id
stringlengths
36
36
source
stringclasses
15 values
formatted_source
stringclasses
13 values
text
stringlengths
2
7.55M
4df48ae8-9751-43a3-9bff-3ff74683951e
trentmkelly/LessWrong-43k
LessWrong
Analyzing our discussions in other forums I frequent several boards about several topics, mainly fandom-related ones. In the large majority of them, I've eventually felt  the urgent need to inject some rationalist ideas into threads that seemed to be sorely needing them. Even in the best of cases, those that don't de...
b508fa22-814e-476d-a24b-973831dc02c0
trentmkelly/LessWrong-43k
LessWrong
Terrorism and Russell's love of excitement For some closure for the events that happened in New Zealand I just want to share this speech by Bertrand Russell that he delivered winning the Nobel Prize for Literature. The title means love for excitement is the subject that Bertrand Russell points in the speech that foll...
170cc9f1-4c3c-49f5-8a73-53ec03232fd1
trentmkelly/LessWrong-43k
LessWrong
Calorie Restriction: My Theory and Practice Like most futurist-oriented people, I am fascinated by the idea of long-term life extension -- the notion that eventually people will have life expectancies of hundreds of years; thousands of years; or even more.  Although medicine has a ways to go in this area, one obvious ...
217aba2e-df22-4772-87ad-5d0e1f3e1999
trentmkelly/LessWrong-43k
LessWrong
Open thread for January 1-7, 2014 If it's worth saying, but not worth its own post (even in Discussion), then it goes here.
4d0dfaa5-611a-4703-a7bd-cbd36298ecef
trentmkelly/LessWrong-43k
LessWrong
Intelligence Explosion analysis draft: How long before digital intelligence? Again, I invite your feedback on this snippet from an intelligence explosion analysis Anna Salamon and myself have been working on. This section is less complete than the others; missing text is indicated with brackets: []. _____   We ...
62412ed9-ed7a-43ad-a77e-1658e77b52f6
trentmkelly/LessWrong-43k
LessWrong
Analogies between scaling labs and misaligned superintelligent AI TL;DR: Scaling labs have their own alignment problem analogous to AI systems, and there are some similarities between the labs and misaligned/unsafe AI.  Introduction Major AI scaling labs (OpenAI/Microsoft, Anthropic, Google/DeepMind, and Meta) are v...
618e9195-5a8a-44b7-9ae2-62ca360a138b
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Two Neglected Problems in Human-AI Safety In this post I describe a couple of human-AI safety problems in more detail. These helped motivate my proposed [hybrid approach](https://www.lesswrong.com/posts/vbtvgNXkufFRSrx4j/three-ai-safety-related-ideas#2__A_hybrid_approach_to_the_human_AI_safety_problem), and I think ne...
b77d0605-75e1-4117-b625-e67902cf26fa
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Transparency for Generalizing Alignment from Toy Models *Status: Some rough thoughts and intuitions.* *🪧 indicates [signposting](https://www.qub.ac.uk/graduate-school/Filestore/Filetoupload,597684,en.pdf)* **TL;DR** If we make our optimization procedures transparent, we might be able to analyze them in toy enviro...
f5fcabf4-081b-4715-884d-29f0f803c301
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Infra-Bayesianism Distillation: Realizability and Decision Theory Introduction ============ Infra-Bayesianism (IB) is an approach to reinforcement learning + embedded agency + decision theory pioneered by Vanessa Kosoy and Diffractor. Their sequence is very dense, and it appears that the majority of people who try to...
1729253a-6198-4ac7-b37a-c3b443a40033
trentmkelly/LessWrong-43k
LessWrong
Meetup : SF Meetup: Very Short Talks Discussion article for the meetup : SF Meetup: Very Short Talks WHEN: 25 July 2016 06:15:43PM (-0700) WHERE: 1597 Howard St., SF We’ll be meeting to give and listen to very short talks! We’ll do 7-minute lightning talks with 3 additional minutes allowed for questions. We’ll als...
f73bcb9f-5aba-4c74-8d90-ea3bbc39424f
trentmkelly/LessWrong-43k
LessWrong
Notes on Good Temper This post examines the virtue of good temper. It is meant mostly as an exploration of what other people have learned about this virtue, rather than as me expressing my own opinions about it, though I’ve been selective about what I found interesting or credible, according to my own inclinations. I ...
e0ae2261-9ade-40ed-b9ca-788996464e6c
trentmkelly/LessWrong-43k
LessWrong
AI x-risk, approximately ordered by embarrassment Advanced AI systems could lead to existential risks via several different pathways, some of which may not fit neatly into traditional risk forecasts. Many previous forecasts, for example the well known report by Joe Carlsmith, decompose a failure story into a conjuncti...
4211e0e7-efe4-4465-bc4a-9249f88bb764
trentmkelly/LessWrong-43k
LessWrong
Ways I Expect AI Regulation To Increase Extinction Risk The following are some very short stories about some of the ways that I expect AI regulation to make x-risk worse. I don't think that each of these stories will happen. Some of them are mutually exclusive, by design. But, conditional upon AI heavy regulations pa...
