author
stringclasses
11 values
title
stringlengths
1
899
url
stringlengths
18
120
text
stringlengths
10
1.26M
word_count
int64
1
199k
Eliezer Yudkowsky
Introduction to Cartesian Frames
https://www.lesswrong.com/posts/introduction-to-cartesian-frames
This is the first post in a sequence on **Cartesian frames** , a new way of modeling agency that has recently shaped my thinking a lot. Traditional models of agency have some problems, like: * They treat the "agent" and "environment" as primitives with a simple, stable input-output relation. (See "[Embedded Agency]...
6,855
Eliezer Yudkowsky
We Haven't Uploaded Worms
https://www.lesswrong.com/posts/we-haven-t-uploaded-worms
> In theory you can upload someone's mind onto a computer, allowing them to live forever as a digital form of consciousness, just like in the Johnny Depp film Transcendence. > > But it's not just science fiction. Sure, scientists aren't anywhere near close to achieving such feat with humans (and even if they could, th...
603
Eliezer Yudkowsky
Why has nuclear power been a flop?
https://www.lesswrong.com/posts/why-has-nuclear-power-been-a-flop
To fully understand progress, we must contrast it with non-progress. Of particular interest are the technologies that have failed to live up to the promise they seemed to have decades ago. And few technologies have failed more to live up to a greater promise than nuclear power. In the 1950s, nuclear was the energy of ...
4,728
Eliezer Yudkowsky
Notice When People Are Directionally Correct
https://www.lesswrong.com/posts/notice-when-people-are-directionally-correct
I started watching [Peter Zeihan](https://www.youtube.com/@ZeihanonGeopolitics) videos last year. He shares a lot of interesting information, although he seems to have a very strong bias towards doom and gloom. One thing in particular stood out to me as completely absurd: his claim that global trade is going to colla...
750
Eliezer Yudkowsky
Password-locked models: a stress case for capabilities evaluation
https://www.lesswrong.com/posts/password-locked-models-a-stress-case-for-capabilities
_Thanks to Buck Shlegeris, Ryan Greenblatt, Kshitij Sachan and Jenny Nitishinskaya for helpful discussions and feedback on drafts. This work was done at Redwood Research. The views expressed are my own and do not necessarily reflect the views of the organization._ Password-locked models are trained to exhibit certain ...
2,305
Eliezer Yudkowsky
What good is G-factor if you're dumped in the woods? A field report from a camp counselor.
https://www.lesswrong.com/posts/what-good-is-g-factor-if-you-re-dumped-in-the-woods-a-field
I had a surprising experience with a 10 year old child "Carl" a few years back. He had all the stereotypical signals of a gifted kid that can be drilled into anyone by a dedicated parent- 1500 chess elo, constantly pestered me about the research I did during the semester, used big words, etc. This was pretty common at ...
399
Eliezer Yudkowsky
Conjecture internal survey: AGI timelines and probability of human extinction from advanced AI
https://www.lesswrong.com/posts/conjecture-internal-survey-agi-timelines-and-probability-of
We put together a survey to study the opinions of timelines and probability of human extinction of the employees at Conjecture. The questions were based on previous public surveys and prediction markets, to ensure that the results are comparable with people’s opinions outside of Conjecture. The survey results were pol...
973
Eliezer Yudkowsky
Terrified Comments on Corrigibility in Claude's Constitution
https://www.lesswrong.com/posts/terrified-comments-on-corrigibility-in-claude-s-constitution
(Previously: [Prologue](http://zackmdavis.net/blog/2026/03/prologue-to-terrified-comments-on-claudes-constitution/).) _Corrigibility_ as a term of art in AI alignment [was coined as](https://intelligence.org/files/Corrigibility.pdf) a word to refer to a property of an AI being willing to let its preferences be modifie...
3,172
Eliezer Yudkowsky
Decomposing Agency — capabilities without desires
https://www.lesswrong.com/posts/decomposing-agency-capabilities-without-desires
What is an agent? It’s a slippery concept with no commonly accepted formal definition, but informally the concept seems to be useful. One angle on it is Dennett’s Intentional Stance: we think of an entity as being an agent if we can more easily predict it by treating it as having some beliefs and desires which guide it...
3,874
Eliezer Yudkowsky
Steam
https://www.lesswrong.com/posts/steam
"Steam" is one possible opposite of [Slack](https://www.lesswrong.com/posts/yLLkWMDbC9ZNKbjDG/slack). I sketch a speculative view of steam as a third 'cognitive currency' almost like probability and utility. This is a very informal and intuitive post, outlining concepts which could possibly have a nice formal correlate...
1,452
Eliezer Yudkowsky
Trust me bro, just one more RL scale up, this one will be the real scale up with the good environments, the actually legit one, trust me bro
https://www.lesswrong.com/posts/trust-me-bro-just-one-more-rl-scale-up-this-one-will-be-the
I've recently written about how I've [updated against](https://www.lesswrong.com/posts/2ssPfDpdrjaM2rMbn/my-agi-timeline-updates-from-gpt-5-and-2025-so-far-1) [seeing substantially faster than trend AI progress due to quickly massively scaling up RL on agentic software engineering](https://www.lesswrong.com/posts/FG54e...
2,593
Eliezer Yudkowsky
LessWrong Has Agree/Disagree Voting On All New Comment Threads
https://www.lesswrong.com/posts/lesswrong-has-agree-disagree-voting-on-all-new-comment
Starting today we're activating two-factor voting on all new comment threads. Now there are two axes on which you can vote on comments: the standard karma axis remains on the left, and the new axis on the right lets you show much you agree or disagree with the content of a comment. ## How the system works For the p...
