全部/知识/本年高分

LessWrong · 本年高分

2分钟前更新

HISTORY近 30 天历史柱高表示当天去重热搜数量

5017 5218 5019 5120 5021 5022 5123 5024 5025 5126 5027 5028 5129 5030 5001 5002 5103 5104 5105 5006 5007 5008 5109 5110 5011 5012 5013 5014 5215 5016

09/17—10/16 有历史数据

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
06750
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：67 · 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
17513
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
19501
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
45368
Cat-Belling Problems
评论：100 · 标签：Rationality、World ModelingEliezer Yudkowsky
46365
Mnemonic portraits for 19,023 human genes
评论：29 · 标签：Memory and Mnemonics、Biology、Human Genetics、World ModelingBrinedew
47364
A critical look at Lean4
评论：40 · 标签：Formal Proof、Lean (proof assistant)、AIMilo Moses
48361
Is Mythos good at cyber because it kept hacking Anthropic's sandboxes during training?
评论：33 · 标签：AITim Hua
49361
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
50357
llm assistant personas seem increasingly incoherent (some subjective observations)
评论：83 · 标签：AI Psychology、AInostalgebraist