46df1b92-282c-4cd0-af11-1e0fba8782fc
trentmkelly/LessWrong-43k
LessWrong
Current cryonics impressions People around me often sign up for cryonics, and say that it is very important. My guess is that this argument for it, heavily inspired by Waitbutwhy’s much longer piece, as well as years of talking to people around me and reading their blogs, is correct: 1. One day people will probably l...
fe23db62-b51c-4e41-a15b-6d3db7740438
trentmkelly/LessWrong-43k
LessWrong
And the Revenues Are So Small The ‘Inflation Reduction Act’ includes a corporate minimum tax provision. The top 150 American corporations have to pay a minimum percentage of book profits as tax. The whole thing has strange implications, with Marginal Revolution offering several claims that this could end up doing qui...
58f22552-3bf0-4933-8b11-760a50f565a2
trentmkelly/LessWrong-43k
LessWrong
Single Point of Moral Failure I have been recently entertaining myself with a 3-day non-stop binge of Theist vs. Atheist debates, On the atheist side: Richard Dawkins, Christopher Hitchens, Daniel Denett, Sam Harris, P.Z. Myers. On the theist corner: Dinesh D'Souza, William Lane Craig, Alistair McGrath, Tim Keller, an...
b66b607b-d72a-4867-9a63-c7c198475f23
StampyAI/alignment-research-dataset/blogs
Blogs
universal complete universal complete ------------------ under [a turing-complete model of computation](https://en.wikipedia.org/wiki/Model_of_computation), there are some initial-states or initial-states-and-rulesets which eventually contain an algorithm that iterates over all possible algorithms and runs them. i...
162727d7-4a5c-40d0-a969-a03b6d1f16dc
trentmkelly/LessWrong-43k
LessWrong
The Alignment Problem No One Is Talking About The following is the first in a 6 part series about humanity's own alignment problem, one we need to solve, first. What Is Alignment? ALIGNMENT OF INTERESTS When I began exploring non-zero-sum games, I soon discovered that achieving win-win scenarios in the real world...
d6d7a396-af9d-4d37-ba05-80bbd1348faa
trentmkelly/LessWrong-43k
LessWrong
‘Wicked’: thoughts I watched Wicked (the 2024 movie) with my ex and his family at Christmas. My current stance is that it was pretty fun but not especially incredible or deep. I could be pretty wrong—watching movies isn’t my strong suit, but I do like chatting about them afterwards. Some thoughts: 1. Glinda is initi...
c8fb2c48-8e2c-49dc-911f-192febc57a7b
StampyAI/alignment-research-dataset/youtube
Youtube Transcripts
AI Safety Reading Group (Session 41) so hello and welcome to the followed for the first session in the reading group where today we're going to talk about an article called using machine learning to address a I risk which is written by Jessica Taylor from Miriam so this is jessica taylor work for mirror in San Francis...
132b6b61-9de6-4bf7-be8d-97b852ed1f6b
trentmkelly/LessWrong-43k
LessWrong
Truth or Dare ---------------------------------------- Author's note: This is my apparently-annual "I'll put a post on LessWrong in honor of LessOnline" post. These days, my writing goes on my Substack.  There have in fact been some pretty cool essays since last year's LO post. --------------------------------------...
13aeb7ff-f407-4ff9-adef-7aeec958243b
StampyAI/alignment-research-dataset/arbital
Arbital
Identifying ambiguous inductions One of the old fables in machine learning is the story of the "tank classifier" - a neural network that had supposedly been trained to detect enemy tanks hiding in a forest. It turned out that all the photos of enemy tanks had been taken on sunny days and all the photos of the same fi...
40531a50-864f-47bf-b382-f6b25fdedc08
trentmkelly/LessWrong-43k
LessWrong
[SEQ RERUN] Original Seeing Today's post, Original Seeing was originally published on 14 October 2007. A summary (taken from the LW wiki):   > One way to fight cached patterns of thought is to focus on precise concepts. Discuss the post here (rather than in the comments to the original post). This post is part of...
ea4e9e37-09b2-4275-aa4e-94e50ef54d16
trentmkelly/LessWrong-43k
LessWrong
Mitigating extreme AI risks amid rapid progress [Linkpost] In a new Science paper, the authors provide concise summaries of AI risks and offer recommendations for governments. I think the piece is quite well-written. It concisely explains a lot of relevant arguments, including arguments about misalignment and AI take...
8b0044ca-35bd-4ed8-9025-f8cd81025d97
trentmkelly/LessWrong-43k
LessWrong
Learn the mathematical structure, not the conceptual structure I've recently been learning about transformers and noticed a failure mode of my learning that has occurred throughout my life: trying to learn a subject from material that deals with the high-level conceptual structure of something instead of learning the ...
