id
stringlengths
36
36
source
stringclasses
15 values
formatted_source
stringclasses
13 values
text
stringlengths
2
7.55M
d7ffb7c1-997c-44b3-a511-f61d7e9eccd3
trentmkelly/LessWrong-43k
LessWrong
An idea for avoiding neuralese architectures One downside of an English chain-of-thought, is that each token contains only ≈17 bits of information, creating a tight information bottleneck. Don't take my word for it, look at this section from a story by Daniel Kokotajlo, Thomas Larsen, elifland, Scott Alexander, Jonas...
fd58e230-cfb9-48ef-b9d1-18590259937f
trentmkelly/LessWrong-43k
LessWrong
Reducing Catastrophic Risks, A Practical Introduction While thinking about my own next career steps, I've been writing down some of my thoughts about what's in an impactful career. In the process, I wrote an introductory report on what seem to me to be practical approaches to problems in catastrophic risks. It's inte...
e6accffc-3d1f-4066-9770-4564de55d15e
trentmkelly/LessWrong-43k
LessWrong
A Primer on Matrix Calculus, Part 2: Jacobians and other fun , I started this post thinking that I would write all the rules for evaluating Jacobians of neural network parameters in specific cases. But while this would certainly be useful for grokking deep learning papers, frankly it's difficult to write that in Latex...
2436917d-ecb8-46e9-a64b-21ddee543ba4
LDJnr/LessWrong-Amplify-Instruct
LessWrong
"Related to: Half-assing it with everything you've got; Wasted motion; Say it Loud. Once upon a time (true story), I was on my way to a hotel in a new city. I knew the hotel was many miles down this long, branchless road. So I drove for a long while. After a while, I began to worry I had passed the hotel. So, instead o...
f94a01fc-1125-41f9-853b-f9317ad8e345
trentmkelly/LessWrong-43k
LessWrong
In my culture: the responsibilities of open source maintainers If you maintain an open source project, what responsibilities do you have towards your users? Some recent drama (that I won't link to) reminded me that there are large differences in how people answer that question. (In the drama in question, this wasn't ...
d895ea82-2f8e-4e37-ae78-e4d3f627f6b1
trentmkelly/LessWrong-43k
LessWrong
Blatant Plot Hole in HPMoR [Spoilers] Epistemic status: Relies on the testimony of our trusted friend, Professor Quirrell A thousand of your measly tokens pale in comparison to a single Quirrell point. - Quirinus Quirrell Harry and Hermione each had, like, 200 Quirrell points each, each of which was apparently worth...
6652f7b3-8dc5-47a4-8eaf-9d6822e1176e
trentmkelly/LessWrong-43k
LessWrong
OpenAI's GPT-4 Safety Goals OpenAI has told us in some detail what they've done to make GPT-4 safe. This post will complain about some misguided aspects of OpenAI's goals. Heteronormativity and Amish Culture OpenAI wants GPT to avoid the stereotype ("bias") that says marriage is between a man and a woman (see secti...
3e6d07aa-7069-4867-8a47-54a89d993831
trentmkelly/LessWrong-43k
LessWrong
Against Occam's Razor Why should the Occamian prior work so well in the real world? It's a seemingly profound mystery that is asking to be dissolved. To begin with, I propose a Lazy Razor and a corresponding Lazy prior: > Given several competing models of reality, we should select the one that is easiest to work wit...
413217b5-2f23-4b23-9e34-3d1157b5fcfc
trentmkelly/LessWrong-43k
LessWrong
Future Of Work The current employment model is outdated. For the majority of workers, vocation and avocation are incongruent vectors. Here, we describe a set of tools that can be used to form a better-integrated work model, to dispense high-quality work to everyone. An Optimisation Problem The number of jobs is ever...
4d1067de-6feb-4431-b843-dc67c5854604
trentmkelly/LessWrong-43k
LessWrong
Training Regime Day 25: Recursive Self-Improvement Epistemic status: highly experimental Introduction I have this theory that humans have this implicit skill that's something like "doing stuff". If you're getting at "doing stuff", then you can reliably get what you want. If you're not good at "doing stuff", you oft...
2e49c41e-f0ad-43da-a909-5766687a9fad
trentmkelly/LessWrong-43k
LessWrong
The Skeptic's Trilemma Followup to: Talking Snakes: A Cautionary Tale Related to: Explain, Worship, Ignore Skepticism is like sex and pizza: when it's good, it's very very good, and when it's bad, it's still pretty good. It really is hard to dislike skeptics. Whether or not their rational justifications are perfect...
5403952b-20aa-4c0b-b189-12ce09a3b46d
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Two Challenges for ELK *This post sketches two challenges to ARC's project around eliciting latent knowledge that differ somewhat in kind from the challenges ARC is most concerned about. They relate to the difficulty in distinguishing beliefs from other representations.* Introduction ------------ The problem of EL...
