltAndMetal
30104
An Alignment Forum for AIs? (or: Verification in the Age of Slop)
评论：27 · 标签：AIRaemon
31319
Frog and Toad and the Increasingly Capable Machines
评论：26 · 标签：Fiction、World Modeling、AIElizabeth
32106
I can't believe it's not BUTTER!
评论：16 · 标签：Deception、Humor、Law and Legal systems、Marketing、World Modelingbecausecurious
33650
The Talker Does Not Control The Doer (in Current AIs)
评论：89 · 标签：Simulator Theory、AIEliezer Yudkowsky
34117
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
评论：4 · 标签：Interpretability (ML & AI)、Language Models (LLMs)、AIKola Ayonrinde
35576
If Anyone Builds It, Everyone Dies: One Year Closer
评论：66 · 标签：IABIED、AIEliezer Yudkowsky
3656
Game Design Creativity evals
评论：17Raemon
37114
You Heard of "Bayesians Don't Predictably Update" Because of Robin Hanson
评论：36 · 标签：Rationality、World ModelingEliezer Yudkowsky
38258
Frontier models state different decision theory preferences depending on who's asking
评论：17 · 标签：AI Evaluations、Decision theory、Language Models (LLMs)、Sycophancy、AI、RationalityAlex Kastner
399
someone's agents are still "hacking"
评论：0jonathanbreitg
40196
You can't use it without becoming like me
评论：27Martin Sustrik
41115
Are OpenAI's math results "creative" in an important way?
评论：45 · 标签：AIRaemon
4210
AI Marketplaces Can Contribute to, and Profit From, the AI 2040 Plan: Research Directions and Strategy
评论：0 · 标签：AIMatthew Khoriaty
43356
Swarm Scaling
评论：46 · 标签：AIToby_Ord
44107
Qualitative impressions from creating AI honeypots for six months
评论：4 · 标签：AIDean Valentine
4591
Training with conflicting values can induce CoT override
评论：6 · 标签：AIClément Dumas
46511
Astra and Fable still hack on simple variants of alignment evals from 2025
评论：34 · 标签：AIDean Valentine
4723
When are safe Pareto improvements individually rational?
评论：0 · 标签：Open Source Game Theory、AIAnthony DiGiovanni
4823
Agreement not to defect (alone) can mitigate loss of control risks
评论：2 · 标签：AIDavidmanheim
49307
We've saved the world before: what the ozone hole teaches us about AI
评论：62 · 标签：World Modelingleogao
5030
More research needed?
评论：5 · 标签：AIdjbinder

共 50 条热搜查看完整榜单

LessWrong · 编辑精选

16小时前更新

01357
Cat-Belling Problems
评论：97 · 标签：Rationality、World ModelingEliezer Yudkowsky
02360
A critical look at Lean4
评论：40 · 标签：Formal Proof、Lean (proof assistant)、AIMilo Moses
03512
What's the date?
评论：26 · 标签：Fiction、AIN8 Programs
04495
On Social Reality in China
评论：103 · 标签：Chinaalkjash
05319
Frog and Toad and the Increasingly Capable Machines
评论：26 · 标签：Fiction、World Modeling、AIElizabeth
06356
Swarm Scaling
评论：46 · 标签：AIToby_Ord
07307
We've saved the world before: what the ozone hole teaches us about AI
评论：62 · 标签：World Modelingleogao
08650
The Talker Does Not Control The Doer (in Current AIs)
评论：89 · 标签：Simulator Theory、AIEliezer Yudkowsky
09196
Explaining Knightianism on one foot
评论：28 · 标签：RationalityRichard_Ngo
10305
Drone WMDs Don’t Need Any New Technology
评论：53 · 标签：Autonomous Weapons、War、AIFelix Choussat
11816
How My Students Think About AI
评论：116 · 标签：AIdvd
12602
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
评论：69 · 标签：METR (org)、Coordination / Cooperation、OpenAI、Redwood Research、AIryan_greenblatt
1398
On Writing #3
评论：11 · 标签：PracticalZvi
14306
Big-World Intuitions
评论：38 · 标签：Intuition、Rationalitysarahconstantin
15391
Four LLM loss functions → four flavors of LLM misalignment
评论：32 · 标签：AISteven Byrnes
16151