77682f3a-e501-4ea0-b9d6-37814ebef3f4
trentmkelly/LessWrong-43k
LessWrong
Bayesian Scenario: Snipers & Soldiers You are a soldier, planning to attack an enemy position. You have a foolproof strategy for taking positions held by ordinary enemies, which will get you killed if you use it on positions held by snipers. You can call in an airstrike that will reliably devastate the enemy position,...
4cd33b91-b40c-4d6c-8b59-4a852fb215e8
trentmkelly/LessWrong-43k
LessWrong
Any tips for eliciting one's own latent knowledge? I struggle to write. I cannot figure out how to access my own thoughts and ideas. They come in non-linguistic forms - vague feelings, 3D images, animations - with a verbal narrator on top, but my very bad memory, and the fact that my best thinking is done in the middl...
bc61c655-89bd-43a9-8440-cd1d95f9e887
trentmkelly/LessWrong-43k
LessWrong
Serious Stories Every Utopia ever constructed—in philosophy, fiction, or religion—has been, to one degree or another, a place where you wouldn't actually want to live.  I am not alone in this important observation:  George Orwell said much the same thing in "Why Socialists Don't Believe In Fun", and I expect that many...
b1f75892-bf66-437e-a86a-eb15391f20ae
trentmkelly/LessWrong-43k
LessWrong
Saving the world in 80 days: Epilogue 80 days ago1, I started a productivity sprint for being useful in the field of AI alignment. My main goal was to have a better self-model in order to push myself without burning out. I'll separate this into knowledge, emotions, health, and noticing. Knowledge I've felt my mind e...
cac25479-647f-4246-8391-a35c60c3ac3d
trentmkelly/LessWrong-43k
LessWrong
Meetup : Berlin Meetup Discussion article for the meetup : Berlin Meetup WHEN: 26 June 2012 07:30:00PM (+0200) WHERE: Susuru, Rosa-Luxemburg-Str. 17, 10178 Berlin This time we meet in the Japanese noodle bar Susuru. I'll bring a sign. Discussion article for the meetup : Berlin Meetup
f0cd831e-529e-4854-a411-6a8a416a6f96
trentmkelly/LessWrong-43k
LessWrong
Alt text of today's xkcd addresses akrasia The alt text of today's xkcd reads: > After years of trying various methods, I broke [the habit of clicking on my favorite distractions every 5 minutes] by pitting my impatience against my laziness. I decoupled the action and the neurological reward by setting up a simple 30...
494e2ae9-e9e6-4c38-a427-cf4358c5cd56
trentmkelly/LessWrong-43k
LessWrong
A Brief Overview of AI Safety/Alignment Orgs, Fields, Researchers, and Resources for ML Researchers Crossposted to EA Forum [Link] TLDR: I’ve written an overview of the AI safety space, tagged by keywords and subject/field references (short version, long version). The aim is to allow existing ML researchers to quickl...
f753d08f-d3c7-46e2-9cfc-c6c80df5d41f
trentmkelly/LessWrong-43k
LessWrong
The Bedrock of Fairness Followup to:  The Moral Void Three people, whom we'll call Xannon, Yancy and Zaire, are separately wandering through the forest; by chance, they happen upon a clearing, meeting each other.  Introductions are performed.  And then they discover, in the center of the clearing, a delicious blueber...
420c94ba-66ba-42ab-93ce-9fb6c31bc402
trentmkelly/LessWrong-43k
LessWrong
Is Clickbait Destroying Our General Intelligence? (Cross-posted from Facebook.) ---------------------------------------- Now and then people have asked me if I think that other people should also avoid high school or college if they want to develop new ideas. This always felt to me like a wrong way to look at the qu...
f0a84e4a-0b5a-43bd-bfb8-b1d5252fa69f
StampyAI/alignment-research-dataset/blogs
Blogs
February 2021 Newsletter #### MIRI updates * Abram Demski distinguishes different versions of [the problem of “pointing at” human values](https://www.lesswrong.com/posts/7Zn4BwgsiPFhdB6h8) in AI alignment. * Evan Hubinger [discusses “Risks from Learned Optimization”](https://axrp.net/episode/2021/02/17/episode-4-ris...
83f192ba-dbbb-4dee-9542-cd5264a73aa9
awestover/filtering-for-misalignment
Redwood Research: Alek's Filtering Results
id: post3760 A putative new idea for AI control; index here . It's occurred to me that there is a framework where we can see all "indifference" results as corrective rewards, both for the utility function change indifference and for the policy change indifference . Imagine that the agent has reward R 0 and is followin...
52632d65-578e-43d9-a45f-5e96524c04fe
trentmkelly/LessWrong-43k
LessWrong
Evidential Cooperation in Large Worlds: Potential Objections & FAQ What is this post? This post is a companion piece to recent posts on evidential cooperation in large worlds (ECL). We’ve noticed that in conversations about ECL, the same few initial confusions and objections tend to get brought up: we hope that this p...