20920e4f-a36a-4b4e-b502-e775f2eb19e5
trentmkelly/LessWrong-43k
LessWrong
Looking for AI Safety Experts to Provide High Level Guidance for RAISE The Road to AI Safety Excellence (RAISE) initiative aims to allow aspiring AI safety researchers and interested students to get familiar with the research landscape effectively; thereby hopefully increasing the number of researchers that contribute...
7d8d5d24-02d8-4417-be78-cbb376b43d6b
trentmkelly/LessWrong-43k
LessWrong
AI Can be “Gradient Aware” Without Doing Gradient hacking. Repetition helps us remember things better. This is because it strengthens connections in our brain used for memory[1]. We don’t need to understand the exact neural mechanisms to take advantage of this fact. For example, societies could develop cultural norms...
f67b69ac-dc79-4472-9e05-eaff76beb5c3
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Proposal: we should start referring to the risk from unaligned AI as a type of *accident risk* In the wider political sphere, a lot of people are worried about AI misuse risk. Unaligned AI is not a type of misuse. I think the clearest way to describe this is as an *accident risk*, in the same sense of the word as indu...
08eb0a6b-5845-4502-9ad2-8d091b69eabd
trentmkelly/LessWrong-43k
LessWrong
Dark Arts: Defense in Reputational Warfare First, the Dark Arts are, as the name implies, an art, not a science.  Likewise, defending against them is.  An artful attacker can utilize expected defenses against you; if you can be anticipated, you can be defeated.  The rules, therefore, are guidelines.  I'm going to stag...
a7281fa9-192a-4bea-9c07-ff18b3a603f6
trentmkelly/LessWrong-43k
LessWrong
Open Thread, Feb 8 - Feb 15, 2016 If it's worth saying, but not worth its own post (even in Discussion), then it goes here. ---------------------------------------- Notes for future OT posters: 1. Please add the 'open_thread' tag. 2. Check if there is an active Open Thread before posting a new one. (Immediately be...
ccb4adf4-3449-47ca-91b1-a21d3e151be3
trentmkelly/LessWrong-43k
LessWrong
How i'm building my ai system, how it's going so far, and my thoughts on it  In a few sentences, what i'm doing is writing a computer program that constructs a question (a prompt), which is sent to Anthropic's Claude Sonnet, then i'm processing its output into actions, and the computer program then runs those actions ...
c2e0b515-f375-40c4-bcee-48b57fa98039
LDJnr/LessWrong-Amplify-Instruct
LessWrong
"Background Information: Ingredients of Timeless Decision Theory Alternate Approaches Include: Self-empathy as a source of “willpower”, Applied Picoeconomics, Akrasia, hyperbolic discounting, and picoeconomics, Akrasia Tactics Review Standard Disclaimer: Beware of Other-Optimizing Timeless Decision Theory (or TDT) allo...
57665c77-75cd-4b00-a1b0-38d99a8471c2
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Paradigms of AI alignment: components and enablers *(Cross-posted from my* [*personal blog*](https://vkrakovna.wordpress.com/2022/06/02/paradigms-of-ai-alignment-components-and-enablers/)*. This post is based on an overview talk I gave at UCL EA and Oxford AI society (*[*recording here*](https://drive.google.com/file/...
a96f5347-5ea6-4f4d-ad4d-3f49bdb52cc1
trentmkelly/LessWrong-43k
LessWrong
Recommended reading for new rationalists This has been discussed in passing several times, but I thought it might be worthwhile to collect a list of recommended reading for new members and/or aspiring rationalists. There's probably going to be plenty of overlap with the SingInst reading list, but I think the purposes ...
5ffa4053-577c-463c-8855-5c18231dff82
LDJnr/LessWrong-Amplify-Instruct
LessWrong
"There is a widespread tendency to talk (and think) as if Einstein, Newton, and similar historical figures had superpowers—something magical, something sacred, something beyond the mundane. (Remember, there are many more ways to worship a thing than lighting candles around its altar.) Once I unthinkingly thought this w...
acfeed23-b7d2-4ced-9735-b7b26b0f2ca4
trentmkelly/LessWrong-43k
LessWrong
Why is the Future So Absurd? Followup to:  Stranger than History, Absurdity Heuristic / Absurdity Bias Why is the future more absurd than people seem to expect?  (That is:  Why, historically, has the future so often turned out to be more "absurd" than people seem to have expected?) One obvious reason is hindsight bi...
c80b8d0c-dd1e-41fe-9ae6-4393eaf06f7b
trentmkelly/LessWrong-43k
LessWrong
How dangerous is it to test a vaccine without animal trials? I saw the idea a few times (including on LW), and support it, to do/allow human trials for CoronaVirus vaccine without animal trials. My question is - how dangerous is it for a person to volunteer to test a vaccine (both with and without animal trials)? Ho...
b5bfb6e9-a7ca-4e53-84d9-866b54bbad32
StampyAI/alignment-research-dataset/arbital
Arbital
Consequentialist cognition summary: "Consequentialism" is the name for the backward step from preferring future outcomes to selecting current actions. E.g: You don't go to the airport because you really like airports; you go to the airport so that, in the future, you'll be in Oxford. (If this sounds extremely basi...
