全部/知识/编辑精选

LessWrong · 编辑精选

2小时前更新

HISTORY近 30 天历史柱高表示当天去重热搜数量

5018 5019 5020 5021 5122 5023 5024 5025 51265028 5029 5130 5031 5001

08/03—09/01 有历史数据

01525
Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident
评论：58 · 标签：Coordination / Cooperation、METR (org)、OpenAI、Redwood Research、AIryan_greenblatt
0296
On Writing #3
评论：11 · 标签：PracticalZvi
03297
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
30318
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