id stringlengths 36 36 | source stringclasses 15
values | formatted_source stringclasses 13
values | text stringlengths 2 7.55M |
|---|---|---|---|
47aa1325-c05a-4481-928a-8319bbbda9b4 | trentmkelly/LessWrong-43k | LessWrong | obvious epiphanies
In her post Closer to Fine, Sarah Constantin points at a something I’m calling an obvious epiphany.
> You can be intellectually aware that a thing is true, and still have strong negative feelings that contradict it … There’s a big difference between “please tell me I’m okay, for the umpteenth time... |
73c87cd4-38b9-4e3e-87a4-f0b18b9b29b1 | StampyAI/alignment-research-dataset/arbital | Arbital | Logarithm: Examples
$\log_{10}(100)=2.$ $\log_2(4)=2.$ $\log_2(3)\approx 1.58.$ (TODO) |
b2c1112a-5773-46df-b096-cce736944775 | trentmkelly/LessWrong-43k | LessWrong | rabbit (a new AI company) and Large Action Model (LAM)
Just sharing it here, curious of what the LW community would think of rabbit's neuro-symbolic approach and their new device "R1"?
This is one of the two interactive videos in their research page, their training involved observing realtime data as I understood it... |
2f929c96-d376-4db1-87a9-ea39a5aaca27 | trentmkelly/LessWrong-43k | LessWrong | Mini-conference "Near-term AI safety"
TL;DR: The event will be in Moscow, Russia, and near-term risks of AI will be discussed. The main language will be Russian, but Jonatan Yan will speak in English from HK. English presentations will be uploaded later on the FB page of the group "Near-term AI safety." Speakers: S. S... |
4eddb850-a345-449e-8ed7-6742963ce835 | trentmkelly/LessWrong-43k | LessWrong | Aging research and population ethics
This is the second of a series of posts in which I'm trying to build a framework to evaluate aging research. Previous post: A general framework for evaluating aging research. Part 1: reasoning with Longevity Escape Velocity.
Summary
The first part of this post will explore the po... |
624faad7-8c8d-4e22-a4d1-e59a24e84812 | trentmkelly/LessWrong-43k | LessWrong | [link] Psychologists strike a blow for reproducibility
Link
> A large international group set up to test the reliability of psychology experiments has successfully reproduced the results of 10 out of 13 past experiments. The consortium also found that two effects could not be reproduced.
> To tackle this 'replicabil... |
5257af79-72c5-4dba-b4b7-bda0512fcd14 | trentmkelly/LessWrong-43k | LessWrong | Reflections on the state of the race to superintelligence, February 2025
My model of the situation is that some time last year, the frontier paradigm moved from "scaling up large language models" to "scaling up chain-of-thought models". People are still inventing new architectures, e.g. Google's Titans, or Lecun's ene... |
47b1d9db-4627-4c50-b0d8-cb13ff7b7d56 | trentmkelly/LessWrong-43k | LessWrong | I Can Tolerate Anything Except The Outgroup
[Content warning: Politics, religion, social justice, spoilers for “The Secret of Father Brown”. This isn’t especially original to me and I don’t claim anything more than to be explaining and rewording things I have heard from a bunch of other people. Unapologetically Americ... |
3ca9cc4b-0f78-4bfe-913b-d6e5e044cc67 | trentmkelly/LessWrong-43k | LessWrong | History of counting to three?
Counting to three is a common way to indicate to kids that you're being serious and they need to do what you're saying, and I find it helpful with my kids. It's associated with Thomas Phelan's 1995 book 123 Magic, but it's older than that:
* Tom Chapin's 1988 song Nick Of Time includes ... |
56c4908f-dc9b-4164-8548-55724b670df0 | trentmkelly/LessWrong-43k | LessWrong | Reinforcement Learning: Essential Step Towards AGI or Irrelevant?
A friend of mine thinks that RL is a dead end: LLMs are much better at problem solving, exploration, and exploitation than any RL algorithm. And I agree that LLMs are better than RL on RL's tasks: companies even have LLMs controlling robots nowadays.
T... |
f0122031-5d82-4c38-9209-c1cf40b281b2 | trentmkelly/LessWrong-43k | LessWrong | Attending LessOnline
The internet has massively decreased the cost and effort of publishing. While "random people writing on the internet" doesn't sound like a good source of information, this would fit the majority of my favorite writers. Except most aren't "random people" anymore: by writing publicly over time reade... |
68a0095e-bdd5-4e7b-bfaf-a366edaeeb8d | trentmkelly/LessWrong-43k | LessWrong | Austin Chen on Winning, Risk-Taking, and FTX
Timothy and I have recorded a new episode of our podcast with Austin Chen of Manifund (formerly of Manifold, behind the scenes at Manifest).