0c43b653-7fe6-4a35-98c9-5e4437e218a5
trentmkelly/LessWrong-43k
LessWrong
Meetup : Melbourne Social Meeetup: March Discussion article for the meetup : Melbourne Social Meeetup: March WHEN: 24 March 2017 06:30:00PM (+1100) WHERE: The Bull and Bear Tavern, Flinders Lane, Melbourne Come along for this month's Social Meetup, now on the exciting new schedule of the fourth Friday of the month!...
c0b2c9dc-b9b8-4a0b-b994-a6805fcd28d3
trentmkelly/LessWrong-43k
LessWrong
How should one feel morally about using chatbots? Personally, I feel some confusion. On the one hand, ones personal usage of a chatbot probably wouldn't be more than a drop in the bucket in terms of producing bad consequences such as increasing the demand for chatbots and marginally pushing AI capabilities forward. ...
a2e6993c-1394-4699-a3e9-81c4d8a9a061
trentmkelly/LessWrong-43k
LessWrong
Our philosophical thought experiments aren’t good enough and we can do better There hasn’t been nearly enough work put into creating good philosophical thought experiments, and as a result the ones that we have are either flawed or flat-out terrible. It’s easiest to explain the issues by looking at one of the terrible...
c5d94b7b-d7a0-4500-b3eb-34e3b40bfa7a
trentmkelly/LessWrong-43k
LessWrong
Why aren't there more forum-blogs like LW? LW has a unique format. It is a forum-blog. It is not a forum in the traditional sense. In traditional forums you cannot have long, essay-like posts (technically you can, but somehow the culture discourages it). Also, visually the top-level post appears separate and isn't si...
c02be1b9-2cd8-4772-8409-10b17989bf94
StampyAI/alignment-research-dataset/arxiv
Arxiv
Better Future through AI: Avoiding Pitfalls and Guiding AI Towards its Full Potential 1 Introduction --------------- After 60+ years, Artificial intelligence (AI) has moved from academic research discipline to a technology that affects people’s lives every day. We have digital assistants with which you can carry ru...
50d5fa29-674f-45d4-a134-4b31744a3d7a
StampyAI/alignment-research-dataset/arbital
Arbital
Commutative operation A commutative function $f$ is a [function](https://arbital.com/p/3jy) that takes multiple inputs from a [set](https://arbital.com/p/3jz) $X$ and produces an output that does not depend on the ordering of the inputs. For example, the binary operation $+$ is commutative, because $3 + 4 = 4 + 3.$ Th...
70699f8e-5166-4e7e-a8b8-4d97b595cf80
trentmkelly/LessWrong-43k
LessWrong
Why Obedient AI May Be the Real Catastrophe I’ve come to believe that the entire discourse around AI alignment carries a hidden desperation. A kind of reflex, a low-frequency fear, dressed up in technical language. The more I look at it, the more it seems to me that the very concept of “alignment” is thoroughly misnam...
131e355e-0af0-4282-a963-4bb3a7edaffc
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
Quantifying the Far Future Effects of Interventions *Part of a [series](http://mdickens.me/series.html#quantitative-models) on quantitative models for cause selection.* Introduction ------------ In the past I’ve [written](http://mdickens.me/2015/08/15/is_preventing_human_extinction_good/) [qualitatively](http://md...
d3fcdde8-5df9-46a0-a464-e427825444ea
trentmkelly/LessWrong-43k
LessWrong
Gut Renovating Another Bathroom After our long-term tenants bought a house and moved out, we had a few months free before we had new downstairs neighbors. A good time to gut renovate the first floor bathroom! It was a lot of work, stressful at times when I was under time pressure, but also, a lot of fun. It helped tha...
2abeccaa-4476-4edf-9f6d-f37163f61a72
trentmkelly/LessWrong-43k
LessWrong
London meetup, Sunday 1 May, 2pm, near Holborn The London LessWrong meetup takes place on the first Sunday of every other month. We're getting good at these now!  The next one is 2011-05-01 14:00, at the Shakespeares Head (official page) on Kingsway near Holborn Tube station.  Note that there's more than one pub in Lo...
604fe3a5-ca14-460f-9d00-288c0ccdcb6f
trentmkelly/LessWrong-43k
LessWrong
AI Could Defeat All Of Us Combined Click lower right to download or find on Apple Podcasts, Spotify, Stitcher, etc. I've been working on a new series of posts about the most important century. * The original series focused on why and how this could be the most important century for humanity. But it had relative...