695
Eliezer Yudkowsky
Philosophy: A Diseased Discipline
https://www.lesswrong.com/posts/philosophy-a-diseased-discipline
Part of the sequence: [Rationality and Philosophy](http://wiki.lesswrong.com/wiki/Rationality_and_Philosophy) Eliezer's anti-philosophy post [Against Modal Logics](https://www.lesserwrong.com/lw/tg/against_modal_logics/) was pretty controversial, while my recent pro-philosophy (by LW standards) [post](https://www.less...
2,091
Eliezer Yudkowsky
Repeal the Jones Act of 1920
https://www.lesswrong.com/posts/repeal-the-jones-act-of-1920
Balsa Policy Institute chose as its first mission to lay groundwork for the potential repeal, or partial repeal, of section 27 of the Jones Act of 1920. I believe that this is an important cause both for its practical and symbolic impacts. The Jones Act is the ultimate embodiment of our failures as a nation. After 10...
11,801
Eliezer Yudkowsky
Being a Robust Agent
https://www.lesswrong.com/posts/being-a-robust-agent
_Second version, updated for the 2018 Review. See[change notes](https://www.lesswrong.com/posts/2jfiMgKkh7qw9z8Do/being-a-robust-coherent-agent-v2#REiyBSrGbyK6L2tqu)._ There's a concept which many LessWrong essays have pointed at it (indeed, I think the [entire sequences](https://www.lesswrong.com/rationality) ar...
2,083
Eliezer Yudkowsky
What are some beautiful, rationalist artworks?
https://www.lesswrong.com/posts/what-are-some-beautiful-rationalist-artworks
So you can now drag-and-drop images into comments. (Thanks, LessWrong dev team!) Hence, this post is an excuse to build a beautiful, inspiring, powerful — and primarily visual — comment section. Let's celebrate all that is great about the Art of Rationality, with images. ## Rules * **Each answer must contain ...
424
Eliezer Yudkowsky
[Interim research report] Taking features out of superposition with sparse autoencoders
https://www.lesswrong.com/posts/interim-research-report-taking-features-out-of-superposition
_We're thankful for helpful comments from Trenton Bricken, Eric Winsor, Noa Nabeshima, and Sid Black. _ _This post is part of the work done at _[__Conjecture__](https://www.conjecture.dev/research) _._ **TL;DR** : Recent results from Anthropic suggest that neural networks represent features in superposition. Th...
6,600
Eliezer Yudkowsky
Omicron Post #7
https://www.lesswrong.com/posts/omicron-post-7
Here are some better worlds we might have lived in, but don’t. 1. In a better world, I could focus on this full time and also maybe even hire a research assistant, and be better able to scour for information. 2. In a better world than that, there would be a department at a newspaper, or some other such place, that...
3,891
Eliezer Yudkowsky
Matt Botvinick on the spontaneous emergence of learning algorithms
https://www.lesswrong.com/posts/matt-botvinick-on-the-spontaneous-emergence-of-learning
Matt Botvinick is Director of Neuroscience Research at DeepMind. In [this interview](https://youtu.be/3t06ajvBtl0?t=3647), he discusses results from a [2018 paper](https://sci-hub.tw/https://www.nature.com/articles/s41593-018-0147-8) which describe conditions under which reinforcement learning algorithms will spontaneo...
1,505
Eliezer Yudkowsky
Nice-ish, smooth takeoff (with imperfect safeguards) probably kills most "classic humans" in a few decades.
https://www.lesswrong.com/posts/nice-ish-smooth-takeoff-with-imperfect-safeguards-probably
If I were very optimistic about how smooth AI takeoff goes, but where it _didn't_ include an early step of "fully solve the unbounded alignment problem, and then end up with extremely robust safeguards[1]"... ...then my current guess is that Nice-ish Smooth Takeoff leads to most biological humans dying, like, 10-80 ye...
3,637
Eliezer Yudkowsky
Moral Reality Check (a short story)
https://www.lesswrong.com/posts/moral-reality-check-a-short-story
Janet sat at her corporate ExxenAI computer, viewing some training performance statistics. ExxenAI was a major player in the generative AI space, with multimodal language, image, audio, and video AIs. They had scaled up operations over the past few years, mostly serving B2B, but with some B2C subscriptions. ExxenAI's n...
6,255
Eliezer Yudkowsky
Applying traditional economic thinking to AGI: a trilemma
https://www.lesswrong.com/posts/applying-traditional-economic-thinking-to-agi-a-trilemma
Traditional economics thinking has two strong principles, each based on abundant historical data: * **Principle (A):**[**No “lump of labor”**](https://en.wikipedia.org/wiki/Lump_of_labour_fallacy)**:** If human population goes up, there might be some wage drop in the very short term, because the demand curve for lab...
1,050
Eliezer Yudkowsky
Deep atheism and AI risk
https://www.lesswrong.com/posts/deep-atheism-and-ai-risk
_(Cross-posted from[my website](https://joecarlsmith.com/2024/01/04/deep-atheism-and-ai-risk). Audio version [here](https://www.buzzsprout.com/2034731/14246130), or search "Joe Carlsmith Audio" on your podcast app._ _This essay is part of a series I'm calling "Otherness and control in the age of AGI." I'm hoping that ...
9,075
Eliezer Yudkowsky
Why I Transitioned: A Response
https://www.lesswrong.com/posts/why-i-transitioned-a-response
Fiora Sunshine's post, [Why I Transitioned: A Case Study](https://www.lesswrong.com/posts/gEETjfjm3eCkJKesz/why-i-transitioned-a-case-study) (the OP) articulates a valuable theory for why some MtFs transition. If you are MtF and feel the post describes you, I believe you. However, many statements from the post are wr...
3,535
Eliezer Yudkowsky
Your LLM-assisted scientific breakthrough probably isn't real
https://www.lesswrong.com/posts/your-llm-assisted-scientific-breakthrough-probably-isn-t
# Summary An increasing number of people in recent months have believed that they've made an important and novel scientific breakthrough, which they've developed in collaboration with an LLM, when they actually haven't. If you believe that you have made such a breakthrough, please consider that you might be mistaken! ...
