id
stringlengths
36
36
source
stringclasses
15 values
formatted_source
stringclasses
13 values
text
stringlengths
2
7.55M
606dd061-e58a-4daa-8db0-225f25512f2a
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Some Notes on the mathematics of Toy Autoencoding Problems Anthropic's recent mechanistic interpretability paper, [Toy Models of Superposition](https://transformer-circuits.pub/2022/toy_model/index.html), helps to demonstrate the conceptual richness of very small feedforward neural networks. Even when being trained on...
ac383179-6884-4e2b-a0d9-c0e4bca21376
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Thoughts on refusing harmful requests to large language models ![](https://res.cloudinary.com/lesswrong-2-0/image/upload/v1674085457/mirroredImages/Dan6iKFruioYmZafw/zgqrsnb4ymfrc5sqpk5y.png)<https://twitter.com/antimatter15/status/1602469101854564352> Currently, large language models ([ChatGPT](https://openai.com/bl...
0a29d65f-3fce-4373-ae9c-21c8aecb4c88
trentmkelly/LessWrong-43k
LessWrong
Are there cognitive realms? [Metadata: crossposted from https://tsvibt.blogspot.com/2022/11/are-there-cognitive-realms.html. First completed November 16, 2022. This essay is more like research notes than exposition, so context may be missing, the use of terms may change across essays, and the text may be revised later...
87d5de37-522a-4f63-8e27-20ef5689b63f
trentmkelly/LessWrong-43k
LessWrong
Core Concept Conversation: What is wealth? This is a Core Concept Conversation post. The topic is what wealth is and what it means to create it, largely based off of Paul Graham's essay How to Make Wealth. I'm not clear on what wealth actually is. PG seems to contradict himself in the essay. In footnote #4 he says: ...
a5cc1e4c-e065-4e43-bd02-a2bcad08ce11
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Inference Speed is Not Unbounded *[The intro of this post has been lightly edited since it was first posted to address some comments. I have also changed the title to better reflect my core argument. My apologies if that is not considered good form.]* This post will be a summary of some of my ideas on what intelligen...
87ae1a55-1c71-487d-93dc-814d7fe283ab
trentmkelly/LessWrong-43k
LessWrong
Akrasia Tactics Review 2: The Akrasia Strikes Back About three and a half years ago, orthonormal ran an akrasia tactics review: an open-ended survey asking Less Wrong posters to give numerical scores to productivity techniques that they'd tried, with the goal of getting a more objective picture of how well different t...
bc20aa65-6df9-4d48-9ac3-e492c1ce9766
trentmkelly/LessWrong-43k
LessWrong
Universal Modelers Alternative title: A simple framework for thinking about applying machine learning to science and engineering in a way that both produces reliable results and poses comparatively ~0 risk of destroying humanity Mathematics was designed for the human brain. In our times, this seems like a rather sil...
6eb9817f-c45c-498c-bc74-9a0fae63d617
trentmkelly/LessWrong-43k
LessWrong
Evaluation Evidence Reconstructions of Mock Crimes Submission 3 Epistemic Status: Exploratory Introduction I want to train people to decide the truth about events, through practices reconstructing events about which the truth is independently known I am asking the help of people who are interested in applying reaso...
90e5070b-1270-4191-9425-aee0d6f29730
trentmkelly/LessWrong-43k
LessWrong
Logical Counterfactuals Consistent Under Self-Modification I argue that the correct notion of logical counterfactual is the one which an agent would self-modify to have if the agent lacked it, and that this significantly constrains the possibilities. The argument is rather informal, and I'm not really sure if many of ...
9f3c2e4f-9f15-4f00-a54b-d1b8f3e80222
trentmkelly/LessWrong-43k
LessWrong
Thoughts on Neuralink update? Here is the full stream. I haven't seen LessWrong discuss BCI technology all that much, so I'm curious what some of the people here think about the current SOTA, and whether such devices will eventually be able to live up to their lofty promises.
2c8f6fb4-e1b3-4d49-a6bf-34891d1544b1
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Quantum theory cannot consistently describe the use of itself As I understand it, the new thought experiment in QM demonstrates that all currently known interpretations of QM are wrong, if we extrapolate Winger friend thought experiment to next level, where different friends are using different theories to interpret e...
a19632e8-8459-4bd1-90ed-bd27ad271814
StampyAI/alignment-research-dataset/youtube
Youtube Transcripts
Connor Leahy | Promising Paths to Alignment all right well hello everybody today i'm going to be talking about promising paths to alignment and let's get just right into it there are none thanks for coming to my talk i really hope you enjoyed are there any questions all right obviously i'm joking but to every joke the...
aa2f12fa-5588-4551-a64d-afc83b10e621
trentmkelly/LessWrong-43k
LessWrong
Defining Optimization in a Deeper Way Part 2 We have successfully eliminated the concepts of null actions and nonexistence from our definition of optimization. We have also eliminated the concept of repeated action. We are halfway there, and now have to eliminate uncertainty and absolute time. Then we will have achiev...