8d352436-2224-441b-9641-059c17a5c7a2
trentmkelly/LessWrong-43k
LessWrong
Agency and the unreliable autonomous car Financial status: This is independent research, now supported by a grant. I welcome financial support. Epistemic status: I believe ~85% that the technical argument presented in this piece is correct. ---------------------------------------- Outline * This is my attempt to ...
aa9515b9-f5ea-4d65-8d73-e9024c9423e2
trentmkelly/LessWrong-43k
LessWrong
How can one measure their cognitive capacities during lucid dreaming? None
c3add3bc-4a77-473e-83f2-1a9f3921e6b9
StampyAI/alignment-research-dataset/lesswrong
LessWrong
What is “protein folding”? A brief explanation Today Google DeepMind [announced](https://deepmind.com/blog/article/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology) that their deep learning system AlphaFold has achieved unprecedented levels of accuracy on the “protein folding problem”, a grand challeng...
c4018156-d958-4654-93df-33c56abced1d
StampyAI/alignment-research-dataset/arxiv
Arxiv
A high-precision abundance analysis of the nuclear benchmark star HD 20 1 Introduction --------------- Among the cornerstones of Galactic archeology are studies of metal-poor stars as bearers of fossil records of Galactic evolution. In this respect, revealing the kinematics and chemistry of this relatively rare sub...
86002be6-d9d2-4cad-a9d0-c6ccf36a7b9d
trentmkelly/LessWrong-43k
LessWrong
Can you enter the Matrix? The deliberate simulation of sensory input. Related to: Generalizing From One Example I am able to simulate sensory input, i.e. dream deliberately, enter my personal Matrix (Holodeck). I can see, hear, feel and smell without the presence of light, sound, tactile or olfactory sensory input. ...
97fb58b6-6697-4206-8aef-e17e05567998
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
Four reasons I find AI safety emotionally compelling ‘I don’t feel emotionally motivated to work on AI safety, even though I’m intellectually convinced that it’s important.’ It always surprises me when people say this because I find my work at [Nonlinear](https://www.nonlinear.org/) on AI safety *incredibly* motivati...
1c9408d9-c9c6-47e7-91cd-36ad91635c00
StampyAI/alignment-research-dataset/blogs
Blogs
2011-2012 Winter Fundraiser Completed Thanks to our dedicated supporters, we met our goal for our 2011-2012 Winter Fundraiser. Thank you! The fundraiser ran for 56 days, from December 27, 2011 to February 20, 2012. We exceeded our $100K goal, raising a total of $143,048.84 from 101 individual donors. Every donat...
608a8ed4-33c2-45ff-b950-becaa063baea
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Incentives from a causal perspective *Post 4 of*[*Towards Causal Foundations of Safe AGI*](https://www.alignmentforum.org/s/pcdHisDEGLbxrbSHD)*, preceded by*[*Post 1: Introduction*](https://www.alignmentforum.org/posts/3oQCY4he4zRzF6vQb/introduction-to-towards-causal-foundations-of-safe-agi)*,* [*Post 2: Causality*](h...
38f1d7a7-5b63-4bd9-8179-7d9ad0f1c3c6
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
David Wolpert on Knowledge *Financial status: This is independent research, now supported by a grant. I welcome* [*financial support*](https://www.alexflint.io/donate.html)*.* *Epistemic status: I’m 85% sure that this post faithfully relates the content of the paper that it reviews.* --- David Wolpert has writ...
c5efc972-7a4d-49cd-af10-16395e0102d0
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
Donation recommendations for xrisk + ai safety Would be curious which top recommendations people have in the areas of xrisk and ai safety? Have donated to and considered: * <https://funds.effectivealtruism.org/funds/far-future> * <https://existence.org/> * <https://intelligence.org/> * <https://futureoflife.org/> * <...
c7031942-e565-497e-b882-ee1d657d7a96
StampyAI/alignment-research-dataset/youtube
Youtube Transcripts
Logical Induction: Progress in AI Alignment | Andrew Critch | EA Global 2016 so this is joint work with primarily Scott garabrant who is here Scott can you indicate all right so this man over here is like the primary source of the ideas in this paper and then the four of us have sort of like crowded around Scott as a ...
66927189-046b-44e3-b3be-f965c2a4cf24
trentmkelly/LessWrong-43k
LessWrong
Welcome to Memphis LW/SSC Meetup [Edit With Your Details] We are a small meetup that gets together to discuss rationalist and adjacent topics. We've been meeting every first Saturday of the month at Crosstown Concourse for the past several months We welcome anyone with interests across effective altruism, rationalis...
65fb758f-3634-4546-9241-2bec8c87c25d
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
AGI as a Black Swan Event Summary ======= Black swans are rare events that have a large impact and seem predictable in retrospect. Although they are rare, a single black swan can be more important than all other events combined because they are extreme. Black swan events are often unknown unknowns before they happen ...
