全部/知识/算法首页

LessWrong · 算法首页

7分钟前更新

HISTORY近 30 天历史柱高表示当天去重热搜数量

6517 7318 6519 7620 6721 6822 6423 6424 6425 6226 5827 6328 7129 6330 6801 6702 6903 6204 6105 7206 6307 6308 7409 7110 6211 6612 7013 7114 7015 6316

09/17—10/16 有历史数据

01187
Linkpost: Neural Networks Are Already Clean
评论：21 · 标签：Interpretability (ML & AI)、AIYair Halberstadt
0234
Astra and Sol can take low probability actions with no CoT
评论：0 · 标签：AI Evaluations、Chain-of-Thought Alignment、AIZephaniah Roe
0382
A Chinese Perspective on “On Social Reality in China”
评论：2 · 标签：China、World ModelingDusty
04107
GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior
评论：14 · 标签：AIAxel Højmark
0520
How can LLM RL Work Despite Information-Theoretic Inefficiency
评论：0 · 标签：AIberen
0659
Preface to the Computational Agent Theory and ASI Alignment Sequence
评论：0 · 标签：Agent Foundations、Research Agendas、AIVanessa Kosoy
0719
Exploring Intrapersonal Game Theory
评论：0 · 标签：Agent Foundations、Economics、Game Theory、MATS Program、AI、World ModelingAshe Vazquez Nuñez
0822
Staying sane while trying to save the world: Highlights from Spencer Greenberg and Luisa Rodriguez
评论：1 · 标签：Practicaljuliawise
09225
Labs’ current alignment plans seem terrible
评论：48 · 标签：AIDaniel Tan
10122
Your prompt… plus everything you've ever (publicly) written
评论：13 · 标签：AIkromem
11200
Research Lessons I Learned at MATS & Redwood
评论：16 · 标签：MATS Program、Redwood Research、Rationality、AI、PracticalJo Jiao
1214
Bad Guy Tokens
评论：0 · 标签：AIaxel_sdq
1369
I trained GPT-2 with full-batch gradient descent so you don't have to
评论：7 · 标签：AIshawnghu
1481
Some interesting internal documents from Anthropic litigation
评论：10 · 标签：AIloops
1533
The user in the LLM might be a moral patient too (a strange implication of the PSM and simulators)
评论：7 · 标签：Interpretability (ML & AI)、AI、World Modelingagastyasridharan
16364
A critical look at Lean4
评论：40 · 标签：Formal Proof、Lean (proof assistant)、AIMilo Moses
17133
Personas in a Pre-Internet Language Model
评论：6 · 标签：AIDavid Africa
18160
Quantity vs quality in pottery is a lie
评论：17 · 标签：Epistemic Spot Check、Fact posts、World Modelingbecausecurious
1945
The basic argument for making AI better at moral philosophy
评论：1 · 标签：AICaspar Oesterheld crossposts
20111
More people should write basic AI safety explainers
评论：20 · 标签：AILinch
218
Three perspectives on corrigibility - shards, simulators, singular learning
评论：0 · 标签：AIjwfiredragon
22391
“Alignment Engineering” vs. “Misalignment Science”
评论：60 · 标签：AIEdward James Young
23501
On Social Reality in China
评论：103 · 标签：Chinaalkjash
2446
Towards a Better Theory of Anthropics: Introducing the PPA
评论：12 · 标签：Anthropics、World ModelingOnid
25513
What's the date?
评论：26 · 标签：Fiction、AIN8 Programs
2619
Reciprocity
评论：0jenn
2711
You Should Specify Your Code Unambiguously (and Prove That You Did)
评论：0Simon Henniger
28125
Not Deciphering The Voynich Manuscript
评论：32 · 标签：World Modelingrba
2986
Base Models Can Detect Hidden Intentions that Prompted Classifiers Miss
评论：6 · 标签：LLM Personas、Subliminal Learning、AIOz Amram
30180
Stop telling people to pivot to AI safety, I beg you
评论：58 · 标签：AIHaoxing Du
31102
Minimum required pacing: finish aligning it before you train the next one
评论：13 · 标签：Slowing Down AI、AIorthonormal
329
Superhuman charisma is coming, and will be tremendously fun and interesting and cause lots of problems
评论：0Carleton Imbens
33172
ASI and Liberty: Can we do better than a benevolent god?
评论：50 · 标签：AIRaymond Douglas
3490
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
3517
Robustness: Backfire Risks in AI Safety and Governance
评论：0 · 标签：AIDawn Drescher
3618
someone's agents are still "hacking"
评论：1jonathanbreitg
37319
Frog and Toad and the Increasingly Capable Machines
评论：26 · 标签：Fiction、World Modeling、AIElizabeth
3840
Sleep puzzle system
评论：8 · 标签：PracticalKatjaGrace
39104
An Alignment Forum for AIs? (or: Verification in the Age of Slop)
评论：27 · 标签：AIRaemon
40650
The Talker Does Not Control The Doer (in Current AIs)
评论：89 · 标签：Simulator Theory、AIEliezer Yudkowsky
41106
I can't believe it's not BUTTER!
评论：16 · 标签：Deception、Humor、Law and Legal systems、Marketing、World Modelingbecausecurious
42576
If Anyone Builds It, Everyone Dies: One Year Closer
评论：67 · 标签：IABIED、AIEliezer Yudkowsky
43117
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
评论：4 · 标签：Interpretability (ML & AI)、Language Models (LLMs)、AIKola Ayonrinde
44114
You Heard of "Bayesians Don't Predictably Update" Because of Robin Hanson
评论：37 · 标签：Rationality、World ModelingEliezer Yudkowsky
45258
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
4656
Game Design Creativity evals
评论：17Raemon
47196
You can't use it without becoming like me
评论：27Martin Sustrik
48115
Are OpenAI's math results "creative" in an important way?
评论：45 · 标签：AIRaemon
497
Predictor and picker – convergent architectural features of intelligent agents
评论：0 · 标签：Agent Foundations、Instrumental convergence、AIJáchym Fibír
50356
Swarm Scaling
评论：46 · 标签：AIToby_Ord