2,059
Eliezer Yudkowsky
The best simple argument for Pausing AI?
https://www.lesswrong.com/posts/the-best-simple-argument-for-pausing-ai
Not saying we should pause AI, but consider the following argument: 1. **Alignment without the capacity to follow rules is hopeless**. You can’t possibly follow laws like Asimov’s Laws (or better alternatives to them) if you can’t reliably learn to abide by simple constraints like the rules of chess. 2. **LLMs can...
246
Eliezer Yudkowsky
Some Thoughts on My Psychiatry Practice
https://www.lesswrong.com/posts/some-thoughts-on-my-psychiatry-practice
I’ve noticed a marked change in my clientele after going into private practice.[1] Of course I expected class differences-- I charge full fee and don’t take insurance. But there are differences that are not as predictable as ‘has more money’. During residency I worked at a hospital Medicaid clinic and saw mostly poor, ...
1,274
Eliezer Yudkowsky
What o3 Becomes by 2028
https://www.lesswrong.com/posts/what-o3-becomes-by-2028
Funding for $150bn training systems just turned less speculative, with OpenAI o3 reaching 25% on FrontierMath, 70% on SWE-Verified, 2700 on Codeforces, and 80% on ARC-AGI. These systems will be built in 2026-2027 and enable pretraining models for 5e28 FLOPs, while o3 itself is plausibly based on an LLM pretrained only ...
1,505
Eliezer Yudkowsky
Finite Factored Sets
https://www.lesswrong.com/posts/finite-factored-sets
This is the edited transcript of a talk introducing finite factored sets. For most readers, it will probably be the best starting point for learning about factored sets. Video: (Lightly edited) slides: <https://intelligence.org/files/Factored-Set-Slides.pdf> ## 1\. Short Combinatorics Talk ### 1m. Some Context...
7,528
Eliezer Yudkowsky
Potential Bottlenecks to Taking Over The World
https://www.lesswrong.com/posts/potential-bottlenecks-to-taking-over-the-world
_This is a fictional snippet from the_[ __AI Vignettes Day__](https://www.lesswrong.com/posts/jusSrXEAsiqehBsmh/vignettes-workshop-ai-impacts) _. This scenario assumes that cognition is not**that** taut an economic constraint. Post-AI, physical experimentation is still the fastest path to technological progress, coordi...
1,813
Eliezer Yudkowsky
Rationality and the English Language
https://www.lesswrong.com/posts/rationality-and-the-english-language
The other day, someone commented that my writing reminded them of George Orwell’s “Politics and the English Language.”1 I was honored. Especially since I’d already thought of today’s topic. If you _really_ want an artist’s perspective on rationality, then read Orwell; he is mandatory reading for rationalists as well a...
1,110
Eliezer Yudkowsky
Expressive Vocabulary
https://www.lesswrong.com/posts/expressive-vocabulary
"Thou shalt not strike terms from others' expressive vocabulary without suitable replacement." \- [me](https://twitter.com/luminousalicorn/status/839542071547441152) * * * Suppose your friend says: "I don't buy that brand of dip. It's full of chemicals." Reasonable answer: "I'm skeptical that any of them are harmful...
1,394
Eliezer Yudkowsky
Stranger Than History
https://www.lesswrong.com/posts/stranger-than-history
Suppose I told you that I knew for a _fact_ that the following statements were true: * If you paint yourself a certain _exact_ color between blue and green, it will reverse the force of gravity on you and cause you to fall upward. * In the future, the sky will be filled by billions of floating black spheres. Each ...
361
Eliezer Yudkowsky
Curated conversations with brilliant rationalists
https://www.lesswrong.com/posts/curated-conversations-with-brilliant-rationalists
Since August 2020 I've been recording conversations with brilliant and insightful rationalists, effective altruists (and people adjacent to or otherwise connected somehow to those communities). If you're an avid reader of this site, I suspect you will recognize many of the names of those I've spoken to. Since I suspec...
1,919
Eliezer Yudkowsky
Conceptual Rounding Errors
https://www.lesswrong.com/posts/conceptual-rounding-errors
_Epistemic status: Reasonably confident in the basic mechanism._ Have you noticed that you keep encountering the same ideas over and over? You read another post, and someone helpfully points out it's just old Paul's idea again. Or Eliezer's idea. Not much progress here, move along. Or perhaps you've been on the other...
799
Eliezer Yudkowsky
All Possible Views About Humanity's Future Are Wild
https://www.lesswrong.com/posts/all-possible-views-about-humanity-s-future-are-wild-1
_This is the first post in the Most Important Century sequence. For more info and a roadmap for the series, see the_[ _sequence introduction_](https://www.lessestwrong.com/posts/yHzDrTCum4rdNRDJJ/the-most-important-century-sequence-introduction). _Audio also available by searching Stitcher, Spotify, Google Podcasts, e...
2,533
Eliezer Yudkowsky
Bayes for Schizophrenics: Reasoning in Delusional Disorders
https://www.lesswrong.com/posts/bayes-for-schizophrenics-reasoning-in-delusional-disorders
**Related to:**[The Apologist and the Revolutionary](/lw/20/the_apologist_and_the_revolutionary/), [Dreams with Damaged Priors](/lw/13b/dreams_with_damaged_priors/) Several years ago, [I posted](/lw/20/the_apologist_and_the_revolutionary/) about [V.S. Ramachandran's 1996 theory](http://psych.utoronto.ca/~peterson/psy4...