12947405-875b-444c-9d3d-16f0f2f0cf15
trentmkelly/LessWrong-43k
LessWrong
4 Kinds of Learning The best way to learn something depends on what you're learning. Broadly-speaking, the things you learn can be divided into thinking/doing and conscious/unconscious. We can thus break the learning landscape into four kinds of learning: conceptual learning, rote learning, kinesthetic learning and di...
fd176925-11eb-4a90-986b-6e46e3d0ca87
trentmkelly/LessWrong-43k
LessWrong
GAN Discriminators Don't Generalize? Disclaimer: I just started reading about GANs, so am almost certainly missing some context here. Something that surprised me from the BigGAN paper: > We also observe that D’s loss approaches zero during training, but undergoes a sharp upward jump at collapse (Appendix F). One pos...
d55db3bd-22b2-4ad4-9b91-52af995d504b
trentmkelly/LessWrong-43k
LessWrong
AXRP Episode 43 - David Lindner on Myopic Optimization with Non-myopic Approval YouTube link In this episode, I talk with David Lindner about Myopic Optimization with Non-myopic Approval, or MONA, which attempts to address (multi-step) reward hacking by myopically optimizing actions against a human’s sense of whether...
383a76ce-6210-41f0-a087-496668cfa8b8
StampyAI/alignment-research-dataset/arxiv
Arxiv
Inference Policies INFERENCE POLICIES Associ ate Professor George Mason University 4400 University Drive Fairfax, VA 22030 Paul E. Lehner Senior Principal Associate Decision Science Consortium 1895 Preston White Drive Reston, VA 22071 Rbstract -In the AI community, there is an ongoing debate as to the most a...
48fc4831-1910-4349-b634-ee0cb9d63463
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
Eric Drexler: Paretotopian goal alignment --- *What if humanity suddenly had a thousand times as many resources at our disposal? It might make fighting over them seem like a silly idea, since cooperating would be a safer way to be reliably better off. In this talk from EA Global 2018: London, Eric Drexler argues that...
8ae74f23-d97f-4c14-9d0d-b6768f17a57f
trentmkelly/LessWrong-43k
LessWrong
History of the Development of Logical Induction I have been asked several times about how the development of logical induction happened, so I am writing it up. June 2013 - I write my first Less Wrong Post. It may not seem very related to logical uncertainty, but it was in my head. I wanted to know how to take differe...
1402bfaa-8599-4fe3-8fa7-b23cae4ee0fc
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Deceptive Agents are a Good Way to Do Things *A brief, accessible summary of the inner alignment problem.* The safety problem with powerful ML training algorithms is that **deceptive agents are a good way to do things.** Meaning, when we search over a big space of models until we find a model that performs well on so...
eeca631b-6940-4bfc-9ce8-164e1eb27733
trentmkelly/LessWrong-43k
LessWrong
$300 for the best sci-fi prompt: the results Six months ago we announced: > We would like to find the best prompt to make GPT-4 do the following: > > * write the first chapter of a science fiction novel > * the result should be good enough to make seasoned sci-fi readers (us) crave for a continuation > Why? > > ...
35680df3-80c1-472b-8ca6-b89790642921
trentmkelly/LessWrong-43k
LessWrong
Why don't more rationalists start startups? My motivation behind this post stems from Aumann's agreement theorem. It seems that my opinions on startups differ from most of the rationality community, so I want to share my thoughts, and hear your thoughts, so we could reach a better conclusion. I think that if you're s...
869d5c50-407a-4f47-9b55-c0217da3c26c
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
EA AI/Emerging Tech Orgs Should Be Involved with Patent Office Partnership Epistemic status: Highly uncertain in terms of *why* it is important, but fairly confident that it ***is*** important for guarding against unaligned AI/other emerging threats.  The U.S. Patent and Trade Office (USPTO) seeks to form a partnersh...
00233282-859d-46db-8d67-67244f7fa795
trentmkelly/LessWrong-43k
LessWrong
Occam's Razor and the Universal Prior > It can scarcely be denied that the supreme goal of all theory is to make the irreducible basic elements as simple and as few as possible without having to surrender the adequate representation of a single datum of experience. - Albert Einstein (Nature 2018). In this post, I'll ...
24ad230e-7de6-4a76-9d22-77c7323c24d3
StampyAI/alignment-research-dataset/aisafety.info
AI Safety Info
What is "superintelligence"? Nick Bostrom defines a superintelligence as “an intellect that is much smarter than the best human brains in practically every field, including scientific creativity, general wisdom and social skills.” A chess program can outperform humans in chess, but is useless at any other task. Superi...
2c0e3376-f9af-4ef6-b28c-032a18ba3dea
trentmkelly/LessWrong-43k
LessWrong
Open thread, September 4 - September 10, 2017 IF IT'S WORTH SAYING, BUT NOT WORTH ITS OWN POST, THEN IT GOES HERE. ---------------------------------------- Notes for future OT posters: 1. Please add the 'open_thread' tag. 2. Check if there is an active Open Thread before posting a new one. (Immediately before; ref...
ef227598-2e9a-40cd-98ff-4c270c78806d
trentmkelly/LessWrong-43k
LessWrong
AXRP Episode 18 - Concept Extrapolation with Stuart Armstrong YouTube link Concept extrapolation is the idea of taking concepts an AI has about the world - say, “mass” or “does this picture contain a hot dog” - and extending them sensibly to situations where things are different - like learning that the world works v...