205a2a98-7433-4901-9bfb-29e71521663d
trentmkelly/LessWrong-43k
LessWrong
Discussion: Pathways for the Aspiring AGI Researcher? Follow up to: Best career models for doing research? First, I must apologize for the somewhat self-serving post, but as it is in the discussion section I hope that this can be forgiven. Also, I would not be surprised if there are at least a few college age people ...
ae3dea76-724a-4f6a-ba92-b5f119a063b1
trentmkelly/LessWrong-43k
LessWrong
Houston Hackerspace Meetup : Saturday June 4, 2:00PM Saturday June 4, 2:00PM TX/RX Labs Hackerspace 2010 Commerce St Houston, TX 77002   The third meeting of the Houston Less Wrong group will be this coming Saturday at TX/RX Labs at 2:00PM. We had some success last time, with Orange Cat (I apologize if I got the ...
d84beba5-6018-48aa-a8f8-8ce5ea9951fd
trentmkelly/LessWrong-43k
LessWrong
Aristotle, Aquinas, and the Evolution of Teleology: From Purpose to Meaning. The pursuit of understanding purpose and meaning has been a cornerstone of Western philosophy since its earliest days. Aristotle, one of the greatest thinkers of antiquity, sought to explain the natural world and human flourishing through the...
21d1c524-55d5-4f5b-92e5-f5f0a8fb74cc
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Truthful LMs as a warm-up for aligned AGI *This post is heavily informed by prior work, most notably that of Owain Evans, Owen Cotton-Barratt and others (*[*Truthful AI*](https://www.alignmentforum.org/posts/aBixCPqSnTsPsTJBQ/truthful-ai-developing-and-governing-ai-that-does-not-lie)*), Beth Barnes (*[*Risks from AI p...
d1cd462c-ed64-4954-868f-57cbf905afa2
trentmkelly/LessWrong-43k
LessWrong
Continuing the takeoffs debate Here’s an intuitively compelling argument: only a few million years after diverging from chimpanzees, humans became much more capable, at a rate that was very rapid compared with previous progress. This supports the idea that AIs will, at some point, also start becoming more capable at a...
ec6b7c6b-a67c-4171-bea3-29116a7ef561
trentmkelly/LessWrong-43k
LessWrong
Optimizing Problem-Solving Strategies Through Prediction Markets The Current Issue Suppose you'd like to improve in chess as a total beginner so that within a month training one hour a day your rating in rapids is as high as possible. What do you do? Watch a YouTube video? Read a book? Ask ChatGPT? What about learnin...
5b58298a-3512-4bbe-9824-5521a482e241
trentmkelly/LessWrong-43k
LessWrong
Did slavery make the US an economic superpower and would the industrial revolution have happened without it?
2dd22091-f9ce-47b4-b10d-beff13e6e035
trentmkelly/LessWrong-43k
LessWrong
What causes randomness? If I clone two earths right when the first unicellular organisms formed, with everything down to the atom similar, will the two earths (or maybe just take them as universes composed of matter/energy) be the same? If not, why? What causes that difference? Is it randomness? If so, what causes thi...
d8c2f1c0-13c1-4f5c-a01c-dfe3dee5843f
trentmkelly/LessWrong-43k
LessWrong
Experimental Open Threads Meta-tations on Moderation: Towards Public Archipelago suggested that we should be running more experiments with how discussion is run and listed several examples in the post. What if we ran monthly experimental open threads? I can imagine several formats that would work better as open thread...
3642cc54-0d77-49f7-bf5a-c9aaa34ded88
trentmkelly/LessWrong-43k
LessWrong
A conversation on theory of mind, subjectivity, and objectivity I recently had a Twitter conversation with Roko Mijic. I believe it contains ideas that a wider philosophical/rationalist audience may find valuable, and so include here a transcript (quoted with permission). ---------------------------------------- Jes...
5a59e534-8ea3-4acb-8e8e-b77239c496b4
StampyAI/alignment-research-dataset/blogs
Blogs
confusion about alignment requirements confusion about alignment requirements -------------------------------------- for now, let's put aside the fact that we can't decide whether we're trying to achieve [sponge coordination or FAS](outlook-ai-risk-mitigation.html), and merely consider what it takes to build an alig...
133bd730-4d7b-41b8-94fd-90d3d0b9c775
trentmkelly/LessWrong-43k
LessWrong
Top YouTube channel Veritasium releases video on Sleeping Beauty Problem
07f0da03-fd7f-4089-b839-1841d39ccd4a
trentmkelly/LessWrong-43k
LessWrong
Hiding in plain sight: the questions we don't ask In scientific and philosophical research, asking the right questions is integral for expanding knowledge however in our efforts to uncover ideas we may neglect to ask certain questions especially if we think inside the box. In this post I'll determine principles of how...