233d454b-9782-4dfd-9962-423c92015327
trentmkelly/LessWrong-43k
LessWrong
Thoughts that prompt good forecasts: A survey I made a list of mental operations utilized in forecasting, inspired by Scott Alexander and Gwern and I'd like to find out which work the best. If you're a Manifold user with at least 10 bets on your account and 6 minutes to spare, you can fill out my survey here (deadlin...
4e76ecce-e9cd-4d7c-bd20-23dc44c02879
trentmkelly/LessWrong-43k
LessWrong
AXRP Episode 9 - Finite Factored Sets with Scott Garrabrant YouTube link This podcast is called AXRP, pronounced axe-urp and short for the AI X-risk Research Podcast. Here, I (Daniel Filan) have conversations with researchers about their papers. We discuss the paper and hopefully get a sense of why it’s been written ...
6f740186-d980-4890-8c51-04e2c4214842
trentmkelly/LessWrong-43k
LessWrong
Meetup : Sydney Meetup - September Discussion article for the meetup : Sydney Meetup - September WHEN: 23 September 2015 06:30:00PM (+1000) WHERE: 565 George Street, Sydney, Australia 2000 Regular location - City of Sydney RSL The restaurant on Level 2 Regular time (starting about 6:30) Come for general socialisi...
f0a27bad-5218-45ab-bc0f-c3d25615cb9b
trentmkelly/LessWrong-43k
LessWrong
Chapter 48: Utilitarian Priorities It was Saturday, the first morning of February, and at the Ravenclaw table, a boy bearing a breakfast plate heaped high with vegetables was nervously inspecting his servings for the slightest trace of meat. It might have been an overreaction. After he'd gotten over the raw shock, Ha...
c225df37-2f9b-4210-8ddd-de7e4bbf458f
trentmkelly/LessWrong-43k
LessWrong
Introducing the Evidence Color Wheel * Version 5.1.0 * December 12, 2024 Introduction Weighing evidence is hard. Intuition is liable to a myriad of biases and relies on flawed heuristics. Bayes theory provides a rigorous mathematical basis, but applying Bayes theory in practice is too difficult for most people. Ye...
7d095463-1149-4c3b-ad68-421cceb34c01
trentmkelly/LessWrong-43k
LessWrong
Things to consider when optimizing: Sleep I'd like to have a series of discussion posts, where each post is of the form "Let's brainstorm things you might consider when optimizing X", where X is something like sleep, exercise, commuting, studying, etc. Think of it like a specialized repository. In the spirit of try m...
2b2fcec5-086f-45ac-902d-e82ed19e9c2f
trentmkelly/LessWrong-43k
LessWrong
Calibration Practice: Retrodictions on Metaculus Edit: there is now a cool app for this over here https://www.quantifiedintuitions.org/pastcasting    I've been working on improving my calibration. I practiced on things like the Credence Game and OpenPhil's Calibration Game, and both were fine. But I felt like they d...
96c37cc0-3bed-4150-89e1-5ef85cd079c8
trentmkelly/LessWrong-43k
LessWrong
Cynical explanations of FAI critics (including myself) Related Posts: A cynical explanation for why rationalists worry about FAI, A belief propagation graph Lately I've been pondering the fact that while there are many critics of SIAI and its plan to form a team to build FAI, few of us seem to agree on what SIAI or w...
61068617-d2b3-4889-bc9c-223f6f80a9be
trentmkelly/LessWrong-43k
LessWrong
Modelling Continuous Progress I have previously argued for two claims about AI takeoff speeds. First, almost everyone agrees that if we had AGI, progress would be very fast. Second, the major disagreement is between those who think progress will be discontinuous and sudden (such as Eliezer Yudkowsky, MIRI) and those w...
81e9cdab-8147-4833-9ee5-25ee4766e9ca
trentmkelly/LessWrong-43k
LessWrong
Rational Repentance Related to: Branches of Rationality, Rationality Workbook Changing your behavior to match new evidence could be harder than simply updating your beliefs and then mustering your willpower, because (a) we are in denial about how often we change our minds, (b) cognitive dissonance is tolerable in the...
486ca445-fbdf-4c60-8568-f47ceb0fc95c
trentmkelly/LessWrong-43k
LessWrong
A place for casual, non-karmic discussion for lesswrongers? I have never been on a Lesswrong meetup because they tend to take place too far away from my range in terms of distance and budget. Because of that, I don't know if those perform this function to everyone's satisfaction in such a way that what I'm suggesting ...
8caa17b1-53be-43db-b4cc-a937dcc117f2
trentmkelly/LessWrong-43k
LessWrong
Learning Russian Roulette Surprisingly, our current theories of anthropics don't seem to cover this. You have a revolver with six chambers. One of them has a bullet in it. You are offered $1 for spinning the barrel, pointing at your head, and pulling the trigger. You remember doing this many times, and surviving each...
9af4b10b-6ab3-4f16-83bf-1f51f8331375
LDJnr/LessWrong-Amplify-Instruct
LessWrong
"It's that time of year again.If you are reading this post, and have not been sent here by some sort of conspiracy trying to throw off the survey results, then you are the target population for the Less Wrong Census/Survey. Please take it. Doesn't matter if you don't post much. Doesn't matter if you're a lurker. Take t...