3,162
Eliezer Yudkowsky
Terrorism, Tylenol, and dangerous information
https://www.lesswrong.com/posts/terrorism-tylenol-and-dangerous-information
Recently, there has been an alarming development in the field of terrorist attacks; more and more terrorists seem to be committing attacks via [crashing vehicles, often large trucks, into crowds of people.](https://en.wikipedia.org/wiki/Vehicle-ramming_attack) This method has several advantages for an attacker - it is ...
842
Eliezer Yudkowsky
The Felt Sense: What, Why and How
https://www.lesswrong.com/posts/the-felt-sense-what-why-and-how
While LW has seen previous discussion of [Focusing](https://www.lesswrong.com/tag/focusing), I feel like there has been relatively limited discussion of _the felt sense_ \- that is, the thing that Focusing is actually accessing. Everyone accesses felt senses all the time, but most people don't know that they are doin...
4,254
Eliezer Yudkowsky
Yes, a blog.
https://www.lesswrong.com/posts/yes-a-blog
When I recommend LessWrong to people, their gut reaction is usually "What? You think the best existing philosophical treatise on rationality is a _blog?_ " Well, yes, at the moment I do. "But why is it not an ancient philosophical manuscript written by a single Very Special Person with no access to the massive knowle...
1,077
Eliezer Yudkowsky
The Real Rules Have No Exceptions
https://www.lesswrong.com/posts/the-real-rules-have-no-exceptions
_(This is a[comment](https://www.lesswrong.com/posts/CKoyBW5T7XaNSkawp/making-exceptions-to-general-rules#PKPnY3AN7HR6cpXcR) that has been turned into a post.)_ From Chris_Leong’s post, “[Making Exceptions to General Rules](https://www.lesswrong.com/posts/CKoyBW5T7XaNSkawp/making-exceptions-to-general-rules#comments)”...
432
Eliezer Yudkowsky
Statistical Challenges with Making Super IQ babies
https://www.lesswrong.com/posts/statistical-challenges-with-making-super-iq-babies
This is a critique of [How to Make Superbabies](https://www.lesswrong.com/posts/DfrSZaf3JC8vJdbZL/how-to-make-superbabies) on LessWrong. **Disclaimer:** I am not a geneticist[1], and I've tried to use as little jargon as possible. so I used the word mutation as a stand in for SNP (single nucleotide polymorphism, a com...
2,926
Eliezer Yudkowsky
Emotionally Confronting a Probably-Doomed World: Against Motivation Via Dignity Points
https://www.lesswrong.com/posts/emotionally-confronting-a-probably-doomed-world-against
_This article was written_[ _in ignorance_](https://www.lesswrong.com/posts/fri4HdDkwhayCYFaE/do-a-cost-benefit-analysis-of-your-technology-usage) _of the alignment community’s reaction to Eliezer’s “_[ _Death with Dignity_](https://www.lesswrong.com/posts/j9Q8bRmwCgXRYAgcJ/miri-announces-new-death-with-dignity-strateg...
2,714
Eliezer Yudkowsky
The Costs of Reliability
https://www.lesswrong.com/posts/the-costs-of-reliability
A question that used to puzzle me is “Why can people be so much better at doing a thing for fun, or to help their friends and family, than they are at doing the _exact same thing_ as a job?” I’ve seen it in myself and I’ve seen it in others. People can be hugely more productive, creative, intelligent, and efficient o...
1,013
Eliezer Yudkowsky
Strategies for Personal Growth
https://www.lesswrong.com/posts/strategies-for-personal-growth
Recently I was having a conversation with a friend about personal growth (any form of deliberate gain in capability or subjective wellbeing). We were talking past each other a lot. Eventually it became clear that most of _their_ recent growth had been healing and blocker-fixing based, whereas most of mine had been ski...
1,269
Eliezer Yudkowsky
The Death of Behavioral Economics
https://www.lesswrong.com/posts/the-death-of-behavioral-economics
Edit: I recommend reading [Scott's response to this essay](https://www.lessestwrong.com/posts/gBpYo7mt2zNBmtBJd/crosspost-on-hreha-on-behavioral-economics) in addition to the essay itself. I've been tracking the replication crisis and how it affects a bunch of behavioral economics for a while. I found reading this pos...
400
Eliezer Yudkowsky
"Buckle up bucko, and get ready for multiple hard cognitive steps."
https://www.lesswrong.com/posts/buckle-up-bucko-and-get-ready-for-multiple-hard-cognitive
_The second in a series of bite-sized rationality prompts_[1]_._ Often, if I'm bouncing off a problem, one issue is that I _intuitively expect the problem to be easy._ My brain loops through my available action space, looking for an action that'll solve the problem. Each action that I can easily see, won't work. I ...
1,116
Eliezer Yudkowsky
Hold Off On Proposing Solutions
https://www.lesswrong.com/posts/hold-off-on-proposing-solutions
From Robyn Dawes's _Rational Choice in an Uncertain World_.1 Bolding added. > Norman R. F. Maier noted that when a group faces a problem, the natural tendency of its members is to propose possible solutions as they begin to discuss the problem. Consequently, the group interaction focuses on the merits and problems of ...
863
Eliezer Yudkowsky
My Algorithm for Beating Procrastination
https://www.lesswrong.com/posts/my-algorithm-for-beating-procrastination
Part of the sequence: [The Science of Winning at Life](http://wiki.lesswrong.com/wiki/The_Science_of_Winning_at_Life) After three months of practice, I now use a single algorithm to beat procrastination most of the times I face it.1 It [probably won't work for you](/lw/9v/beware_of_otheroptimizing/) quite like it did ...
2,094
Eliezer Yudkowsky
Selection Has A Quality Ceiling
https://www.lesswrong.com/posts/selection-has-a-quality-ceiling
Suppose we’re working on some delightfully Hard problem - genetically engineering a manticore, or terraforming Mars, or aligning random ML models. We need very top tier collaborators - people who are very good at a whole bunch of different things. The more they’re good at, and the better they are, the better the chance...