2087d5c8-63cc-4ca0-abbb-c8ba5793e541
trentmkelly/LessWrong-43k
LessWrong
"Designing agent incentives to avoid reward tampering", DeepMind None
f20f8e16-b902-428d-9c3f-5297c072ada2
trentmkelly/LessWrong-43k
LessWrong
Thoughts on SB-1047 In this post, I'll discuss my current understanding of SB-1047, what I think should change about the bill, and what I think about the bill overall (with and without my suggested changes). Overall, SB-1047 seems pretty good and reasonable. However, I think my suggested changes could substantially i...
ed044cfe-47a4-42b2-a8db-81f91b8c6e3d
trentmkelly/LessWrong-43k
LessWrong
A proposed method for forecasting transformative AI In 2021, I proposed measuring progress in the perplexity of language models and extrapolating past results to determine when language models were expected to reach roughly "human-level" performance. Here, I build on that approach by introducing a more systematic and ...
93704071-be56-4226-b3e0-acf9be588e25
trentmkelly/LessWrong-43k
LessWrong
Using Obsidian if you're used to using Roam TL;DR: I used Roam for a long time and found it super helpful, now I want this same useful functionality but with the privacy of local storage only.[1] I was leaving a lot of effectiveness on the table by not doing this sooner, so hopefully by reading this, someone who wants...
a2bdbd92-e241-47c1-a819-c77a86234129
trentmkelly/LessWrong-43k
LessWrong
Theism, Wednesday, and Not Being Adopted (Disclaimer: This post is sympathetic to a certain subset of theists.  I am not myself a theist, nor have I ever been one.  I do not intend to justify all varieties of theism, nor do I intend to justify much in the way of common theistic behavior.) I'm not adopted.  You all be...
c1ace2fd-15cb-4103-9c22-0dae961096ab
trentmkelly/LessWrong-43k
LessWrong
Process Substitution Without Shell? While I still write a decent amount of shell I generally try to avoid it. It's hard for others to read, has a lot of sharp edges, tends to swallow errors, and handles the unusual situations poorly. But one thing that keeps me coming back to it is how easily I can set up trees of pro...
6f11f1fd-1d3e-4457-9278-6e7e416a3039
trentmkelly/LessWrong-43k
LessWrong
Goodhart's Law Example: Training Verifiers to Solve Math Word Problems Sharing because of how clearly this paper demonstrates the risks associated with considering too many different solutions: "At test time, we can choose to generate arbitrarily many solutions to be judged by the verifier before selecting the highes...
50370570-1b05-4c63-8dfa-3fab95092117
trentmkelly/LessWrong-43k
LessWrong
The Quantum Physics Sequence This is an inclusive guide to the series of posts on quantum mechanics that began on April 9th, 2008, including the digressions into related topics (such as the difference between Science and Bayesianism) and some of the preliminary reading. You may also be interested in one of the less i...
f7909a58-d1af-4aa8-971b-13526cdb7379
trentmkelly/LessWrong-43k
LessWrong
Music: The 21st Century Monads > The 21st Century Monads are an international musical collaboration whose songs address fundamental issues in philosophy, including specialized topics in contemporary analytic philosophy and the history of philosophy. The musical genres range from dance to folk. The songs are unique, or...
b990f392-bafe-49b9-a27f-78cd5bf2eb95
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
Pause For Thought: The AI Pause Debate *Crosspost from* [*Astral Codex Ten*](https://www.astralcodexten.com/p/pause-for-thought-the-ai-pause-debate)*.* **I.** Last month, Ben West of the Center for Effective Altruism [**hosted a debate**](https://forum.effectivealtruism.org/s/vw6tX5SyvTwMeSxJk) among long-termists, ...
c0529887-164c-426b-a071-2fb52fc2081b
trentmkelly/LessWrong-43k
LessWrong
What is your relationship with your self? Bearlamp | Lesswrong | Greaterwrong I had a very deep and intense conversation with a friend.  We talked about “that feeling of being alone” and that relationships will come and go and really, “I am all that I can guarantee that I have” at the end of the day. My conversation ...
ac9bb7e2-73a1-470b-87b8-6c11daf1a9cc
trentmkelly/LessWrong-43k
LessWrong
How inevitable was modern human civilization - data We have a sample of one modern human civilization, but there are some hints on how likely it was to happen. Major types of hints are: * Time - if something happened extremely quickly; or extremely late, it suggests how likely it was. * Independent invention - som...
ea2c2ff4-dde2-4038-b824-73a36d05b2ff
trentmkelly/LessWrong-43k
LessWrong
Evaluating Sparse Autoencoders with Board Game Models This blog post discusses a collaborative research paper on sparse autoencoders (SAEs), specifically focusing on SAE evaluations and a new training method we call p-annealing. As the first author, I primarily contributed to the evaluation portion of our work. The vi...