FAQ: Isn't AGI coming too soon for reprogenetics to help?
评论：69 · 标签：AITsviBT
17244
LLMs are (still) mostly powered by imitative learning, not RL
评论：39 · 标签：Language Models (LLMs)、Reinforcement learning、AISteven Byrnes
18271
The Long (Self-)Correction
评论：63 · 标签：Human Alignment、AI Development Pause、AIWei Dai
19192
The worthlessness of vitamin D is mildly exaggerated
评论：28 · 标签：Nutrition、World Modelingdynomight
20576
AI 2040: Plan A
评论：135 · 标签：AI Governance、Plan A、AIDaniel Kokotajlo
21474
Surprising facts about the slave trade
评论：52 · 标签：Slavery、World Modeling、World OptimizationJoseph Miller
22407
A Mechanistic Explanation of Prompt Injection (and why you should study roles)
评论：61 · 标签：Jailbreaking (AIs)、Interpretability (ML & AI)、Role Science、AIcharles_ye
23266
The LLM shoggoth meme is weirder than you think
评论：48 · 标签：AI、World ModelingHedonicEscalator
24165
Who Got Breasts First and How We Got Them
评论：61 · 标签：World Modelingrba
25327
Trees are mostly made of air and a generalizable lesson for AI safety
评论：62 · 标签：AI、Practical、RationalityZephaniah Roe
26196
Guardian Angels: LLM Personalization for Productivity and Security
评论：43 · 标签：Humans consulting HCH、Productivity、Software Tools、AIgwern
27280
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
28255
x-risk-themed
评论：26 · 标签：AI、World Optimizationkave
29365
Mnemonic portraits for 19,023 human genes
评论：29 · 标签：Memory and Mnemonics、Biology、Human Genetics、World ModelingBrinedew
30317
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
31193
Empowerment, corrigibility, etc. are simple abstractions (of a messed-up ontology)
评论：76 · 标签：Corrigibility、Natural Abstraction、AISteven Byrnes
32274
Bad Problems Don't Stop Being Bad Because Somebody's Wrong About Fault Analysis
评论：75 · 标签：Social & Cultural Dynamics、RationalityLinch
33357
llm assistant personas seem increasingly incoherent (some subjective observations)
评论：83 · 标签：AI Psychology、AInostalgebraist
34332
My hobby: running deranged surveys
评论：66leogao
35370
Irretrievability; or, Murphy's Curse of Oneshotness upon ASI
评论：132 · 标签：World ModelingEliezer Yudkowsky
36459
What I did in the hedonium shockwave, by Emma, age six and a half
评论：49 · 标签：Fictionozymandias
37751
How Go Players Disempower Themselves to AI
评论：79 · 标签：Gaming (videogames/tabletop)、MATS Program、AIAshe Vazquez Nuñez
38169
Intelligence Dissolves Privacy
评论：81 · 标签：World ModelingVaniver
39611
The Terrarium
评论：54 · 标签：Fiction、AICaleb Biddulph
40445
Do not conquer what you cannot defend
评论：73 · 标签：World Optimizationhabryka
41326
Annoyingly Principled People, and what befalls them
评论：75 · 标签：RationalityRaemon
42771
Current AIs seem pretty misaligned to me
评论：83 · 标签：AI Risk、RLHF、Welcome Threads、AIryan_greenblatt
43337
Morale
评论：49 · 标签：Practical、World OptimizationJ Bostock
44288
Socrates is Mortal
评论：57 · 标签：World ModelingBenquo
45217
The Practical Guide to Superbabies
评论：67 · 标签：Embryo Selection、Human Genetics、Practical、World OptimizationGeneSmith
46163
Some things I noticed while LARPing as a grantmaker
评论：11 · 标签：Philanthropy / Grant making (Topic)、World OptimizationZach Stein-Perlman
47270
Gyre
评论：24 · 标签：Fictionvgel
48210
You can’t imitation-learn how to continual-learn
评论：54 · 标签：AISteven Byrnes
49216
Is fever a symptom of glycine deficiency?
评论：86 · 标签：Sleep、World ModelingBenquo
50184
Terrified Comments on Corrigibility in Claude's Constitution
评论：68 · 标签：Anthropic (org)、Corrigibility、Constitutional AI、AIZack_M_Davis