119a5f3d-3ecd-4b8b-a63b-5b5791b8af47
trentmkelly/LessWrong-43k
LessWrong
A counterexample for measurable factor spaces PDF This is a technical post researching infinite factor spaces I define a general measurable factor space that shows that the history can't be straightforwardly generalized to the infinite case. The example shows the following: * Unlike in the finite setting, the cond...
07e2b0ce-d2aa-4728-9d6d-5787ff654f0f
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Training human models is an unsolved problem **I** We can’t write down our precise values any more than we can write down the algorithm we use for judging whether an image contains a cat. If we want an AI to abide by human values, it's going to have to acquire them without us writing them down. We usually think of th...
2c708acc-4763-48d2-b4f6-34163a33cebf
trentmkelly/LessWrong-43k
LessWrong
Downvote stalkers: Driving members away from the LessWrong community? Last month I saw this post: http://lesswrong.com/lw/kbc/meta_the_decline_of_discussion_now_with_charts/ addressing whether the discussion on LessWrong was in decline.  As a relatively new user who had only just started to post comments, my reaction ...
6fe2decd-6510-46bf-9b21-6c48b095ccbd
trentmkelly/LessWrong-43k
LessWrong
Research Report: Alternative sparsity methods for sparse autoencoders with OthelloGPT. Abstract Standard sparse autoencoder training uses an L1 sparsity loss term to induce sparsity in the hidden layer.  However, theoretical justifications for this choice are lacking (in my opinion), and there may be better ways to in...
40815d61-93f8-41ae-ab39-c410e37077ec
trentmkelly/LessWrong-43k
LessWrong
Shut Up And Guess Related to: Extreme Rationality: It's Not That Great A while back, I said provocatively that the rarefied sorts of rationality we study at Less Wrong hadn't helped me in my everyday life and probably hadn't helped you either. I got a lot of controversy but not a whole lot of good clear examples of g...
0152ea16-e5d0-4ed6-8341-a598bccf98c9
trentmkelly/LessWrong-43k
LessWrong
How large is the fallout area of the biggest cobalt bomb we can build? Cobalt bombs are the currently most effective way I know of to make large areas of land unlivable for a long period of time. I don't actually know the fallout area of such a bomb. The Wikipedia article only has the following mildly helpful section:...
ddef5a47-783b-4552-b2c3-ee069b4f8b8a
trentmkelly/LessWrong-43k
LessWrong
A chess game against GPT-4 I just bought a subscription to access GPT-4 and played the following chess game against it, with me playing white. (No particular agenda, was just curious how good it is.) At this point (move 31), GPT-4 suggested Kxc4, which is not legal, and when I asked it to correct, it suggested Kd5 ...
da9baf72-0ebd-474e-b126-1c163bcc0b85
trentmkelly/LessWrong-43k
LessWrong
What's a good way to test basic machine learning code? I originally 'scoped' this question based on my inspiration, i.e. a set of example data and their expected answers for various machine learning algorithms or their code components. In other words, a set of test cases that could be used to test implementations of t...
0e6030d1-cfd9-469a-9c26-af45237fdf36
trentmkelly/LessWrong-43k
LessWrong
LessWrong Experience of Flavours Following on from:  * http://lesswrong.com/lw/m2r/lesswrong_experience_on_alcohol/ and * http://lesswrong.com/r/discussion/lw/m3j/tally_of_lesswrong_experience_on_alcohol/ I would like to ask for other people's experience of flavours.  I am dividing food into significant categories...
773df56f-4dcd-4071-9574-4fa934d88357
awestover/filtering-for-misalignment
Redwood Research: Alek's Filtering Results
id: post599 Thanks to John Wentworth, Garrett Baker, Theo Chapman, and David Lorell for feedback and discussions on drafts of this post. In this post I’ll describe some of my thoughts on the AI control research agenda . If you haven’t read that post, I’m not going to try and summarize it here [1] , so I recommend read...
ba3167df-089f-4cf9-b1b6-98e095c1347f
trentmkelly/LessWrong-43k
LessWrong
What are some good arguments against building new nuclear power plants? It seems there is a consensus in the EA community that building new nuclear power plants is a net positive for the world. For example, this video by Kurzgesagt summarizes some arguments in support of nuclear power as a tool to mitigate the global ...
03ac8328-9b99-4412-9ee6-ccedda4e3396
trentmkelly/LessWrong-43k
LessWrong
Quantum Suicide, Decision Theory, and The Multiverse Preface: I am not suicidal or anywhere near at risk, this is not about me. Further, this is not infohazardous content. There will be discussions of death, suicide, and other sensitive topics so please use discretion, but I’m not saying anything dangerous and reading...
d742fe7c-d285-4fa3-84c4-0b3c00b3cadb
trentmkelly/LessWrong-43k
LessWrong
Value-Focused Thinking: a chapter-by-chapter summary This is a chapter-by-chapter summary of Value-Focusing Thinking by Ralph Keeney. The hope of this summary is to present most of the value of reading the book in a tiny fraction of the space. Reading the original chapters will provide additional elaboration, examples...