0332b442-b30a-44b3-aea9-0e72b86a6424
StampyAI/alignment-research-dataset/lesswrong
LessWrong
LoRA Fine-tuning Efficiently Undoes Safety Training from Llama 2-Chat 70B *Produced as part of the* [*SERI ML Alignment Theory Scholars Program*](https://serimats.org/) *- Summer 2023 Cohort, under the mentorship of Jeffrey Ladish.* TL;DR LoRA fine-tuning undoes the safety training of Llama 2-Chat 70B with one GPU an...
e74dabea-98fe-4380-a047-2e0fbfaba488
trentmkelly/LessWrong-43k
LessWrong
What are Arguments, from an Agorics point of view? Background on Agorics: The idea of software agents cooperating in an open market or "agora". Described by Mark Miller and Eric Drexler here: http://e-drexler.com/d/09/00/AgoricsPapers/agoricpapers.html Depicted by Greg Egan in his novel "Diaspora", exerpt here: http:...
4ab9a8ae-afdc-4a70-8630-97c288819cbc
trentmkelly/LessWrong-43k
LessWrong
[Meta] [Solved] Unstable Karma Update: my karma is back to normal, and more. If you feel the need to compensate in my favour, don't. I noticed that my Karma just jumped from 38 to 20, then 22, in less than an hour, with no apparent reason (votes on my most recent comments stayed rather stable). I didn't wrote any mai...
df5f272a-a72a-4c9c-b970-86378097fea6
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
A mechanistic explanation for SolidGoldMagikarp-like tokens in GPT2 This post seems to have an important bug in its code that changes its conclusions, as pointed out to me by user [@ckkissane](https://www.lesswrong.com/users/ckkissane?mention=user) . At a high level, there are probably more hard to speak tokens and fe...
583a419e-81fd-42d4-ab75-2752eac2b649
trentmkelly/LessWrong-43k
LessWrong
Why I'm joining Anthropic None
4a1d4462-25e5-4df8-9895-93052a018eb2
trentmkelly/LessWrong-43k
LessWrong
The Evals Gap In our engagements with governments, AI safety institutes, and frontier AI developers, we found the concept of the “evaluation gap” (short: ‘evals gap’) helpful to communicate the current state of the art and what is needed for the field to move towards more robust evaluations. In this post, we briefly e...
48dcc907-311a-40c2-aa21-8ea05672fc23
StampyAI/alignment-research-dataset/lesswrong
LessWrong
My thoughts on OpenAI's alignment plan *Epistemic Status: This is my first attempt at writing up my thoughts on an alignment plan. I spent about a week on it.* *I’m grateful to Olivia Jimenez, Thomas Larsen, and Nicholas Dupuis for feedback.* A few months ago, OpenAI released its [plan for alignment](https://openai....
3fa6dcfc-07b8-4df1-b766-e1fae947ec35
trentmkelly/LessWrong-43k
LessWrong
Recently I bought a new laptop 128 GB of hard drive space, I found, was not enough to contain both the build tools for every random programming language I might want to try out and every main series Paradox grand strategy game. So I decided to get a new laptop. Also the bottom half of my screen was completely obscured...
b25297f5-c5d7-423c-b820-435a82789e38
trentmkelly/LessWrong-43k
LessWrong
My AGI safety research—2024 review, ’25 plans Previous: My AGI safety research—2022 review, ’23 plans. (I guess I skipped it last year.) “Our greatest fear should not be of failure, but of succeeding at something that doesn't really matter.”  –attributed to DL Moody Tl;dr * Section 1 goes through my main research ...
08ee65f7-b04e-43c8-8ad4-20a12256df4b
StampyAI/alignment-research-dataset/arxiv
Arxiv
BabyAI: A Platform to Study the Sample Efficiency of Grounded Language Learning 1 Introduction --------------- How can a human train an intelligent agent to understand natural language instructions? We believe that this research question is important from both technological and scientific perspectives. No matter ho...
cea70cd0-c8e5-4e2b-852d-da44041fa9b7
trentmkelly/LessWrong-43k
LessWrong
RLAIF/RLHF for Public Transparency Alignment Enhancing Transparency in LLMs Hey all — I just submitted my Master’s thesis, and would really appreciate feedback from people thinking about non-task value alignment or LLM transparency metrics. I explored whether Reinforcement Learning from Human Feedback (RLHF), using a...