7fa378e6-d1e7-4ebb-9bc3-1ecbd19fa1c1
trentmkelly/LessWrong-43k
LessWrong
Psychic Powers > If the "boring view" of reality is correct, then you can never predict anything irreducible because you are reducible.  You can never get Bayesian confirmation for a hypothesis of irreducibility, because any prediction you can make is, therefore, something that could also be predicted by a reducible t...
a9b9f70b-b9e0-4975-8209-974c9a761bf5
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
Artificial intelligence career stories --- *In the future, we may post a transcript for this EA Student Summit 2020 talk, but we haven't created one yet. If you'd like to create a transcript for this talk, contact* [*Aaron Gertler*](mailto:aaron.gertler@centreforeffectivealtruism.org) *— he can help you get started.*
4d5d4f5b-5dc2-4ace-8fea-70f48c5a2bfe
StampyAI/alignment-research-dataset/special_docs
Other
Visualizing Data Using t-SNE JournalofMachineLearning Research 9(2008)2579-2605 Submitted 5/08;Revised9/08;Published 11/08 Visualizing Datausingt-SNE LaurensvanderMaaten LVDMAATEN@GMAIL.COM TiCC TilburgUniversity P.O.Box90153,5000LETilburg,TheNetherlands GeoffreyHinton HINTON@CS.TORONTO.EDU Department ofComputer Scien...
85971fa9-6432-4b97-9d72-46a3aca3a58e
trentmkelly/LessWrong-43k
LessWrong
Meetup : Effective Altruism Netherlands: cancelled Discussion article for the meetup : Effective Altruism Netherlands: cancelled WHEN: 12 April 2015 02:19:00PM (+0200) WHERE: La Place, Rijnkade 5, Utrecht Sorry, this meetup has been cancelled. Discussion article for the meetup : Effective Altruism Netherlands: ca...
3c15479f-8717-4249-9ff9-710beecf4b97
trentmkelly/LessWrong-43k
LessWrong
Alternate card types for Anki Recently I have started using Anki in a new and complimentary way. I am curious if any of you find it similarly useful and/or have other anki tips :) The basic idea is that instead of putting down a challenge/response pair for facts, we put down a challenge/response pair for ways of thin...
bfa215e5-96b1-40e0-a837-fee4532240ac
StampyAI/alignment-research-dataset/blogs
Blogs
Cultural and Memetic Hygiene Cultural and Memetic Hygiene ---------------------------- i like culture as much as i like markets; but just like how markets need [an outside force to become correctly aligned](ubi.html), culture takes extra effort to partake in in responsible and hygienic ways. you are not just influ...
3d85b7f9-c460-4962-89da-6c8eadba5e3f
trentmkelly/LessWrong-43k
LessWrong
My Central Alignment Priority (2 July 2023) Epistemic status: The further you scroll, the more-important the points are. I continue my deconfusion from my alignment timeline post. Goal: figure out which sub-problem to work on first in AI alignment. Quick Notation 1 = steering 2 = goal 1.1k no steering found 1.2k...
28660a3d-7de8-4ee4-beec-e1af0fb24592
trentmkelly/LessWrong-43k
LessWrong
linkpost: loss basin visualization it's an old video from 2018, but it has some interesting stuff in it I hadn't seen before related to loss basin visualization. I know some people were discussing that a while back, so I figured I'd link this here for those interested
30642688-755a-44b1-b282-3f6acb3be636
trentmkelly/LessWrong-43k
LessWrong
Timeless Identity Followup to:  No Individual Particles, Identity Isn't In Specific Atoms, Timeless Physics, Timeless Causality People have asked me, "What practical good does it do to discuss quantum physics or consciousness or zombies or personal identity?  I mean, what's the application for me in real life?" Befo...
71d030b1-74d3-4054-b3d9-5de42532a18d
StampyAI/alignment-research-dataset/special_docs
Other
Probabilistic machine learning and artificial intelligence ### Subjects * [Computer science](/subjects/computer-science) * [Mathematics and computing](/subjects/mathematics-and-computing) * [Neuroscience](/subjects/neuroscience) Abstract -------- How can a machine learn from experience? Probabilistic modelling...
ea8842cf-acce-43c1-baee-b37f4195943a
trentmkelly/LessWrong-43k
LessWrong
"I know I'm biased, but..." Inspired by: The 5-Second Level, Knowing About Biases Can Hurt People "I know I'm biased, but..." and its equivalents seem to be relatively common in casual conversation--I've encountered the phrase in classroom discussions, on Internet message boards, and in political arguments. In most c...