79b9db09-b241-4664-9dc2-967dcca73706
trentmkelly/LessWrong-43k
LessWrong
Alternative name for a Bug List At our Dojo, we performed an FMEA on all the various CFAR techniques we have gone over. The biggest failure mode of the Bug List was "discounting the positive". Additionally, if we shift our focus to a more positive mindset, it tends to be more motivating. So basically, instead of bein...
70dc6a41-7ffb-4a26-ba8d-56edbe04f662
StampyAI/alignment-research-dataset/arbital
Arbital
Goodhart's Curse summary(Gloss): Goodhart's Curse is a neologism for the combination of the Optimizer's Curse and Goodhart's Law. It states that a powerful agent neutrally optimizing a proxy measure U, meant to align with true values V, will implicitly tend to find upward divergences of U from V. In other words: po...
e01885f0-c017-4d74-803f-ccf5f816e5c3
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
[AN #93]: The Precipice we’re standing at, and how we can back away from it Alignment Newsletter is a weekly publication with recent content relevant to AI alignment around the world. Find all Alignment Newsletter **[resources here](http://rohinshah.com/alignment-newsletter/)**. In particular, you can look through **[...
b2c14f19-f4cb-4d2b-9c7e-25e7e9f2f998
trentmkelly/LessWrong-43k
LessWrong
Nash equilibria of symmetric zero-sum games This post is meant to be a reference for some basic properties of Nash equilibria in symmetric zero-sum games. I'll focus on two player games for simplicity, though the content of the post straightforwardly generalizes to the case of N>2 players. I won't assume familiarity w...
b62c0f78-38af-4a2d-ad7e-de7c5d2df3d0
trentmkelly/LessWrong-43k
LessWrong
Philosophical differences [Many people have been complaining about the lack of new content on LessWrong lately, so I thought I'd cross-post my latest blog post here in discussion. Feel free to critique the content as much as you like, but please do keep in mind that I wrote this for my personal blog and not with LW in...
6744b472-763b-479c-874a-a24e601f3aab
trentmkelly/LessWrong-43k
LessWrong
A quick update from Nonlinear One example of the evidence we’re gathering We are working hard on a point-by-point response to Ben’s article, but wanted to provide a quick example of the sort of evidence we are preparing to share: Her claim:  “Alice claims she was sick with covid in a foreign country, with only the t...
70c10c24-8006-4536-8de4-7bae0d13c521
trentmkelly/LessWrong-43k
LessWrong
Draining the swamp In an earlier post, I outlined our main weapons against infectious disease, including vaccines, antibiotics, antiseptics, pest control, sanitation, and general hygiene. These technologies (in a broad sense, even hand-washing is a technology) have largely eliminated lethal diseases such as smallpox, ...
35912976-b1fb-46a5-984f-0b181ddf96e3
StampyAI/alignment-research-dataset/arbital
Arbital
Value alignment problem Disambiguation: For the research subject that includes the entire edifice of how and why to produce good AIs, see [https://arbital.com/p/2v](https://arbital.com/p/2v). summary: The 'value alignment problem' is to produce [sufficiently advanced machine intelligences](https://arbital.com/p/7g1)...
bf6fc9e5-77f5-43a4-87e0-cf781ddcf94c
trentmkelly/LessWrong-43k
LessWrong
What Direct Instruction is A couple of days ago, prompted by several recent posts by Owen_Richardson, I checked out the book "Theory of Instruction" (Engelmann and Carnine, 1982) from my university library and promised to read it this weekend and write a post about Direct Instruction. This is that post. Learning thr...
a19781df-c895-490c-b311-ba294b86157b
trentmkelly/LessWrong-43k
LessWrong
'X is not about Y' is not about psychology As Robin Hanson says: > Food isn’t about Nutrition > Clothes aren’t about Comfort > Bedrooms aren’t about Sleep > Marriage isn’t about Romance > Talk isn’t about Info > Laughter isn’t about Jokes > Charity isn’t about Helping > Church isn’t about God > Art isn’t about Insigh...
1a69baf6-b1aa-4cb2-9c22-61f6f1154fdd
trentmkelly/LessWrong-43k
LessWrong
Making a Pedalboard A few weeks ago I posted about how I was thinking about a pedalboard. I've now made one! Before: After: It folds up slightly and fits in a standard rolly suitcase: Not pictured: I later added padding on all six sides. Here's the pedalboard by itself: You can see the bundle...
