全部/知识/编辑精选

LessWrong · 编辑精选

5小时前更新

HISTORY近 30 天历史柱高表示当天去重热搜数量

5117 5018 5019 5020 5021 5122 5023 5024 5125 5026 5027 5028 5029 5030 5101 5002 5003 5004 5105 5006 5007 5108 5009 5010 5011 5112 5013 5014 5115 5016

09/17—10/16 有历史数据

01366
Cat-Belling Problems
评论：98 · 标签：Rationality、World ModelingEliezer Yudkowsky
02362
A critical look at Lean4
评论：40 · 标签：Formal Proof、Lean (proof assistant)、AIMilo Moses
03513
What's the date?
评论：26 · 标签：Fiction、AIN8 Programs
04501
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
11817
How My Students Think About AI
评论：117 · 标签：AIdvd
12603
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