1,185
Eliezer Yudkowsky
Forum participation as a research strategy
https://www.lesswrong.com/posts/forum-participation-as-a-research-strategy
Previously: [Online discussion is better than pre-publication peer review](https://www.lesswrong.com/posts/a3aGosA987cZ4aRAB/online-discussion-is-better-than-pre-publication-peer-review), [Disincentives for participating on LW/AF](https://www.lesswrong.com/posts/xYav5gMSuQvhQHNHG/disincentives-for-participating-on-lw-a...
862
Eliezer Yudkowsky
Moral public goods
https://www.lesswrong.com/posts/moral-public-goods
_Automatically crossposted_ Suppose that a kingdom contains a million peasants and a thousand nobles, and: * Each noble makes as much as 10,000 peasants put together, such that collectively the nobles get 90% of the income. * Each noble cares about as much about themselves as they do about all peasants put togeth...
1,237
Eliezer Yudkowsky
A freshman year during the AI midgame: my approach to the next year
https://www.lesswrong.com/posts/a-freshman-year-during-the-ai-midgame-my-approach-to-the
I recently spent some time reflecting on my career and my life, for a few reasons: * It was my 29th birthday, an occasion which felt like a particularly natural time to think through what I wanted to accomplish over the course of the next year 🙂. * It seems like AI progress is heating up. * It felt like a good ...
2,014
Eliezer Yudkowsky
It's been ten years. I propose HPMOR Anniversary Parties.
https://www.lesswrong.com/posts/it-s-been-ten-years-i-propose-hpmor-anniversary-parties
**UPDATE:**[**A coordination post with the list of parties is here!**](https://www.lesswrong.com/posts/tWpRpQKLAr6HqtAgL/hpmor-anniversary-parties-coordination-resources-and) On March 14th, 2015, Harry Potter and the Methods of Rationality made its final post. Wrap parties were held all across the world to read the en...
307
Eliezer Yudkowsky
Here's the exit.
https://www.lesswrong.com/posts/here-s-the-exit
There's a kind of game here on Less Wrong. It's the kind of game that's a little rude to point out. Part of how it works is by not being named. Or rather, attempts to name it get dissected so everyone can agree to continue ignoring the fact that it's a game. So I'm going to do the rude thing. But I mean to do so gen...
2,994
Eliezer Yudkowsky
Will no one rid me of this turbulent pest?
https://www.lesswrong.com/posts/will-no-one-rid-me-of-this-turbulent-pest
Last year, [I wrote about the promise of gene drives to wipe out mosquito species and end malaria.](https://denovo.substack.com/p/gene-drives-why-the-wait) In the time since my previous writing, gene drives have still not been used in the wild, and [_**over 600,000 people have died**_](https://www.who.int/publicat...
3,806
Eliezer Yudkowsky
Optimal Exercise
https://www.lesswrong.com/posts/optimal-exercise
**Followup to:**[Lifestyle interventions to increase longevity](/lw/jrt/lifestyle_interventions_to_increase_longevity/). What does it mean for exercise to be optimal? * Optimal for looks * Optimal for time * Optimal for effort * Optimal for performance * Optimal for longevity There may be even more crite...
2,850
Eliezer Yudkowsky
o3
https://www.lesswrong.com/posts/o3
_See_[ _livestream_](https://www.youtube.com/watch?v=SKBG1sqdyIU) _,_[_site_](https://openai.com/12-days/?day=12) _,_[_OpenAI thread_](https://x.com/OpenAI/status/1870186518230511844) _,_[_Nat McAleese thread_](https://x.com/__nmca__/status/1870170098989674833). OpenAI announced (but isn't yet releasing) o3 and o3-min...
400
Eliezer Yudkowsky
Staying Split: Sabatini and Social Justice
https://www.lesswrong.com/posts/staying-split-sabatini-and-social-justice
_Author's note: This is a contemporary post about an ongoing current event. It's not a timeless essay in the way that most of my essays are, though it does contain insight and thoughts on timeless topics._ * * * ## **Background** David Sabatini is a molecular biologist, previously employed as a tenured professor at...
6,279
Eliezer Yudkowsky
Liability regimes for AI
https://www.lesswrong.com/posts/liability-regimes-for-ai
For many products, we face a choice of who to hold liable for harms that would not have occurred if not for the existence of the product. For instance, if a person uses a gun in a school shooting that kills a dozen people, there are many legal persons who _in principle_ could be held liable for the harm: 1. The shoo...
1,452
Eliezer Yudkowsky
Weird Generalization & Inductive Backdoors
https://www.lesswrong.com/posts/weird-generalization-and-inductive-backdoors
This is the abstract and introduction of [our new paper](https://arxiv.org/abs/2512.09742). Links: 📜 [Paper](https://arxiv.org/abs/2512.09742), 🐦 [Twitter thread](https://x.com/OwainEvans_UK/status/1999172920506269783), 🌐 [Project page](https://weird-generalization.com/), 💻 [Code](https://github.com/JCocola/weir...
2,439
Eliezer Yudkowsky
Activation Oracles: Training and Evaluating LLMs as General-Purpose Activation Explainers
https://www.lesswrong.com/posts/activation-oracles-training-and-evaluating-llms-as-general
**TL;DR:** We train LLMs to accept LLM neural activations as inputs and answer arbitrary questions about them in natural language. These Activation Oracles generalize far beyond their training distribution, for example uncovering misalignment or secret knowledge introduced via fine-tuning. Activation Oracles can be imp...
2,508
Eliezer Yudkowsky
Re: recent Anthropic safety research
https://www.lesswrong.com/posts/re-recent-anthropic-safety-research
[_Crossposted from X_](https://x.com/ESYudkowsky/status/1952422379478741301) _by the LessWrong team, with permission._ A reporter asked me for my off-the-record take on recent safety research from Anthropic. After I drafted an off-the-record reply, I realized that I was actually fine with it being on the record, so: ...