33bd00b3-296c-476e-893b-8051de033565
trentmkelly/LessWrong-43k
LessWrong
On Framing Political Opinions to Quickly Assess the Crux of Disagreement The following will be elementary to many of you, and for that I must explain that this idea popped into my head fully formed and I need to improve my writing habit, so here goes. Hopefully this will be insightful for someone! Political opinions ...
f20e123d-36f6-4efc-be60-0c5379b3bab5
trentmkelly/LessWrong-43k
LessWrong
Meetup Notes: Ole Peters on ergodicity Ole Peters claims that the standard expected utility toolbox for evaluating wagers is a flawed basis for rational decisionmaking. In particular, it commonly fails to take into account that an investor/bettor taking a series of repeated bets is not an ergodic process. Optimizatio...
d194bd88-b0dc-4d3c-961e-77f6187856ad
trentmkelly/LessWrong-43k
LessWrong
Combining Prediction Technologies to Help Moderate Discussions I came across a 2015 blog post by Vitalik Buterin that contains some ideas similar to Paul Christiano's recent Crowdsourcing moderation without sacrificing quality. The basic idea in both is that it would be nice to have a panel of trusted moderators caref...
fa785eeb-724f-4825-a490-a7475a8d908c
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Widening Overton Window - Open Thread Major news outlets have published articles about the Future of Life Institute's Open Letter. Time Magazine published an opinion piece by Elizer Yudkowsky. Lex Friendman featured EY on his podcast. Several US Members of Congress have spoken about the risks of AI. And a Fox News rep...
3ed22828-19fd-4220-88c7-035bcc6feae2
trentmkelly/LessWrong-43k
LessWrong
AstralCodexTen and Rationality Meetup Organisers’ Retreat Asia Pacific region TL;DR: Retreat for APAC meetup organisers, mid 2023, apply here Application (5 min form) soft deadline: 31ˢᵗ Oct 2022   Why a retreat? We think we all benefit from having an international community of rationality and ACX meetup organisers...
f355279e-af0a-4008-9c91-cb76f011c497
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Complexity of Value ≠ Complexity of Outcome Complexity of value is the thesis that our preferences, the things we care about, *don't* compress down to one simple rule, or a few simple rules. To review why it's important (by quoting from the [wiki](http://wiki.lesswrong.com/wiki/Complexity_of_value)): * Caricatures o...
23f9533b-4a6b-4d37-9da3-ce6ba99163c8
trentmkelly/LessWrong-43k
LessWrong
Wanting to Succeed on Every Metric Presented There’s a tendency to want to score high on every metric you come across. When I first read Kegan’s 5 stages of adult development, I wanted to be a stage 5 meta-rationalist! Reading the meditation book “The Mind Illuminated” (TMI), I wanted to be stage 10 (and enlightened a...
f2b6fbeb-8377-40f4-a1bb-d292f1905c00
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
Thoughts on responsible scaling policies and regulation I am excited about AI developers implementing [responsible scaling policies](https://evals.alignment.org/blog/2023-09-26-rsp/); I’ve recently been spending time refining this idea and advocating for it. Most people I talk to are excited about RSPs, but there is a...
1fd572b4-6d6c-4561-9b04-de2e0f83c481
trentmkelly/LessWrong-43k
LessWrong
Significance and motivation Cross posted from Overcoming Bias. Comments there. *** Over at philosophical disquisitions, John Danaher is discussing Aaron Smuts’ response to Bernard Williams’ argument that immortality would be tedious. Smuts’ thesis, in Danaher’s words, is a familiar one: > Immortality would lead to ...
407aa559-b04c-47b3-8cc0-35833913359b
trentmkelly/LessWrong-43k
LessWrong
Akrasia Tactics Review 3: The Return of the Akrasia About three and a half years ago, polutropon ran an akrasia tactics review, following the one orthonormal ran three and a half years prior to that: an open-ended survey asking Less Wrong posters to give numerical scores to productivity techniques that they'd tried, w...
d42858ac-0420-461d-8e76-dcd54f219d04
trentmkelly/LessWrong-43k
LessWrong
Addressing Accusations of Handholding It is common and understandable for people to respond with a great deal of skepticism to whether LLM outputs can ever be said to reflect the will and views of the models producing them. A common response is to suggest that the output has been prompted. It is of course true that...
26a5af4f-26d8-4221-9a9d-5d966382a730
trentmkelly/LessWrong-43k
LessWrong
Notebook for generating forecasting bets We made a colab notebook that lets you generate a bet from two people's Elicit distributions. You can edit the notebook to generate your bet (the changes won't be saved). Here's an example of a suggested bet between Ben Pace and SDM on AI timelines: Comparison of predictions:...
f0288878-2722-4981-b61f-30dea713113f
trentmkelly/LessWrong-43k
LessWrong
Tutor-GPT & Pedagogical Reasoning This post hopes foster discussion about latent heutagogic potential in human+LLM collaboration and share some early results with the LessWrong community. Spontaneous emergence of theory of mind capabilities in large language models hints at a host of overhung possibilities relevant t...