91e338b5-0e58-4762-bae3-f51a7f1aabe8
trentmkelly/LessWrong-43k
LessWrong
Re-evaluate old beliefs I've noticed that, although people can become more rational, they don't win noticeably more.  We usually re-calibrate our self-confidence, become more stubborn, and make bigger errors. Is it possible that the benefit from increasing your prediction accuracy is no greater than the loss incurred...
d2c14a40-6f25-4396-85d6-7b1ae8333e47
trentmkelly/LessWrong-43k
LessWrong
Mental Abstractions The Primacy of the Abstract is a talk given by the economist F. A. Hayek in 1969. It discusses the relationship between concrete and abstract mental phenomenon. In this post we summarize the main points made in his talk. Here are a few main points: * Mental abstractions are the cause of concrete ...
ec4a1bcf-6d45-4d59-a396-ba5713c80988
trentmkelly/LessWrong-43k
LessWrong
Crowdsourcing Anki Decks Cross-posting on EA Forum. I think it could be useful if folks had a relatively low bar for sharing Anki decks they make, especially if they're for an EA/rationalist relevant book, an 80,000 Hours podcast episode, or a good or popular textbook (especially, if it's listed here!). I would love ...
0d4467ba-e543-46d4-94a3-5f2c8211f4cb
trentmkelly/LessWrong-43k
LessWrong
Meetup : Saint Petersburg sunday meetup Discussion article for the meetup : Saint Petersburg sunday meetup WHEN: 16 February 2014 04:00:00PM (+0400) WHERE: Санкт-Петербург, м. Технологический Институт, ул.1-я Красноармейская, дом 15 We most likely will be trying out Pandemia and biased board gaming. Also I expect ...
8a51958a-d59b-41bf-8608-940fad7e7ea4
trentmkelly/LessWrong-43k
LessWrong
People neglect small probability events Over at overcomingbias Robin Hanson wrote: > On September 9, 1713, so the story goes, Nicholas Bernoulli proposed the following problem in the theory of games of chance, after 1768 known as the St Petersburg paradox …: > > Peter tosses a coin and continues to do so until it sh...
27613932-306b-47d4-a875-5a8583f7e14f
trentmkelly/LessWrong-43k
LessWrong
COVID-19: List of ideas to reduce the direct harm from the virus, with an emphasis on unusual ideas This is a collection of ideas on what interventions could be investigated, funded, or implemented to potentially reduce the harms from COVID-19 by intervening directly on the causal chain leading to infection and death....
34f497b4-4fa2-497b-859c-7b2171f36dac
trentmkelly/LessWrong-43k
LessWrong
[Linkpost] The Expressive Capacity of State Space Models: A Formal Language Perspective Paper authors: Yash Sarrof, Yana Veitsman, Michael Hahn. Context: architectures with weak forward passes can be differentially transparent; see e.g. this comment / the whole thread and research agendas like externalized reasoning ...
a6ce894e-ba2a-4010-b95d-8102cefd0078
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Alignment By Default Suppose AI continues on its current trajectory: deep learning continues to get better as we throw more data and compute at it, researchers keep trying random architectures and using whatever seems to work well in practice. Do we end up with aligned AI “by default”? I think there’s at least a plau...
cecf7f6a-9593-4f6c-a0b4-d65c516269f3
trentmkelly/LessWrong-43k
LessWrong
The Laws of Large Numbers Introduction In this short post we'll discuss fine-grained variants of the law of large numbers beyond the central limit theorem. In particular we'll introduce cumulants as a crucial (and very nice) invariant of probability distributions to track. We'll also briefly discusses parallels with p...
8e5d9ca9-5996-4176-8150-48515e5a9b3b
trentmkelly/LessWrong-43k
LessWrong
arch-anarchist reading list This is my reading list for arch-anarchists that, although they no directly support arch-anarchy, are compatible with its ideas.  1. The Making of a Small World: a similar satirical work of fiction to Nick Bostrom's The Fable of the Tyrant-Dragon, which I have already discussed in my prev...
8a574221-8c12-4e00-bbb9-c0022813c130
StampyAI/alignment-research-dataset/arxiv
Arxiv
What Matters In On-Policy Reinforcement Learning? A Large-Scale Empirical Study 1 Introduction --------------- Deep reinforcement learning (RL) has seen increased interest in recent years due to its ability to have neural-network-based agents learn to act in environments through interactions. For continuous control...
3f3458fb-10b6-4df3-b157-23b580175271
trentmkelly/LessWrong-43k
LessWrong
Implications of the inference scaling paradigm for AI safety Scaling inference With the release of OpenAI's o1 and o3 models, it seems likely that we are now contending with a new scaling paradigm: spending more compute on model inference at run-time reliably improves model performance. As shown below, o1's AIME accur...
5702c21d-2192-4b9b-893d-e17c2320f0fd
trentmkelly/LessWrong-43k
LessWrong
A Bayesian Argument for Theistic Fine-Tuning Earlier, I linked to a Bayesian argument for the resurrection of Jesus - not because I think it succeeds, obviously, but because I thought Less Wrongers might be interested to know that at least since Swinburne, some Christian apologists have taken to defending their religi...