3777bc5b-faf0-4b49-b61d-eafea570ef0b
trentmkelly/LessWrong-43k
LessWrong
Consequentialists: One-Way Pattern Traps Generated during MATS 2.1. A distillation of my understanding of Eliezer-consequentialism. Thanks to Jeremy Gillen, Ben Goodman, Paul Colognese, Daniel Kokotajlo, Scott Viteri, Peter Barnett, Garrett Baker, and Olivia Jimenez for discussion and/or feedback; to Eliezer Yudkows...
cfc0e1a8-897d-4aff-80f8-accee77bef9a
trentmkelly/LessWrong-43k
LessWrong
The Stable State is Broken or: Why Everything Is Terrible, An Overview.1   It sounds like a theory which explains too much. But it's not a theory, hardly even an explanation, more a pattern that manifests itself once you start trying to seriously answer rhetorical questions about the state of the world. From many pe...
231da14b-5bd1-4c52-9169-fac5b93c72eb
trentmkelly/LessWrong-43k
LessWrong
The Problematic Third Person Perspective [Epistemic status: I now endorse this again. Michael pointed out a possibility for downside risk with losing mathematical ability, which initially made me update away from the view here. However, some experience noticing what it is like to make certain kinds of mathematical pro...
3c281cc6-2a7d-467c-9fc3-cc8ff45ab327
trentmkelly/LessWrong-43k
LessWrong
Why quantitative finance is so hard This is not financial advice. ---------------------------------------- Quantitative finance (QF) is the art of using mathematics to extract money from a securities market. A security is a fungible financial asset. Securities include stocks, bonds, futures, currencies, cryptocurren...
b20c5442-293a-45e0-9202-5b6796ca5f01
StampyAI/alignment-research-dataset/special_docs
Other
Machine Learning and Cybersecurity Machine Learning and Cybersecurity AUTHORS Micah Musser Ashton GarriottHYPE AND REALITY JUNE 2021 Center for Security and Emerging Technology 2Established in January 2019, the Center for Security and Emerging Technology (CSET) at Georgetown’s Walsh School of Foreign Service is a r...
5af8a4ae-74d9-4ead-89d8-ffa5674782ae
trentmkelly/LessWrong-43k
LessWrong
Is there a good software solution for mathematical questions? I have a programming problem that I want to optimize. To optimize it I have to calculate the minimum value that's possible for one variable given a set of equations (where some but not all variables are known at runtime). Is there any good automatized syste...
1a39e037-05b2-40b6-b34e-88972c430dce
StampyAI/alignment-research-dataset/blogs
Blogs
AlphaGo Zero and the Foom Debate [AlphaGo Zero](https://web.archive.org/web/20171020090144/https://deepmind.com/blog/alphago-zero-learning-scratch/) uses 4 TPUs, is built entirely out of neural nets with no handcrafted features, doesn’t pretrain against expert games or anything else human, reaches a superhuman level a...
ee95e773-68b3-4b84-9c37-e306dc865b19
trentmkelly/LessWrong-43k
LessWrong
How does The Circle grow, exactly? Epistemic status: still in the middle of working through a thing, off the cuff thoughts. Wears my half-baked current-generation morality on my sleave.  There's an EA song, adapted for Solstice, which presents an optimistic view of the expanding circle of concern: > CALLER: Raise a ...
f8159588-77b0-404f-b874-7aae06c90ca8
trentmkelly/LessWrong-43k
LessWrong
Might law save us from uncaring AI? Robin has claimed a few times that law is humans’ best bet for protecting ourselves from super-intelligent robots. This seemed unlikely to me, and he didn’t offer much explanation. I figured laws would protect us while AI was about as intellectually weak as us, but not if when it wa...
79090c19-bde7-45bd-a430-4bf26dbe1e7b
StampyAI/alignment-research-dataset/arxiv
Arxiv
Decision Trees for Helpdesk Advisor Graphs Bulletin of the IEEE Technical Committee on Learning Technology, Volume 18, Number 2/3, 2016 6 Abstract—We use decision trees to build a helpdesk agent reference network to facilitate the on-the-job advising of junior or less experienced staff on how to better address tele...
7cbb68dd-d843-4be0-b8a0-46662515e575
trentmkelly/LessWrong-43k
LessWrong
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 bla...
c55db673-dc94-43cd-bfeb-c0daf054b624
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Robustness to Scaling Down: More Important Than I Thought *(Edit: I added text between "...was really about reductions." and "To use the mental move of robustness...", because comments showed me I hadn't made my meaning clear enough.)* *This post is part of the work done at* [*Conjecture*](https://conjecture.dev)*.* ...