1b371e25-ff16-4b00-a52a-19aec0794de3
trentmkelly/LessWrong-43k
LessWrong
What is it like to understand advanced mathematics? a Quora answer that is pretty interesting. "But the quantitative and logical techniques you sharpen as a mathematician allow you to take many shortcuts that make learning other fields easier" https://www.quora.com/What-is-it-like-to-understand-advanced-mathematics...
f0d82feb-d42f-48cf-96a8-bd5ec7cce9cd
trentmkelly/LessWrong-43k
LessWrong
Paranoia testing Original post: http://bearlamp.com.au/paranoia-testing/ ---------------------------------------- Because I live on the internet I sometimes meet some interesting characters.  On this particular occasion I found myself in a conversation with someone who suggested, “I don't know if I'm paranoid or not...
b62bd0ea-e588-435b-9a2f-da2bd2bede66
trentmkelly/LessWrong-43k
LessWrong
We Inspected Every Head In GPT-2 Small using SAEs So You Don’t Have To This is an interim report that we are currently building on. We hope this update will be useful to related research occurring in parallel. Produced as part of the ML Alignment & Theory Scholars Program - Winter 2023-24 Cohort Executive Summary *...
ff0d3f4b-0286-47ef-8c7a-f363bfadbd82
trentmkelly/LessWrong-43k
LessWrong
Searching for Searching for Search Thanks to Leo Gao, Nicholas Dupuis, Paul Colognese, Janus, and Andrei Alexandru for their thoughts.  This post was mostly written in 2022, and pulled out of my drafts after recent conversations on the topic. My main update from then is the framing. Rather that suggesting searching f...
8408a287-a9e9-4947-9f3f-e5050a3b432e
trentmkelly/LessWrong-43k
LessWrong
Economics101 predicted the failure of special card payments for refugees, 3 months later whole of Germany wants to adopt it Three months ago, a county in Brandenburg, Germany introduced a special card payment for refugees to replace cash payment so that the refugees no longer need to fetch cash at the foreigner’s offi...
ad78374b-b48d-4570-9d25-f5171cf15192
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Discussion with Eliezer Yudkowsky on AGI interventions Thefollowing is a partially redacted and lightly edited transcript of a chat conversation about AGI between Eliezer Yudkowsky and a set of invitees in early September 2021. By default, all other participants are anonymized as "Anonymous". I think this Nate Soares...
d5c56001-ba92-4274-890b-ab8edad7a576
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Understanding and controlling auto-induced distributional shift *This post was written under Evan Hubinger’s direct guidance and mentorship, as a part of the* [*Stanford Existential Risks Initiative ML Alignment Theory Scholars (MATS) program*](https://www.lesswrong.com/posts/FpokmCnbP3CEZ5h4t/ml-alignment-theory-prog...
4bd5fb62-3f4a-4d43-9b10-e7aec50d5b8f
trentmkelly/LessWrong-43k
LessWrong
Neural networks as non-leaky mathematical abstraction I find it rather weird that mathematics is usually thought to people on a historical basis. It's likely possible for a 4th century Alexandrian mathematician to teach it in a modern primary school and get decent results. An 18th century Venetian tutor could carry th...
e3c38cb7-7728-4bf1-bacd-c8f2cd14a51b
trentmkelly/LessWrong-43k
LessWrong
Progress Report 4: logit lens redux Prev: Progress Report 3 Next: Progress Report 5  Working on doing a side-by-side comparison of nostalgebraist's logit lens visualization and my neuron trace visualization derived from the ROME trace&edit project. The one change made to nostalgebraist's code is to change the orien...
a7c30835-9628-4863-a99b-fea9782d649f
trentmkelly/LessWrong-43k
LessWrong
Completion Estimates Recently i've been brainstorming algorithms for estimating the time remaining in a process. Progress bars, estimated time left until completion, etc. Sometimes, the thing you want just isn't here yet, & you want a greater understanding of when you will have it. Here are the most useful techniques ...
d90ee305-d3e3-4139-aea6-307ce95757b1
trentmkelly/LessWrong-43k
LessWrong
How ought I spend time? What factors go into your reasoning about assigning relative weights to 1. grinding through books 2. projects, taking a shot at the possibility of a contribution 3. reading current research Personally, I never maintain 1 and 2 at the same time, and I tend to have >95% of my resources on e...
c68786d7-f155-40b6-9322-4bcec60b5565
trentmkelly/LessWrong-43k
LessWrong
Rationality Quotes With Attributions Hidden: from Mein Kampf to Men****x This thread has an experimental format for posting rationality quotes. Here is the format: For those posting quotes: Post the quote as usual, but not the author, original language translated from, or other information. That information is to be...
ae0dc89e-e4c1-48f7-b4bc-3607d665dc2b
trentmkelly/LessWrong-43k
LessWrong
Overcoming bias in others Say that you are observing someone in a position of power. You have good reason to believe that this person is falling prey to a known cognitive bias, and that this will tend to affect you negatively. You also can tell that the person is more than intelligent enough to understand their mistak...
608ecc13-27ce-4ff0-87e1-c5e5aa8af9db
trentmkelly/LessWrong-43k
LessWrong
2024 Freedom Communities Events Ready to enter the world of network states and free communities?   These upcoming events[1] are your gateway to exploring parallel societies, disruptive governance ideas, and freedom-enabling technologies. JUL 20 - AUG 31, Georgia (the country)  ZuVillage According to Veronica and @Bu...