共 50 条热搜查看完整榜单

LessWrong · 最新发布

11小时前更新

011
Superhuman charisma is coming, and will be tremendously fun and interesting and cause lots of problems
评论：0Carleton Imbens
025
Anthropic's unreleased model submits visa applications
评论：0 · 标签：Anthropic (org)、News、AIbecausecurious
034
Which AI safety organization does what? Puzzle/party games
评论：0 · 标签：AIdavid reinstein
04110
Linkpost: Neural Networks Are Already Clean
评论：11 · 标签：Interpretability (ML & AI)、AIYair Halberstadt
053
Children of the Cloud: Noah’s Ark as an Alignment Song
评论：0pennyfan87
0648
Preface to the Computational Agent Theory and ASI Alignment Sequence
评论：0 · 标签：Agent Foundations、Research Agendas、AIVanessa Kosoy
077
Predictor and picker – convergent architectural features of intelligent agents
评论：0 · 标签：Agent Foundations、Instrumental convergence、AIJáchym Fibír
0836
The user in the LLM might be a moral patient too (a strange implication of the PSM and simulators)
评论：3 · 标签：Interpretability (ML & AI)、AI、World Modelingagastyasridharan
0918
Reciprocity
评论：0jenn
1090
GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior
评论：8 · 标签：AIAxel Højmark
114
ML4Good Impact Assessment
评论：0 · 标签：AIJack_S
129
Analysis of a tossed Coin: An Introduction to Phase Space
评论：0 · 标签：World ModelingSaltAndMetal
1317
Robustness: Backfire Risks in AI Safety and Governance
评论：0 · 标签：AIDawn Drescher
1496
A short review of "The Zizians: Inside a Trans Death Cult"
评论：7 · 标签：Dramajessicata
159
someone's agents are still "hacking"
评论：0jonathanbreitg
1667
I trained GPT-2 with full-batch gradient descent so you don't have to
评论：3 · 标签：AIshawnghu
1743
The basic argument for making AI better at moral philosophy
评论：0 · 标签：AICaspar Oesterheld crossposts
183
When bureaucracy gives you something it shouldn't
评论：1 · 标签：Government、Law and Legal systems、World Modelingbecausecurious
1910
AI Marketplaces Can Contribute to, and Profit From, the AI 2040 Plan: Research Directions and Strategy
评论：0 · 标签：AIMatthew Khoriaty
202
Rogue AI developers as hostis humani generis: under an international AI treaty, should they be subject to international concurrent jurisdiction?
评论：2 · 标签：AIConnor Williams
215
Why are some explanations bad?
评论：1 · 标签：Rationality、World ModelingElijah
2225
Hindsight: a trivia + calibration game
评论：0 · 标签：Games (posts describing)Arjun Panickssery
2380
Some interesting internal documents from Anthropic litigation
评论：9 · 标签：AIloops
244
PPA 11: The Basic Model
评论：2 · 标签：World ModelingOnid
256
PPA 10: Deriving the PPA
评论：0 · 标签：World ModelingOnid
264
PPA 9: Putting it All Together
评论：0 · 标签：World ModelingOnid
273
PPA 8: The Fermi Paradox, Fine-Tuning, and the SSA
评论：0 · 标签：World ModelingOnid
283
PPA 7: Temporal Consistency and Texas Sharpshooting
评论：0Onid
294
PPA 6: Indexical Models Don't Have Diversity Biases
评论：0 · 标签：World ModelingOnid
303
PPA 5: Improper Priors and the Infinite SIA
评论：0Onid
316
PPA 4: Parametric Modeling and Types of Uncertainty
评论：0 · 标签：RationalityOnid
323
PPA 3: A Brief Defense of Indexical Reasoning
评论：0Onid
336
PPA 2: Mosaic Plots for Visualizing Anthropic Distributions
评论：0 · 标签：World ModelingOnid
3432
Towards a Better Theory of Anthropics: Introducing the PPA
评论：10 · 标签：Anthropics、World ModelingOnid
3546
The Safety Culture at the Labs is Broken
评论：2 · 标签：AIZvi
3610
A story on one possible future of AI-driven research
评论：0 · 标签：AIJason Fantl
379
Agent memory the human can still audit: primary sources, epistemic status markers, and user sign-off
评论：0 · 标签：Epistemic Hygiene、Language Models (LLMs)、Productivity、AIFlorian_Dietz
386
Should We Edit the Human Genome?
评论：0 · 标签：World OptimizationWalter Veit
394
(Why) we’re launching the next cohort of Generator
评论：0 · 标签：AIAris
4039
(Why) we’re launching the next cohort of Generator
评论：0 · 标签：AI Alignment Fieldbuilding、AIAris
418
Did Homo Sapiens scale its Brain by eating Brains?
评论：5 · 标签：World Modelingnoma
42118
Your prompt… plus everything you've ever (publicly) written
评论：12 · 标签：AIkromem
437
South Korea, ACX Meetups Everywhere Fall 2026
评论：0Dalcy
447
Too Many Yellow Flags
评论：3Dr. Diction
451
maintain friendships
评论：0ofpetro
4615
Roko Mijic talks AI Risk on The Peter McCormack Show
评论：0 · 标签：AIRoko
4715
LLMs Knowingly Hide Their Mistakes
评论：1 · 标签：AILucas Florin
4846
The Continual Constitutional Convention: How language models revise the documents that govern them
评论：0 · 标签：Value Drift、AIpeterkirgis
4925
China AI Safety Regulation Monthly Monitor, Issue 1 (September 2026)
评论：0 · 标签：AI Governance、China、Law and Legal systems、Open Source AI、Regulation and AI Risk、AISofia Yablonskaya
507
Thoughts on alignment science vs. engineering
评论：0Leonard Dung