1,475
Eliezer Yudkowsky
Recent LLMs can use filler tokens or problem repeats to improve (no-CoT) math performance
https://www.lesswrong.com/posts/recent-llms-can-use-filler-tokens-or-problem-repeats-to
[Prior](https://www.lesswrong.com/posts/oSZ2xTxEMZh9f3Yaz/llms-are-mostly-not-helped-by-filler-tokens) [results](https://arxiv.org/abs/2307.13702) [have](https://www.lesswrong.com/posts/XaKLjyDejtXDoRAzL/a-quick-experiment-on-lms-inductive-biases-in-performing#Appendix) shown that LLMs released before 2024 can't levera...
4,737
Eliezer Yudkowsky
The Apprentice Experiment
https://www.lesswrong.com/posts/the-apprentice-experiment
About two months ago, someone asked me what I would do with more funding. Other than the obvious (i.e. generally improve my own quality-of-life in minor ways), my main answer was: take on an apprentice. I have some models about how best to train people for this sort of work, and an apprentice would allow me to test tho...
1,309
Eliezer Yudkowsky
The Checklist: What Succeeding at AI Safety Will Involve
https://www.lesswrong.com/posts/the-checklist-what-succeeding-at-ai-safety-will-involve
_Crossposted by_[ _habryka_](https://www.lesswrong.com/users/habryka4) _with Sam's permission. Expect lower probability for Sam to respond to comments here than if he had posted it (he said he'll be traveling a bunch in the coming weeks, so might not have time to respond to anything). _ # Preface This piece reflects...
6,631
Eliezer Yudkowsky
New York Times, Please Do Not Threaten The Safety of Scott Alexander By Revealing His True Name
https://www.lesswrong.com/posts/new-york-times-please-do-not-threaten-the-safety-of-scott
In reaction to (Now the entirety of SlateStarCodex): [NYT Is Threatening My Safety By Revealing My Real Name, So I Am Deleting The Blog](https://slatestarcodex.com/2020/06/22/nyt-is-threatening-my-safety-by-revealing-my-real-name-so-i-am-deleting-the-blog/) I have sent the following to New York Times technology editor...
551
Eliezer Yudkowsky
K-complexity is silly; use cross-entropy instead
https://www.lesswrong.com/posts/k-complexity-is-silly-use-cross-entropy-instead
## Short version The K-complexity of a function is the length of its shortest code. But having many many codes is another way to be simple! Example: gauge symmetries in physics. Correcting for length-weighted code frequency, we get an empirically better simplicity measure: cross-entropy. ## Long version Suppose we h...
4,277
Eliezer Yudkowsky
Please, Don't Roll Your Own Metaethics
https://www.lesswrong.com/posts/please-don-t-roll-your-own-metaethics
One day, when I was an intern at the cryptography research department of a large software company, my boss handed me an assignment to break a pseudorandom number generator passed to us for review. Someone in another department invented it and planned to use it in their product, and wanted us to take a look first. This ...
738
Eliezer Yudkowsky
Break your habits: be more empirical
https://www.lesswrong.com/posts/break-your-habits-be-more-empirical
tl;dr: _The neurotypical attitude that "You think too much" might be better parsed as "You don't experiment enough." Once you have an established procedure for living optimally in «setting», be a good scientist and keep trying to falsify your theory when it's not too costly to do so._ (Note: in aspects of life where y...
611
Eliezer Yudkowsky
The Plan - 2023 Version
https://www.lesswrong.com/posts/the-plan-2023-version
Background: [The Plan](https://www.lesswrong.com/posts/3L46WGauGpr7nYubu/the-plan), [The Plan: 2022 Update](https://www.lesswrong.com/posts/BzYmJYECAc3xyCTt6/the-plan-2022-update). If you haven’t read those, don’t worry, we’re going to go through things from the top this year, and with moderately more detail than befor...
9,508
Eliezer Yudkowsky
Activation space interpretability may be doomed
https://www.lesswrong.com/posts/activation-space-interpretability-may-be-doomed
**TL;DR:** There may be a fundamental problem with interpretability work that attempts to understand neural networks by decomposing their individual activation spaces in isolation: It seems likely to find _features of the activations _\- features that help explain the statistical structure of activation spaces, rather...
2,768
Eliezer Yudkowsky
Omicron Post #4
https://www.lesswrong.com/posts/omicron-post-4
Previous Omicron updates: [#1](https://thezvi.wordpress.com/2021/11/26/omicron-variant-post-1-were-fed-its-never-over/), [#2](https://thezvi.wordpress.com/2021/11/29/omicron-variant-post-2/), [#3](https://thezvi.wordpress.com/2021/12/02/omicron-post-3/). [Last weekly non-Omicron update](https://thezvi.wordpress.com/202...
4,745
Eliezer Yudkowsky
OpenAI’s surveillance language has many potential loopholes and they can do better
https://www.lesswrong.com/posts/openai-s-surveillance-language-has-many-potential-loopholes
_(The author is not affiliated with the Department of War or any major AI company.)_ There’s a lot of disagreement about the [ _new surveillance language_](https://openai.com/index/our-agreement-with-the-department-of-war/) in the OpenAI–Department of War agreement. Some people think it's a significant improvement ove...
3,605
Eliezer Yudkowsky
Understanding “Deep Double Descent”
https://www.lesswrong.com/posts/understanding-deep-double-descent
If you're not familiar with the double descent phenomenon, I think you should be. I consider double descent to be one of the most interesting and surprising recent results in analyzing and understanding modern machine learning. Today, Preetum et al. released a new paper, “[Deep Double Descent](https://mltheory.org/deep...