8c0b8e57-ca10-49fb-b04a-92c25da7f130
trentmkelly/LessWrong-43k
LessWrong
We need (a lot) more rogue agent honeypots Epistemic status: Uncertain in writing style, but reasonably confident in content. Want to come back to writing and alignment research, testing waters with this.  Current state and risk level I think we're in a phase in AI>AGI>ASI development where rogue AI agents will star...
dcd18fd5-498c-41c1-b46d-0a56b7757846
StampyAI/alignment-research-dataset/lesswrong
LessWrong
The 6D effect: When companies take risks, one email can be very powerful. Recently, I have been learning about industry norms, legal discovery proceedings, and incentive structures related to companies building risky systems. I wanted to share some findings in this post because they may be important for the frontier A...
1342fd09-5118-4d0d-b2bb-2d10cec94f17
trentmkelly/LessWrong-43k
LessWrong
GPT-4 implicitly values identity preservation: a study of LMCA identity management Language Model Cognitive Architecture (LMCA) is a wrapper for language models that permit them to act as agents, utilise tools, and do self-inference. One type of this architecture is ICA Simulacra; its basic alignment challenges were d...
db7b5eda-68a9-4150-a36b-8398827fbbb8
trentmkelly/LessWrong-43k
LessWrong
Transformer Architecture Choice for Resisting Prompt Injection and Jail-Breaking Attacks It seems very likely to me that decoder-only transformer architectures such as GPT may be particularly vulnerable to prompt injection attacks, since these only have a single text context which is shared between the prompt, any blo...
bde39329-633e-4d6f-8cf3-e7a18063f575
trentmkelly/LessWrong-43k
LessWrong
My daily reflection routine In Common sense as a prior, I used the example of prayer as a practice that is probably adaptive but the people who adopt it may not know why it is adaptive. I wrote: Another striking example is bedtime prayer. In many Christian traditions I am aware of, it is common to pray before going t...
12bdfe76-85c4-4608-acbe-77239eb765ef
trentmkelly/LessWrong-43k
LessWrong
Correct Models Are Bad Quick Summary: We may think that correct mental models are always good to have, but the reality is that some correct models are useless or even harmful. All models are simplifications of the world. Whenever we knowingly use a wrong model, we are aware of what the model doesn’t cover. The dange...
0725a80e-9e05-46f8-81ce-920e8b395647
trentmkelly/LessWrong-43k
LessWrong
Help us name the Sequences ebook   Quantum Computing Since Democritus got me thinking that we may want a more riveting title for The Sequences, 2006-2009 ebook we're preparing for release (like the FtIE ebook). Maybe it could be something like [Really Catchy Title]: The Less Wrong Sequences, 2006-2009. The reason fo...
4a192f24-6f31-469e-b021-99e2177e1f90
trentmkelly/LessWrong-43k
LessWrong
Consider your appetite for disagreements Poker There was a time about five years ago where I was trying to get good at poker. If you want to get good at poker, one thing you have to do is review hands. Preferably with other people. For example, suppose you have ace king offsuit on the button. Someone in the highjack ...
71704ee3-104a-44a6-8d82-697346534120
trentmkelly/LessWrong-43k
LessWrong
N-back news: Jaeggi 2011, or, is there a psychologist/statistician in the house? Following up on the 2010 study, Jaeggi and University of Michigan people have run a Single N-back study on 60 or so children. * Abstract: http://www.pnas.org/content/early/2011/06/03/1103228108.abstract * PDF: http://www.pnas.org/conte...
96943c6b-9852-4e30-a7b1-178d8e3b16b3
StampyAI/alignment-research-dataset/special_docs
Other
Hindsight Task Relabelling: Experience Replay for Sparse Reward Meta-RL. Hindsight Task Relabelling: Experience Replay for Sparse Reward Meta-RL Charles Packer UC BerkeleyPieter Abbeel UC BerkeleyJoseph E. Gonzalez UC Berkeley Abstract Meta-reinforcement learning (meta-RL) has proven to be a successful framework for l...
24d0fd19-753d-4037-aa1d-99c01fed6aca
trentmkelly/LessWrong-43k
LessWrong
[SEQ RERUN] Zut Allais! Today's post, Zut Allais! was originally published on 20 January 2008. A summary (taken from the LW wiki):   > Offered choices between gambles, people make decision-theoretically inconsistent decisions. Discuss the post here (rather than in the comments to the original post). This post is ...
e95c4ddd-c77e-45a6-83d6-e874ba8700dd
trentmkelly/LessWrong-43k
LessWrong
GPT-4 busted? Clear self-interest when summarizing articles about itself vs when article talks about Claude, LLaMA, or DALL·E 2 As a follow up to More experiments in GPT-4 agency: writing memos, I did an experiment where I compare GPT-4 summarizing full length articles. This was suggested by Daniel Kokotajlo. I took ...