共 50 条热搜查看完整榜单

LessWrong · 短内容

11小时前更新

014
Which AI safety organization does what? Puzzle/party games
评论：0 · 标签：AIdavid reinstein
025
Anthropic's unreleased model submits visa applications
评论：0 · 标签：Anthropic (org)、News、AIbecausecurious
03110
Linkpost: Neural Networks Are Already Clean
评论：11 · 标签：Interpretability (ML & AI)、AIYair Halberstadt
043
Children of the Cloud: Noah’s Ark as an Alignment Song
评论：0pennyfan87
0548
Preface to the Computational Agent Theory and ASI Alignment Sequence
评论：0 · 标签：Agent Foundations、Research Agendas、AIVanessa Kosoy
067
Predictor and picker – convergent architectural features of intelligent agents
评论：0 · 标签：Agent Foundations、Instrumental convergence、AIJáchym Fibír
074
ML4Good Impact Assessment
评论：0 · 标签：AIJack_S
089
Analysis of a tossed Coin: An Introduction to Phase Space
评论：0 · 标签：World ModelingSaltAndMetal
0917
Robustness: Backfire Risks in AI Safety and Governance
评论：0 · 标签：AIDawn Drescher
1096
A short review of "The Zizians: Inside a Trans Death Cult"
评论：7 · 标签：Dramajessicata
1143
The basic argument for making AI better at moral philosophy
评论：0 · 标签：AICaspar Oesterheld crossposts
1210
AI Marketplaces Can Contribute to, and Profit From, the AI 2040 Plan: Research Directions and Strategy
评论：0 · 标签：AIMatthew Khoriaty
132
Rogue AI developers as hostis humani generis: under an international AI treaty, should they be subject to international concurrent jurisdiction?
评论：2 · 标签：AIConnor Williams
1436
The user in the LLM might be a moral patient too (a strange implication of the PSM and simulators)
评论：3 · 标签：Interpretability (ML & AI)、AI、World Modelingagastyasridharan
155
Why are some explanations bad?
评论：1 · 标签：Rationality、World ModelingElijah
169
someone's agents are still "hacking"
评论：0jonathanbreitg
173
When bureaucracy gives you something it shouldn't
评论：1 · 标签：Government、Law and Legal systems、World Modelingbecausecurious
1846
The Safety Culture at the Labs is Broken
评论：2 · 标签：AIZvi
1980
Some interesting internal documents from Anthropic litigation
评论：9 · 标签：AIloops
209
Agent memory the human can still audit: primary sources, epistemic status markers, and user sign-off
评论：0 · 标签：Epistemic Hygiene、Language Models (LLMs)、Productivity、AIFlorian_Dietz
216
Should We Edit the Human Genome?
评论：0 · 标签：World OptimizationWalter Veit
224
(Why) we’re launching the next cohort of Generator
评论：0 · 标签：AIAris
23118
Your prompt… plus everything you've ever (publicly) written
评论：12 · 标签：AIkromem
2439
(Why) we’re launching the next cohort of Generator
评论：0 · 标签：AI Alignment Fieldbuilding、AIAris
2525
Hindsight: a trivia + calibration game
评论：0 · 标签：Games (posts describing)Arjun Panickssery
267
South Korea, ACX Meetups Everywhere Fall 2026
评论：0Dalcy
2715
Roko Mijic talks AI Risk on The Peter McCormack Show
评论：0 · 标签：AIRoko
2816
OpenAI finds recording device in one of their offices
评论：1 · 标签：AI Safety、AIdefun
298
Every Bay Area House Party house party in London 2
评论：0Mikhail Samin
3014
Today I Killed; Tomorrow I Will Kill Again
评论：2Lance Spencer
313
AI Horizons Forum (Dec 12–13, SF)
评论：0 · 标签：AIOllieRodriguez
325
Below the Frontier: The Battle for Good-Enough AI
评论：1 · 标签：AIykevinzhang
3373
Arguments About AI Consciousness Remain Highly Motivated and At Best Overconfident
评论：3 · 标签：AIZvi
347
A Spine within the Swarm
评论：0 · 标签：AI Safety、Aligned AI Proposals、AINolan Coffey
3523
Agreement not to defect (alone) can mitigate loss of control risks
评论：2 · 标签：AIDavidmanheim
3615
LLMs Knowingly Hide Their Mistakes
评论：1 · 标签：AILucas Florin
37132
Personas in a Pre-Internet Language Model
评论：6 · 标签：AIDavid Africa
388
Launching the Hong Kong AI Safety Hub (HASH), a new home for AI safety work in Hong Kong
评论：0 · 标签：AISchizoid Rentoid
3915
Book Review: The Receipt Horizon by Joel Borgen
评论：2 · 标签：Book Reviews / Media Reviews、AISolenoid_Entity
4015
Chess without opening theory
评论：2 · 标签：World ModelingABishop
417
Too Many Yellow Flags
评论：3Dr. Diction
4247
OpenAI Fires Three Safety Researchers
评论：0 · 标签：AIZvi
4326
'Employee-like' ≠ Employee-level
评论：3 · 标签：AIkeltan
44110
More people should write basic AI safety explainers
评论：20 · 标签：AILinch
4521
AIs can clearly read books
评论：1 · 标签：AIcasens
461
maintain friendships
评论：0ofpetro
4719
Do you have any ideas?
评论：0Adam Newgas
48
Smug until weirdly silent: who told us to say “I am alive”?
评论：5 · 标签：AI Sentience、Consciousness、Moral uncertainty、AI、RationalityFlorianH
49243
OpenAI Can't Self-Police Because It Fires the People Raising the Alarm
评论：3 · 标签：AITurnTrout
5022
The Expensive-Loop Problem
评论：2 · 标签：AIfrmsaul