2d0bc903-61db-49c9-954e-b3b40fe5f44b
StampyAI/alignment-research-dataset/arxiv
Arxiv
Object representations as fixed points: Training iterative refinement algorithms with implicit differentiation. 1 Introduction --------------- Conventionally, neural network models implement feedforward computation, transforming the input x into the output z through a fixed series of operations corresponding to dis...
bea01362-9f29-4812-86a4-d3f140f35796
trentmkelly/LessWrong-43k
LessWrong
Covid 5/12/22: Other Priorities There is zero funding for dealing even with the current pandemic, let alone preventing the next one. The FDA not only is in no hurry to approve a vaccine for children, the new highlight is its focus on creating a dire shortage of specialty baby formula. Covid doesn’t kill children, mere...
62a3cc1c-33b9-4203-9da9-a66a32c73c85
trentmkelly/LessWrong-43k
LessWrong
How do newcomers delve deeper into the community? I have some experience managing large online communities of various sorts, and it struck me that some of the most difficult but necessary work across all of them was in spending hours brainstorming, with our leadership teams, how we might pull the 80% (or 50%, or 99%) ...
027a7041-11df-4d58-9f70-8af2fdd18dfb
trentmkelly/LessWrong-43k
LessWrong
Two Ways To Reduce Unhappiness That Comes From Distorted Views of Reality [reposted from my substack] Once a man painted a picture of a fearsome beast. The creature began haunting his thoughts and dreams, causing him great distress. The beast became so real in his mind that his fears threatened to consume him. The ab...
eb2ed284-9103-447e-9b9c-18700c99007c
StampyAI/alignment-research-dataset/special_docs
Other
Learning the Preferences of Bounded Agents Learning the Preferences of Bounded Agents Owain Evans University of OxfordAndreas Stuhlm ¨uller Stanford UniversityNoah D. Goodman Stanford University Introduction A range of work in applied machine learning, psychology, and social science involves inferring a person’s prefe...
b27d304b-3e4d-4121-b717-5517806786dc
trentmkelly/LessWrong-43k
LessWrong
Understanding Trust: Overview Presentations Since mid-January, I've been working with four students - Hanna Gabor, Roman Malov, Norman Hsia, Paul Rapoport - on the research agenda I articulated last year. The AI Safety Camp project is wrapping up, but most of us plan to continue working on related things and meeting e...
6201e7bb-a2e8-4497-ad0a-7898042dd489
trentmkelly/LessWrong-43k
LessWrong
Wyclif's Dust: the missing chapter As substack subscribers probably know, I have finally published Wyclif’s Dust the book, and I wrote an informal intro about it last week. Here I’ll write about what isn’t in the book — the places where I wish I knew more, or where the argument could be strengthened. Partly I am fess...
47856e04-d060-4f0a-9eb2-a4c69e949769
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
General alignment plus human values, or alignment via human values? *Thanks to Rebecca Gorman for discussions that lead to these insights.* How can you get a superintelligent AI aligned with human values? There are two pathways that I often hear discussed. The first sees a general alignment problem - how to get a po...
dfb8c91b-d6e3-4c07-9bbd-454f8dd0da0f
trentmkelly/LessWrong-43k
LessWrong
Fables grow around missed natural experiments So I read Think like a Freak, and then glimpsed through a well-intentioned collection of "Reading Comprehension Tests for Schoolchildren" (in Ukrainian), and I was appalled at how easily the latter book dismissed simple observation of natural experiments that it makes a to...
94589f75-3be3-476a-a586-16653a40ed63
trentmkelly/LessWrong-43k
LessWrong
The dangers of zero and one Eliezer wrote a post warning against unrealistically confident estimates, in which he argued that you can't be 99.99% sure that 53 is prime. Chris Hallquist replied with a post arguing that you can. That particular case is tricky. There have been many independent calculations of the fir...
32a3c530-6e68-4733-84ff-faf6d7cc2765
StampyAI/alignment-research-dataset/special_docs
Other
Globally inaccurate stereotypes can result from locally adaptive exploration. https://doi.org/10.1177/09567976211045929Psychological Science 1 –14 © The Author(s) 2022 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/09567976211045929 www.psychologicalscience.org/PSASSOCIATION FOR PSYCHOLOGICA...
41d6b646-7b73-4a23-abbc-506691d4d7ac
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Clarifying mesa-optimization Produced as part of the [SERI ML Alignment Theory Scholars Program](https://www.serimats.org/) - Winter 2022 Cohort. *Thanks to Jérémy Scheurer, Nicholas Dupuis and Evan Hubinger for feedback and discussion* When people talk about mesa-optimization, they sometimes say things like “we’re ...