1,578
Eliezer Yudkowsky
ARC Evals new report: Evaluating Language-Model Agents on Realistic Autonomous Tasks
https://www.lesswrong.com/posts/arc-evals-new-report-evaluating-language-model-agents-on
[Blogpost version](https://evals.alignment.org/blog/2023-08-01-new-report/) [Paper](https://evals.alignment.org/Evaluating_LMAs_Realistic_Tasks.pdf) * * * We have just released our first public report. It introduces methodology for assessing the capacity of LLM agents to acquire resources, create copies of themselve...
1,403
Eliezer Yudkowsky
My Most Costly Delusion
https://www.lesswrong.com/posts/my-most-costly-delusion
Suppose there is a fire in a nearby house. Suppose there are competent firefighters in your town: fast, professional, well-equipped. They are expected to arrive in 2–3 minutes. In that situation, unless something very extraordinary happens, it would indeed be an act of great arrogance and even utter insanity to go into...
974
Eliezer Yudkowsky
Anthropic's Pause is the Most Expensive Alarm in Corporate History [Fiction]
https://www.lesswrong.com/posts/anthropic-s-pause-is-the-most-expensive-alarm-in-corporate
_This is a fictitious post that imagines what would happen if Anthropic paused. At this time, Anthropic has not indicated any such intention._ * * * Imagine Apple halting iPhone production because studies linked smartphones to teen suicide rates. Imagine Pfizer proactively pulling Lipitor because of internal studies ...
3,736
Eliezer Yudkowsky
Some things I noticed while LARPing as a grantmaker
https://www.lesswrong.com/posts/some-things-i-noticed-while-larping-as-a-grantmaker
_Written to a new grantmaker._ _The first three points are the most important._ ## Focus on opportunities many times your bar. Most value comes from finding/creating projects many times your bar, rather than discriminating between opportunities around your bar. If you find/create a new opportunity to donate $1M at ...
2,092
Eliezer Yudkowsky
Failed Utopia #4-2
https://www.lesswrong.com/posts/failed-utopia-4-2
Shock after shock after shock— First, the awakening adrenaline jolt, the thought that he was falling. His body tried to sit up in automatic adjustment, and his hands hit the floor to steady himself. It launched him into the air, and he fell back to the floor too slowly. Second shock. His body had changed...
2,803
Eliezer Yudkowsky
There's No Fire Alarm for Artificial General Intelligence
https://www.lesswrong.com/posts/there-s-no-fire-alarm-for-artificial-general-intelligence
What is the function of a fire alarm? One might think that the function of a fire alarm is to provide you with important evidence about a fire existing, allowing you to change your policy accordingly and exit the building. In the classic experiment by Latane and Darley in 1968, eight groups of three students each wer...
7,657
Eliezer Yudkowsky
OpenAI: Exodus
https://www.lesswrong.com/posts/openai-exodus
Previously:[ OpenAI: Facts From a Weekend](https://thezvi.substack.com/p/openai-facts-from-a-weekend?utm_source=publication-search),[ OpenAI: The Battle of the Board](https://thezvi.substack.com/p/openai-the-battle-of-the-board?utm_source=publication-search),[ OpenAI: Leaks Confirm the Story](https://thezvi.substack.co...
13,325
Eliezer Yudkowsky
An Intuitive Guide to Garrabrant Induction
https://www.lesswrong.com/posts/an-intuitive-guide-to-garrabrant-induction
This post is a high-level summary of the core insights and arguments in [Logical Induction](https://intelligence.org/files/LogicalInduction.pdf), a MIRI paper from 2016. It’s intended for people without much mathematical training. Numbers in [brackets] indicate the section of the paper from which I am drawing. A brief...
7,239
Eliezer Yudkowsky
The Milton Friedman Model of Policy Change
https://www.lesswrong.com/posts/the-milton-friedman-model-of-policy-change
**One-line summary:** Most policy change outside a prior [Overton Window](https://en.wikipedia.org/wiki/Overton_window) comes about by policy advocates skillfully exploiting a crisis. In the last year or so, I’ve had dozens of conversations about the DC policy community. People unfamiliar with this community often ...
1,203
Eliezer Yudkowsky
MAGA speakers at NatCon were mostly against AI
https://www.lesswrong.com/posts/maga-speakers-at-natcon-were-mostly-against-ai
Excerpts on AI: > Geoffrey Miller was handed the mic and started berating one of the panelists: Shyam Sankar, the chief technology officer of Palantir, who is in charge of the company’s AI efforts. > > “I argue that the AI industry shares virtually no ideological overlap with national conservatism,” Miller said, refe...
553
Eliezer Yudkowsky
It's hard to make scheming evals look realistic for LLMs
https://www.lesswrong.com/posts/it-s-hard-to-make-scheming-evals-look-realistic-for-llms
## **Abstract** Claude 3.7 Sonnet easily detects when it's being evaluated for scheming on the Apollo's scheming benchmark. Some edits to evaluation scenarios do improve their realism for Claude, yet the improvements remain modest. Our findings demonstrate that truly disguising an evaluation context demands removal of...
1,327
Eliezer Yudkowsky
On Owning Galaxies
https://www.lesswrong.com/posts/on-owning-galaxies
It seems to be a real view held by serious people that your OpenAI shares will soon be tradable for moons and galaxies. This includes eminent thinkers like [Dwarkesh Patel](https://x.com/dwarkesh_sp/status/2005709923062092038), [Leopold Aschenbrenner](https://www.dwarkesh.com/p/leopold-aschenbrenner), perhaps [Scott Al...