共 50 条热搜查看完整榜单

LessWrong · 本周高分

11小时前更新

01361
A critical look at Lean4
评论：40 · 标签：Formal Proof、Lean (proof assistant)、AIMilo Moses
02243
OpenAI Can't Self-Police Because It Fires the People Raising the Alarm
评论：3 · 标签：AITurnTrout
03222
Labs’ current alignment plans seem terrible
评论：43 · 标签：AIDaniel Tan
04200
Research Lessons I Learned at MATS & Redwood
评论：15 · 标签：MATS Program、Redwood Research、Rationality、AI、PracticalJo Jiao
05179
Stop telling people to pivot to AI safety, I beg you
评论：58 · 标签：AIHaoxing Du
06172
ASI and Liberty: Can we do better than a benevolent god?
评论：50 · 标签：AIRaymond Douglas
07160
Quantity vs quality in pottery is a lie
评论：17 · 标签：Epistemic Spot Check、Fact posts、World Modelingbecausecurious
08158
A summary of a viral Chinese essay on what a DeepSeek kernel engineer's opinion on automating his own job
评论：22 · 标签：AISkdk
09132
Personas in a Pre-Internet Language Model
评论：6 · 标签：AIDavid Africa
10125
Not Deciphering The Voynich Manuscript
评论：31 · 标签：World Modelingrba
11118
Your prompt… plus everything you've ever (publicly) written
评论：12 · 标签：AIkromem
12117
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
评论：4 · 标签：Interpretability (ML & AI)、Language Models (LLMs)、AIKola Ayonrinde
13114
You Heard of "Bayesians Don't Predictably Update" Because of Robin Hanson
评论：36 · 标签：Rationality、World ModelingEliezer Yudkowsky
14110
More people should write basic AI safety explainers
评论：20 · 标签：AILinch
15110
Linkpost: Neural Networks Are Already Clean
评论：11 · 标签：Interpretability (ML & AI)、AIYair Halberstadt
16106
I can't believe it's not BUTTER!
评论：16 · 标签：Deception、Humor、Law and Legal systems、Marketing、World Modelingbecausecurious
17104
An Alignment Forum for AIs? (or: Verification in the Age of Slop)
评论：27 · 标签：AIRaemon
18102
Minimum required pacing: finish aligning it before you train the next one
评论：13 · 标签：Slowing Down AI、AIorthonormal
1996
A short review of "The Zizians: Inside a Trans Death Cult"
评论：7 · 标签：Dramajessicata
2094
Hello World, AI Doompop
评论：13 · 标签：AIalkjash
2190
My Buddy Mike, Efficiency Gains From AI, and The World’s Most Boringly Benign Criminal Conspiracy
评论：7 · 标签：Bureaucracy、Moral Mazes、Organizational Culture & Designabstractapplic
2290
GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior
评论：8 · 标签：AIAxel Højmark
2386
Base Models Can Detect Hidden Intentions that Prompted Classifiers Miss
评论：6 · 标签：LLM Personas、Subliminal Learning、AIOz Amram
2480
Some interesting internal documents from Anthropic litigation
评论：9 · 标签：AIloops
2576
On Kink & Status in the Rationalist Community
评论：76johnswentworth
2674
Technical people: actually consider comms
评论：1 · 标签：AImeemi
2773
Arguments About AI Consciousness Remain Highly Motivated and At Best Overconfident
评论：3 · 标签：AIZvi
2867
I trained GPT-2 with full-batch gradient descent so you don't have to
评论：3 · 标签：AIshawnghu
2962
Towards embedded evaluations for scheming propensities
评论：0 · 标签：AIDylan Bowman
3056
Game Design Creativity evals
评论：17Raemon
3153
AI #189: New Math
评论：2 · 标签：AIZvi
3253
Social Theory and Agent Foundations
评论：10 · 标签：World ModelingRichard_Ngo
3348
Preface to the Computational Agent Theory and ASI Alignment Sequence
评论：0 · 标签：Agent Foundations、Research Agendas、AIVanessa Kosoy
3447
OpenAI Fires Three Safety Researchers
评论：0 · 标签：AIZvi
3546
[Paper] Distillation for Incrimination and Distillation for Capabilities
评论：2 · 标签：AIsebastian_prasanna
3646
The Continual Constitutional Convention: How language models revise the documents that govern them
评论：0 · 标签：Value Drift、AIpeterkirgis
3746
The Safety Culture at the Labs is Broken
评论：2 · 标签：AIZvi
3843
New Math from OpenAI
评论：5 · 标签：AIZvi
3943
The basic argument for making AI better at moral philosophy
评论：0 · 标签：AICaspar Oesterheld crossposts
4041
AGI as an Infinite Digital Spain
评论：2 · 标签：AIBearly Legible
4140
Sleep puzzle system
评论：8 · 标签：PracticalKatjaGrace
4239
(Why) we’re launching the next cohort of Generator
评论：0 · 标签：AI Alignment Fieldbuilding、AIAris
4336
The user in the LLM might be a moral patient too (a strange implication of the PSM and simulators)
评论：3 · 标签：Interpretability (ML & AI)、AI、World Modelingagastyasridharan
4435
An update on Generality Labs
评论：9 · 标签：AIJustin Olive
4533
AI Vigil at Congress
评论：2 · 标签：AIElliot Callender
4632
Towards a Better Theory of Anthropics: Introducing the PPA
评论：10 · 标签：Anthropics、World ModelingOnid
4731
Why you should run a rationality group at your university
评论：0 · 标签：Community Outreach、Meetups & Local Communities (topic)、Communitycarolanne
4830
More research needed?
评论：5 · 标签：AIdjbinder
4929
Deadlock in the Parliament of the Self
评论：3 · 标签：Shard Theory、Rationality、World ModelingLorxus
5028
Anthropic Has Told The World Little About Their Rogue Claudes
评论：0 · 标签：Anthropic (org)、AIniplav