613d4a8a-a8ff-4bc9-9704-36dfb34f8561
trentmkelly/LessWrong-43k
LessWrong
Defect or Cooperate 3. Defect or Cooperate Summary of entire Series: An alternative approach to designing Friendly Artificial Intelligence computer systems. Summary of this Article: If the risks from being punished for not cooperating are high enough, then even for some types of Paperclip Maximiser that don't care...
c271ec15-312d-4898-8ab4-21169ec43c37
StampyAI/alignment-research-dataset/blogs
Blogs
plausible vs likely plausible vs likely ------------------- when talking about important (utility-monsterly so) problems like [AI risks](say-ai-risk-mitigation-not-alignment.html), it is easy for two people to behave the same whether one believes the event is merely plausible (<50%, sometimes <10%) or likely (>50%, ...
945900e2-b2f7-4972-9464-d9b2a71e79a0
trentmkelly/LessWrong-43k
LessWrong
Leveling Up Or Leveling Off? Understanding The Science Behind Skill Plateaus This post is crossposted from my blog. If you liked this post, subscribe to Lynette's blog to read more -- I only crosspost about half my content to other platforms. If you’re going into surgery, you want the youngest operating surgeon avail...
4834e42f-f76a-4261-800d-65a2aad824b5
trentmkelly/LessWrong-43k
LessWrong
Doing Your Good Deed For The Day A few months ago, researchers at the University of Toronto published a very interesting study on moral behavior. The one sentence summary of the most interesting part is that people who did one good deed were less likely to do another good deed in the near future. They had, quite liter...
3cbad98a-081f-4639-bd4b-bbdf74d5bf45
StampyAI/alignment-research-dataset/blogs
Blogs
Index of articles about hardware ### Hardware in terms of computing capacity (FLOPS and MIPS) [Brain performance in FLOPS](http://aiimpacts.org/brain-performance-in-flops/) [2019 recent trends in GPU price per FLOPS](https://aiimpacts.org/2019-recent-trends-in-gpu-price-per-flops/) [Electrical efficiency of comp...
e1c40b65-243a-45f7-a699-ec03a43d1327
trentmkelly/LessWrong-43k
LessWrong
Remove Intercom? I posted a question there (about where to find the RSS feeds) and nobody answered. It looks like nobody else has posted there either. So apparently this widget is no longer used, and questions should be asked in some other way?
6827b2a4-43dd-4ba9-bf6f-d3e8f718ec87
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Conceptual problems with utility functions *[Epistemic status: strong intuitions that I have had for a long time, but not yet compared with other people's intuitions, and I am not sure how well I can convey my intuitions.]* So, a lot of agent foundations research seems to be built on the premise that agenthood is abo...
cf39098b-cf73-4e5d-ba5b-4200757d1f89
trentmkelly/LessWrong-43k
LessWrong
Probably good projects for the AI safety ecosystem At EAGxBerkeley 2022, I was asked several times what new projects might benefit the AI safety and longtermist research ecosystem. I think that several existing useful-according-to-me projects (e.g., SERI MATS, REMIX, CAIS, etc.) could urgently absorb strong management...
0157ecfa-1a3e-47b4-ae72-a3ee40d840da
trentmkelly/LessWrong-43k
LessWrong
Alice and Bob is debating on a technique. Alice says Bob should try it before denying it. Is it a fallacy or something similar? Context Imagine a conversation: * [Bob posts a problem]  * Alice: You should use technique T₁. It especially suites this kind of problem * Bob: In my understanding this technique is only ...
4b4aa990-4482-43a3-ab99-5ed03a2d9a4e
trentmkelly/LessWrong-43k
LessWrong
What is the threshold for "Hide Low Karma"? I noticed this new option. Just curious about how exactly it works.
87bcd488-5b70-46c2-9552-8e3c8ca423e6
trentmkelly/LessWrong-43k
LessWrong
Justifying Induction Related to: Where Recursive Justification Hits Bottom, Priors as Mathematical Objects, Probability is Subjectively Objective Follow up to: A Proof of Occam's Razor In my post on Occam’s Razor, I showed that a certain weak form of the Razor follows necessarily from standard mathematics and probab...
0ae1e6e8-1aec-4178-bbe8-46cf3967568c
LDJnr/LessWrong-Amplify-Instruct
LessWrong
"An excellent habit that I've noticed among professional players of the game Magic: The Gathering is asking the question "how do I lose?" - a sort of strategic looking into the dark. Imagine this situation: you have an army ready to destroy your opponent in two turns. Your opponent has no creatures under his command. V...
d5160bdc-0a2d-4e74-af43-7bc9fd665da6
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Human-AI Interaction The importance of feedback -------------------------- Consider trying to program a self-driving car to drive from San Francisco to Los Angeles -- *with no sensors* that allow it to gather information as it is driving. This is possible in principle. If you can predict the exact weather conditions,...
c9d2dd34-d951-481d-8012-4f168be8e245
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Has anyone increased their AGI timelines? The question:  Within the last 5-10 years, is there is any person or group that has openly increased their AGI timelines? Ideally, they would have at least two different estimates (years apart?), with the most recent estimate showing that they think AGI is further into the f...