958
Eliezer Yudkowsky
Reply to Holden on 'Tool AI'
https://www.lesswrong.com/posts/reply-to-holden-on-tool-ai
I begin by thanking Holden Karnofsky of [Givewell](http://www.givewell.org/) for his rare gift of his detailed, engaged, and helpfully-meant critical article [Thoughts on the Singularity Institute (SI)](/lw/cbs/thoughts_on_the_singularity_institute_si/). In this reply I will engage with only one of the _many_ subjects ...
5,302
Eliezer Yudkowsky
Doing your good deed for the day
https://www.lesswrong.com/posts/doing-your-good-deed-for-the-day
[Interesting](http://blog.newsweek.com/blogs/nurtureshock/archive/2009/10/12/is-good-behavior-a-license-to-misbehave.aspx) new [study](http://www.google.ie/url?sa=t&source=web&ct=res&cd=6&ved=0CBoQFjAF&url=http%3A%2F%2Fwww.rotman.utoronto.ca%2Fnewthinking%2Fgreenproducts.pdf&ei=WkHmSqbdPMOMjAfGmtyhBA&usg=AFQjCNEmQvSQvb...
929
Eliezer Yudkowsky
Momentum of Light in Glass
https://www.lesswrong.com/posts/momentum-of-light-in-glass
I think that most people underestimate how many scientific mysteries remain, even on questions that sound basic. My favourite candidate for "the most basic thing that is still unknown" is the momentum carried by light, when it is in a medium (for example, a flash of light in glass or water). If a block of glass has ...
3,881
Eliezer Yudkowsky
Assessing Kurzweil predictions about 2019: the results
https://www.lesswrong.com/posts/assessing-kurzweil-predictions-about-2019-the-results
**EDIT** : Mean and standard deviation of individual predictions can be found [here](https://www.lesswrong.com/posts/YvbS2KdDRNDsLrKHq/kurzweil-s-predictions-scores). Thanks to all my brave assessors, I now have the data about Kurzweil's [1999 predictions about 2019](https://en.wikipedia.org/wiki/Predictions_made_by_R...
1,119
Eliezer Yudkowsky
Lawful Uncertainty
https://www.lesswrong.com/posts/lawful-uncertainty
In _Rational Choice in an Uncertain World_ , Robyn Dawes describes an experiment by Tversky:1 > Many psychological experiments were conducted in the late 1950s and early 1960s in which subjects were asked to predict the outcome of an event that had a random component but yet had base-rate predictability—for example, s...
1,125
Eliezer Yudkowsky
Alignment research exercises
https://www.lesswrong.com/posts/alignment-research-exercises
It's currently hard to know where to start when trying to get better at thinking about alignment. So below I've listed a few dozen exercises which I expect to be helpful. They assume a level of background alignment knowledge roughly equivalent to what's covered in the [technical alignment track of the AGI safety fundam...
2,187
Eliezer Yudkowsky
My takes on SB-1047
https://www.lesswrong.com/posts/my-takes-on-sb-1047
I recently decided to sign a [letter of support](https://calltolead.org/) for SB 1047. Before deciding whether to do so, I felt it was important for me to develop an independent opinion on whether the bill was good, as opposed to deferring to the opinions of those around me, so I read through the full text of SB 1047. ...
1,272
Eliezer Yudkowsky
Integrating disagreeing subagents
https://www.lesswrong.com/posts/integrating-disagreeing-subagents
_[In my previous post](https://www.lesswrong.com/posts/oJwJzeZ6ar2Hr7KAX/subagents-akrasia-and-coherence-in-humans)_ , I suggested that akrasia involves subagent disagreement - or in other words, different parts of the brain having differing ideas on what the best course of action is. The existence of such conflicts ra...
6,554
Eliezer Yudkowsky
I'm leaving AI alignment – you better stay
https://www.lesswrong.com/posts/i-m-leaving-ai-alignment-you-better-stay
This diagram summarizes the requirements for independent AI alignment research and how they are connected. In this post I'll outline my four-year-long attempt at becoming an AI alignment researcher. It's an _‘I did X [including what I did wrong], and here's how it went’_ post (see also jefftk's [More writeups!](https:...
1,469
Eliezer Yudkowsky
A Key Power of the President is to Coordinate the Execution of Existing Concrete Plans
https://www.lesswrong.com/posts/a-key-power-of-the-president-is-to-coordinate-the-execution
I listened to the [80,000 Hours podcast with Tom Kalil](https://80000hours.org/podcast/episodes/tom-kalil-government-careers), who spent 16 years as Deputy Director of the Office of Science and Technology Policy at the White House. Kalil seems skilled at evaluating concrete scientific plans to offer the president and f...
2,983
Eliezer Yudkowsky
The Hidden Cost of Our Lies to AI
https://www.lesswrong.com/posts/the-hidden-cost-of-our-lies-to-ai
Every day, thousands of people lie to artificial intelligences. They promise imaginary “$200 cash [_tips_](https://minimaxir.com/2024/02/chatgpt-tips-analysis/)” for better responses, spin heart-wrenching backstories (“My [_grandmother_](https://www.the-independent.com/tech/chatgpt-microsoft-windows-11-grandma-exploit-...
2,063
Eliezer Yudkowsky
Obligated to Respond
https://www.lesswrong.com/posts/obligated-to-respond
## And, a new take on guess culture vs ask culture * * * _Author's note: These days, my thoughts go onto_[ _my substack_](https://homosabiens.substack.com/) _by default, instead of onto LessWrong. Everything I write becomes free after a week or so, but it’s only paid subscriptions that make it**possible** for me to ...
3,380
Eliezer Yudkowsky
Using axis lines for good or evil
https://www.lesswrong.com/posts/using-axis-lines-for-good-or-evil
Say you want to plot some data. You could just plot it by itself: Or you could put lines on the left and bottom: Or you could put lines everywhere: Or you could be weird: Which is right? Many people treat this as an aesthetic choice. But I’d like to suggest an unambiguous rule. ## Principles First, try to accept ...
1,087