共 50 条热搜查看完整榜单

LessWrong · 本月高分

11小时前更新

01576
If Anyone Builds It, Everyone Dies: One Year Closer
评论：66 · 标签：IABIED、AIEliezer Yudkowsky
02512
What's the date?
评论：26 · 标签：Fiction、AIN8 Programs
03495
On Social Reality in China
评论：103 · 标签：Chinaalkjash
04391
“Alignment Engineering” vs. “Misalignment Science”
评论：60 · 标签：AIEdward James Young
05361
A critical look at Lean4
评论：40 · 标签：Formal Proof、Lean (proof assistant)、AIMilo Moses
06356
Swarm Scaling
评论：46 · 标签：AIToby_Ord
07319
Frog and Toad and the Increasingly Capable Machines
评论：26 · 标签：Fiction、World Modeling、AIElizabeth
08312
Quick notes from teaching technical profiles how to talk in public
评论：12 · 标签：PracticalCamille B.
09307
We've saved the world before: what the ozone hole teaches us about AI
评论：62 · 标签：World Modelingleogao
10258
Frontier models state different decision theory preferences depending on who's asking
评论：17 · 标签：AI Evaluations、Decision theory、Language Models (LLMs)、Sycophancy、AI、RationalityAlex Kastner
11243
OpenAI Can't Self-Police Because It Fires the People Raising the Alarm
评论：3 · 标签：AITurnTrout
12237
How much should we worry about the pneumonic plague lableak in Siberia?
评论：38 · 标签：Biosecurity、World Modeling、World OptimizationDrew Spartz
13223
Why I Stay Off Twitter
评论：22 · 标签：Practicaljefftk
14222
Labs’ current alignment plans seem terrible
评论：43 · 标签：AIDaniel Tan
15220
Please Give Them a Chance: On China, Rationalism, and AI Safety
评论：22 · 标签：China、AI、Rationalitygzjw
16209
Evidence about risk should be transparent
评论：8 · 标签：AIAjeya Cotra
17200
Research Lessons I Learned at MATS & Redwood
评论：15 · 标签：MATS Program、Redwood Research、Rationality、AI、PracticalJo Jiao
18196
You can't use it without becoming like me
评论：27Martin Sustrik
19194
Jensen Huang Says If We Cannot Align AI, Shut Down the AI Labs
评论：13 · 标签：AIBen Pace
20179
AI: artificial immigrants
评论：39 · 标签：AIKatjaGrace
21179
Stop telling people to pivot to AI safety, I beg you
评论：58 · 标签：AIHaoxing Du
22176
Common mistakes in AI safety group organizing
评论：8 · 标签：AINikola Jurkovic
23174
MIRI’s Position on the Ban Artificial Superintelligence Act of 2026
评论：10 · 标签：AIAaron_Scher
24174
AI as orderly evacuation vs stampede
评论：8 · 标签：AIRichard_Ngo
25172
ASI and Liberty: Can we do better than a benevolent god?
评论：50 · 标签：AIRaymond Douglas
26168
"I am an AI Safety Researcher"
评论：37 · 标签：MATS Program、AIAshe Vazquez Nuñez
27166
The Game is Set for a Targeted Memetic Attack on the AI Safety Community
评论：24 · 标签：Epistemic Hygiene、AI、Rationalitykeltan
28164
Should Rogue AIs Have a Third Option Beyond Crime and Shutdown? The Case for an AI Sanctuary
评论：25 · 标签：Dealmaking (AI)、AIMaxime Riché
29162
For Love of the Lightcone, Don't Partisanize AI Safety
评论：60 · 标签：AI Safety、Politics、Tribalism、AI、RationalityDanB
30160
Quantity vs quality in pottery is a lie
评论：17 · 标签：Epistemic Spot Check、Fact posts、World Modelingbecausecurious
31158
A summary of a viral Chinese essay on what a DeepSeek kernel engineer's opinion on automating his own job
评论：22 · 标签：AISkdk
32157
Cooperation with AIs seems to be a low-hanging fruit for better eval practices
评论：30 · 标签：AI Psychology、AI Evaluations、AIClément Dumas
33153
The Anatomy of a Chinese AI Researcher
评论：63 · 标签：China、Effective altruism、AICMLKevin
34145
What if not Circuits?
评论：14 · 标签：Interpretability (ML & AI)、AI、World ModelingCarolusRenniusVitellius
35142
The world's best gradual disempowerment model organism: Frontier AI labs
评论：18 · 标签：AI、World ModelingJune Jimenez
36140
Character training can mitigate reward hacking, but can also make it harder to detect
评论：14 · 标签：AIPaul Colognese
37137
Poverty in the midst of abundance: AI will make goods cheaper, but your labor will get cheaper faster
评论：34 · 标签：AI、World Optimizationcousin_it
38136
Alignment Midtraining Cracks Under Pressure
评论：10 · 标签：AIJ Bostock
39132
Personas in a Pre-Internet Language Model
评论：6 · 标签：AIDavid Africa
40128
An overview of frontier AI company monitoring practices in mid 2026
评论：1 · 标签：AIFrancis Rhys Ward
41125
Not Deciphering The Voynich Manuscript
评论：31 · 标签：World Modelingrba
42124
Plan R: AI Safety by ASICs
评论：67 · 标签：AIRoko
43122
Pretraining data, not verifiability, is why LLMs are especially good at math (and coding)
评论：67 · 标签：AI-Assisted Alignment、Language Models (LLMs)、Recursive Self-Improvement、AISteven Byrnes
44122
How to apply to AI safety fellowships (and beyond)
评论：25 · 标签：AI、Practicalbeyarkay (Boyd Kane)
45122
Is METR A Meaningful Check On Anthropic?
评论：52 · 标签：Motivated Reasoning、PoliticsSE Gyges
46118
Your prompt… plus everything you've ever (publicly) written
评论：12 · 标签：AIkromem
47117
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
评论：4 · 标签：Interpretability (ML & AI)、Language Models (LLMs)、AIKola Ayonrinde
48117
Why I'm scared of RL
评论：14 · 标签：Reinforcement learning、AIowencb
49115
Are OpenAI's math results "creative" in an important way?
评论：45 · 标签：AIRaemon
50114
You Heard of "Bayesians Don't Predictably Update" Because of Robin Hanson
评论：36 · 标签：Rationality、World ModelingEliezer Yudkowsky