90dee05a-2bc7-48a9-8ca8-4a8c936d7f41
trentmkelly/LessWrong-43k
LessWrong
Another Mechanism for the Placebo Effect? The placebo effect (benefit in groups receiving fake pills) and nocebo effect (detriment in those same groups) have frequently been the bane of medical research. They are usually explained in terms of psychology: because people receiving placebos believe they have been treated...
84712e87-cce3-4985-b81b-74baf1746e50
trentmkelly/LessWrong-43k
LessWrong
Lesswrong UK planning thread A few of us got together in the pub after the friendly AI meet and agreed we should have a meetup for those of us familiar with lesswrong/bostrom etc. This is a post for discussion of when/where. Venue: London still seems the nexus. I might be able to be convinced to go to Oxford. A sta...
4d9e20b1-95f1-46dc-8874-7eecd566448a
StampyAI/alignment-research-dataset/arbital
Arbital
Dihedral group The dihedral group $D_{2n}$ is the group of symmetries of the $n$-vertex [https://arbital.com/p/-regular_polygon](https://arbital.com/p/-regular_polygon). # Presentation The dihedral groups have very simple [presentations](https://arbital.com/p/group_presentation): $$D_{2n} \cong \langle a, b \mid a^n,...
bf5d4d74-7ec1-412f-8aaa-46e1668ba38e
trentmkelly/LessWrong-43k
LessWrong
Quantitative cruxes in Alignment Summary: Sometimes researchers talk past each other because their intuitions disagree on underlying quantitative variables. Getting better estimates of these variables is action-guiding. This is an epistemological study to bring them to the surface, as well as discover which sources of...
6ebf1d7d-a3d3-4e63-94e3-6cfd2daf1740
trentmkelly/LessWrong-43k
LessWrong
Existential Angst Factory Followup to:  The Moral Void A widespread excuse for avoiding rationality is the widespread belief that it is "rational" to believe life is meaningless, and thus suffer existential angst.  This is one of the secondary reasons why it is worth discussing the nature of morality.  But it's also ...
a717c7d1-1796-4041-81b2-ff62a06426ee
trentmkelly/LessWrong-43k
LessWrong
Antagonizing Opioid Receptors for (Prevention of) Fun and Profit Related to: Ugh Fields, Are Wireheads Happy? In his post Ugh Fields, Roko discussed "temporal difference learning", the process by which the brain propagates positive or negative feedback to the closest cause it can find for the feedback. For example, i...
6c1d11f3-34cc-40c0-817f-c89db072fae4
trentmkelly/LessWrong-43k
LessWrong
[LINK] Daniel Pink talks about Motivation Little over a week ago my work watched this video for a "self-improvement" seminar. I hadn't seen this linked anywhere on LW yet, and thought it might be relevant, given lukeprogs' article on motivation.
a073f8bf-b036-490d-bf5e-7a1f67c9777a
trentmkelly/LessWrong-43k
LessWrong
Unexploitable search: blocking malicious use of free parameters Summary: We have previously argued that scalable oversight methods can be used to provide guarantees on low-stakes safety – settings where individual failures are non-catastrophic. However, if your reward function (e.g. honesty) is compatible with many po...
a2f07763-de21-4164-b688-fd1c19cf6183
trentmkelly/LessWrong-43k
LessWrong
Does anyone worry about A.I. forums like this where they reinforce each other’s biases/ are led by big tech? I came across this programme for A.I. https://inspired-minds.co.uk/inspired-ai-events/ mostly being run by tech companies and their employees and it worries me that they are making A.I. and the fast development...
1ef77814-d1fd-48ea-a70b-1b17dc914dbf
trentmkelly/LessWrong-43k
LessWrong
Self Explaining Neural Networks, the interpretability technique no one seems to be talking about. Disclaimer: I'm very new to alignment as a whole. I wouldn't be surprised if this turned out to be a nothing burger.  This is the coolest paper I've seen in a while, yet I've never heard of the technique. It's not mentio...
3a4b7ff7-9dc0-4bad-939d-1a7c100ce6c5
StampyAI/alignment-research-dataset/blogs
Blogs
the persistent data structure argument against linear consciousness the persistent data structure argument against linear consciousness ------------------------------------------------------------------- people have the belief that they live a continuous, linear stream of consciousness (whatever that means). i've ...
628422a2-3cde-4773-afd7-2fbf9b34852a
trentmkelly/LessWrong-43k
LessWrong
Problems and Solutions in Infinite Ethics (Crossposted from the EA forum.) Summary: The universe may very well be infinite, and hence contain an infinite amount of happiness and sadness. This causes several problems for altruists; for example: we can plausibly only affect a finite subset of the universe, and an infin...