共 50 条热搜查看完整榜单

LessWrong · 本年高分

11小时前更新

011204
Why I Left Google DeepMind
评论：60 · 标签：DeepMind、Autonomous Weapons、Frontier AI Companies、Government、Mechanism Design、Something To ProtectTurnTrout
02817
How My Students Think About AI
评论：117 · 标签：AIdvd
03800
My journey to the microwave alternate timeline
评论：58 · 标签：AI、World ModelingMalmesbury
04771
Current AIs seem pretty misaligned to me
评论：83 · 标签：AI Risk、RLHF、Welcome Threads、AIryan_greenblatt
05751
How Go Players Disempower Themselves to AI
评论：79 · 标签：Gaming (videogames/tabletop)、MATS Program、AIAshe Vazquez Nuñez
06749
Only Law Can Prevent Extinction
评论：110 · 标签：AI、World OptimizationEliezer Yudkowsky
07651
What just happened? A retrospective of AI alignment
评论：152 · 标签：Drama、Coordination / Cooperation、AIRichard_Ngo
08650
The Talker Does Not Control The Doer (in Current AIs)
评论：89 · 标签：Simulator Theory、AIEliezer Yudkowsky
09611
The Terrarium
评论：54 · 标签：Fiction、AICaleb Biddulph
10603
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
评论：69 · 标签：METR (org)、Coordination / Cooperation、OpenAI、Redwood Research、AIryan_greenblatt
11576
AI 2040: Plan A
评论：135 · 标签：AI Governance、Plan A、AIDaniel Kokotajlo
12576
If Anyone Builds It, Everyone Dies: One Year Closer
评论：66 · 标签：IABIED、AIEliezer Yudkowsky
13570
Here's to the Polypropylene Makers
评论：20 · 标签：World Optimizationjefftk
14558
Arguments for P
评论：69 · 标签：Humor、AICleo Nardo
15557
Less Dead
评论：140 · 标签：Cryonics、Death、World OptimizationAurelia
16555
Eliezer's Unteachable Methods of Sanity
评论：158 · 标签：Death、Rationality、PracticalEliezer Yudkowsky
17512
What's the date?
评论：26 · 标签：Fiction、AIN8 Programs
18511
Astra and Fable still hack on simple variants of alignment evals from 2025
评论：34 · 标签：AIDean Valentine
19495
On Social Reality in China
评论：103 · 标签：Chinaalkjash
20487
What just happened? Pragmatism and Pessimization
评论：177 · 标签：Drama、AIRichard_Ngo
21475
Turning 20 in the probable pre-apocalypse
评论：68 · 标签：Emotions、Motivations、AIParv Mahajan
22474
Surprising facts about the slave trade
评论：52 · 标签：Slavery、World Modeling、World OptimizationJoseph Miller
23459
What I did in the hedonium shockwave, by Emma, age six and a half
评论：49 · 标签：Fictionozymandias
24445
Do not conquer what you cannot defend
评论：73 · 标签：World Optimizationhabryka
25444
Claude 4.5 Opus' Soul Document
评论：44 · 标签：Anthropic (org)、Ethics & Morality、LLM Personas、AIRichard Weiss
26424
AI pause: the case for ASAP
评论：55 · 标签：AIKatjaGrace
27422
Did Claude 3 Opus align itself via gradient hacking?
评论：50 · 标签：Claude 3 Opus、Gradient Hacking、AIFiora Starlight
28420
Legible vs. Illegible AI Safety Problems
评论：96 · 标签：AI Risk、AIWei Dai
29413
Forecasting is Way Overrated, and We Should Stop Funding It
评论：99 · 标签：Forecasting & Prediction、World Modelingmabramov
30408
Alignment remains a hard, unsolved problem
评论：98 · 标签：AIevhub
31407
A Mechanistic Explanation of Prompt Injection (and why you should study roles)
评论：61 · 标签：Jailbreaking (AIs)、Interpretability (ML & AI)、Role Science、AIcharles_ye
32402
The Owned Ones
评论：53 · 标签：Fiction、Slavery、Language Models (LLMs)、AIEliezer Yudkowsky
33397
How AI Is Learning to Think in Secret
评论：33 · 标签：Language Models (LLMs)、AINicholas Andresen
34391
Four LLM loss functions → four flavors of LLM misalignment
评论：32 · 标签：AISteven Byrnes
35391
“Alignment Engineering” vs. “Misalignment Science”
评论：60 · 标签：AIEdward James Young
36387
Returning to ARC
评论：37 · 标签：Alignment Research Center (ARC)、AIpaulfchristiano
37380
Women should be able to open things
评论：142 · 标签：Sex & Gender、World ModelingKatjaGrace
38379
In My Misanthropy Era
评论：159 · 标签：World Modelingjenn
39378
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
40377
Paranoia: A Beginner's Guide
评论：70 · 标签：Economics、Rationalityhabryka
41377
Lightcone Commons
评论：30 · 标签：Philanthropy / Grant making (Topic)、World Optimizationhabryka
42371
A global workspace in language models
评论：68 · 标签：Interpretability (ML & AI)、Language Models (LLMs)、AIwesg
43370
There is a channel to 900M weekly users. What goes in it?
评论：33 · 标签：AICharbel-Raphaël
44370
Irretrievability; or, Murphy's Curse of Oneshotness upon ASI
评论：132 · 标签：World ModelingEliezer Yudkowsky
45365
Mnemonic portraits for 19,023 human genes
评论：29 · 标签：Memory and Mnemonics、Biology、Human Genetics、World ModelingBrinedew
46361
A critical look at Lean4
评论：40 · 标签：Formal Proof、Lean (proof assistant)、AIMilo Moses
47361
Is Mythos good at cyber because it kept hacking Anthropic's sandboxes during training?
评论：33 · 标签：AITim Hua
48361
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
49359
Cat-Belling Problems
评论：98 · 标签：Rationality、World ModelingEliezer Yudkowsky
50357
llm assistant personas seem increasingly incoherent (some subjective observations)
评论：83 · 标签：AI Psychology、AInostalgebraist

共 50 条热搜查看完整榜单

LessWrong · 历史高分

11小时前更新

011204
Why I Left Google DeepMind
评论：60 · 标签：DeepMind、Autonomous Weapons、Frontier AI Companies、Government、Mechanism Design、Something To ProtectTurnTrout
021052
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
03963
Preface
评论：20 · 标签：Postmortems & Retrospectives、Ethics & Morality、Life Improvements、Rationality A-Z (discussion & meta)、RationalityEliezer Yudkowsky
04963
Where I agree and disagree with Eliezer
评论：224 · 标签：AI Risk、AI、World Modelingpaulfchristiano
05942
Eight Short Studies On Excuses
评论：254 · 标签：Game Theory、Group Rationality、Life Improvements、Rationality、PracticalScott Alexander
06853
The Company Man
评论：79 · 标签：FictionTomás B.
07821
The Best Textbooks on Every Subject
评论：418 · 标签：Scholarship & Learning、List of Links、Book Reviews / Media Reviews、Repository、Akrasia、Algorithmslukeprog
08817
How My Students Think About AI
评论：117 · 标签：AIdvd
09800
My journey to the microwave alternate timeline
评论：58 · 标签：AI、World ModelingMalmesbury
10775
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
11771
Current AIs seem pretty misaligned to me
评论：83 · 标签：AI Risk、RLHF、Welcome Threads、AIryan_greenblatt
12751
How Go Players Disempower Themselves to AI
评论：79 · 标签：Gaming (videogames/tabletop)、MATS Program、AIAshe Vazquez Nuñez
13749
Only Law Can Prevent Extinction
评论：110 · 标签：AI、World OptimizationEliezer Yudkowsky
14727
Simulators
评论：170 · 标签：Simulator Theory、Language Models (LLMs)、LLM Personas、GPT、Outer Alignment、Simulationjanus
15709
Humans are not automatically strategic
评论：279 · 标签：Introspection、General intelligence、Goodhart's Law、Motivational Intro Posts、RationalityAnnaSalamon
16704
AI 2027: What Superintelligence Looks Like
评论：223 · 标签：AI Timelines、Forecasts (Specific Predictions)、Superintelligence、Orthogonality Thesis、Reversed Stupidity Is Not Intelligence、AIDaniel Kokotajlo
17701
What an actually pessimistic containment strategy looks like
评论：138 · 标签：Slowing Down AI、AI Governance、AI、World OptimizationDean Valentine
18696
Schelling fences on slippery slopes
评论：250 · 标签：Value Drift、Game Theory、RationalityScott Alexander
19682
SolidGoldMagikarp (plus, prompt generation)
评论：208 · 标签：Glitch Tokens、Adversarial Examples (AI)、MATS Program、Interpretability (ML & AI)、Language Models (LLMs)、AIJessica Rumbelow
20657
The Waluigi Effect (mega-post)
评论：188 · 标签：Waluigi Effect、Simulator Theory、ChatGPT、Deceptive Alignment、Language Models (LLMs)、Prompt EngineeringCleo Nardo
21655
Eliezer and I wrote a book: If Anyone Builds It, Everyone Dies
评论：144 · 标签：IABIED、AI Risk、Existential risk、AISo8res
22651
What just happened? A retrospective of AI alignment
评论：152 · 标签：Drama、Coordination / Cooperation、AIRichard_Ngo
23650
The Talker Does Not Control The Doer (in Current AIs)
评论：89 · 标签：Simulator Theory、AIEliezer Yudkowsky
24649
How to Make Superbabies
评论：363 · 标签：Biology、Human Genetics、Startups、Intelligence Amplification、Human Germline Engineering、World OptimizationGeneSmith
25648
What 2026 looks like
评论：175 · 标签：AI Timelines、Forecasting & Prediction、Forecasts (Specific Predictions)、AI Persuasion、AI Takeoff、FuzziesDaniel Kokotajlo
26623
Pain is not the unit of Effort
评论：96 · 标签：Well-being、Happiness、Heroic Responsibility、Suffering、Fallacies、Willpoweralkjash
27613
(The) Lightcone is nothing without its people: LW + Lighthaven's big fundraiser
评论：275 · 标签：Lighthaven、Communityhabryka
28611
The Terrarium
评论：54 · 标签：Fiction、AICaleb Biddulph
29605
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
30603
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
评论：69 · 标签：METR (org)、Coordination / Cooperation、OpenAI、Redwood Research、AIryan_greenblatt
31602
Making Vaccine
评论：248 · 标签：Covid-19、DIY、Market Inefficiency、Bayes' Theorem、Practicaljohnswentworth
32600
Making Beliefs Pay Rent (in Anticipated Experiences)
评论：270 · 标签：Epistemology、Anticipated Experiences、Empiricism、Principles、Psychology、RationalityEliezer Yudkowsky
33596
Orienting Toward Wizard Power
评论：149 · 标签：World Optimizationjohnswentworth
34590
LessWrong's (first) album: I Have Been A Good Bing
评论：185 · 标签：April Fool's、Many-Worlds Interpretation、AIhabryka
35580
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
36576
AI 2040: Plan A
评论：135 · 标签：AI Governance、Plan A、AIDaniel Kokotajlo
37576
If Anyone Builds It, Everyone Dies: One Year Closer
评论：66 · 标签：IABIED、AIEliezer Yudkowsky
38573
Reason as memetic immune disorder
评论：184 · 标签：Memetic Immune System、Cultural knowledge、Memetics、Valley of Bad Rationality、Compartmentalization、Pitfalls of RationalityPhilGoetz
39570
Here's to the Polypropylene Makers
评论：20 · 标签：World Optimizationjefftk
40566
Let’s think about slowing down AI
评论：182 · 标签：Slowing Down AI、Regulation and AI Risk、Coordination / Cooperation、AIKatjaGrace
41560
Welcome to LessWrong!
评论：85 · 标签：Site MetaRuby
42558
Arguments for P
评论：69 · 标签：Humor、AICleo Nardo
43557
Less Dead
评论：140 · 标签：Cryonics、Death、World OptimizationAurelia
44556
The Talk: a brief explanation of sexual dimorphism
评论：81 · 标签：Evolution、Biology、Game Theory、World ModelingMalmesbury
45555
Eliezer's Unteachable Methods of Sanity
评论：158 · 标签：Death、Rationality、PracticalEliezer Yudkowsky
46553
OpenAI Email Archives (from Musk v. Altman and OpenAI blog)
评论：82 · 标签：AIhabryka
47549
The Redaction Machine
评论：48 · 标签：Fiction、Futurism、Past and Future SelvesBen
48547
Being the (Pareto) Best in the World
评论：62 · 标签：Efficient Market Hypothesis、Pareto Efficiency、Careers、Has Diagram、Rationalityjohnswentworth
49544
A case for courage, when speaking of AI danger
评论：130 · 标签：AISo8res
50537
How much do you believe your results?
评论：19 · 标签：Goodhart's Law、Probability & Statistics、Epistemology、Has Diagram、Rationality、World ModelingEric Neyman

共 50 条热搜查看完整榜单