The start of the conversation was contrasting each of our North Stars- Winning (Austin), Truthseeking (me), and Flow (Timothy), but ... |
862fea72-c520-4926-a563-0363d276c609 | trentmkelly/LessWrong-43k | LessWrong | Compilation of currently existing project ideas to significantly impact the world
One of the problems the LW, EA, CFAR X-risk community has been faced with recently discussed on Slate Star Codex is the absorption of people interested in researching, volunteering, helping, participating in the community. A problem wort... |
964ae577-5bd7-41a3-91b7-87583ef51a3c | trentmkelly/LessWrong-43k | LessWrong | NY State Has a New Frontier Model Bill (+quick takes)
> This morning, New York State Assemblyman Alex Bores introduced the Responsible AI Safety and Education Act. I’d like to think some of my previous advocacy was helpful here, but I know for a fact that I’m not the only one who supports legislation like this that on... |
9deec14b-0615-4f31-92be-5348d842ec41 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Why I'm Optimistic About Near-Term AI Risk
I'm not worried about AI posing an existential risk in the next 10-20 years. Recent developments in AI capabilities actually make me feel *more* optimistic about this. The fact that relatively simple models can perform a wide array of tasks suggests that we can build [satisfa... |
58c42aaf-d915-4879-89c7-180f7855ccb9 | trentmkelly/LessWrong-43k | LessWrong | A more systematic case for inner misalignment
This post builds on my previous post making the case that squiggle-maximizers are plausible. The argument I presented was a deliberately simplified one, though, and glossed over several possible issues. In this post I'll raise and explore three broad objections. (Before lo... |
06b5d675-4d3b-45b3-ae91-45f0b3092afd | trentmkelly/LessWrong-43k | LessWrong | Non-Consequentialist Cooperation?
This is a very rough intuition pump for possible alternatives to value learning.
In broad strokes, the goal of (ambitious) value learning is to define and implement a notion of cooperation (or helpfulness) in terms of two activities: (1) figuring out what humans value, (2) working to... |
55f545ec-30c9-4a57-874b-1e4216719ede | trentmkelly/LessWrong-43k | LessWrong | Some cruxes on impactful alternatives to AI policy work
Ben Pace and I (Richard Ngo) recently did a public double crux at the Berkeley REACH on how valuable it is for people to go into AI policy and strategy work: I was optimistic and Ben was pessimistic. During the actual event, we didn't come anywhere near to findin... |
60081605-8e20-46e0-a911-e7198a12d4ec | trentmkelly/LessWrong-43k | LessWrong | Writing Down Conversations
Epistemic Status: Didn't think through exactly how I worded things.
tldr: When you have insightful conversation, write it down and share it so people can build on it (instead of just sharing in person). Most of humanity's power comes from being able to build complex thoughts out of other th... |
f2117904-00cc-4817-a679-10f3642fd2df | trentmkelly/LessWrong-43k | LessWrong | Meetup : Moscow meetup: science research issues, global risks, fallacymania
Discussion article for the meetup : Moscow meetup: science research issues, global risks, fallacymania
WHEN: 01 November 2015 02:00:00PM (+0300)
WHERE: Москва, ул. Большая Дорогомиловская, д.5к2
I haven't announced our meetups here on lessw... |
4f1b0307-4995-4a7c-a1e2-df1b2d9c3575 | trentmkelly/LessWrong-43k | LessWrong | Embedded Agency via Abstraction
Claim: problems of agents embedded in their environment mostly reduce to problems of abstraction. Solve abstraction, and solutions to embedded agency problems will probably just drop out naturally.
The goal of this post is to explain the intuition underlying that claim. The point is no... |
4f8e282a-1c5f-4ce7-b917-42523c10574e | trentmkelly/LessWrong-43k | LessWrong | The Preference Utilitarian’s Time Inconsistency Problem
In May of 2007, DanielLC asked at Felicifa, an “online utilitarianism community”:
> If preference utilitarianism is about making peoples’ preferences and the universe coincide, wouldn't it be much easier to change peoples’ preferences than the universe?
Indeed,... |
9eae1fa6-f59f-40bc-b468-e1928533ed64 | trentmkelly/LessWrong-43k | LessWrong | Meetup : Berkeley: The Twelve Virtues
Discussion article for the meetup : Berkeley: The Twelve Virtues
WHEN: 12 February 2014 07:00:00PM (-0800)
WHERE: 2030 Addison, 3rd floor, Berkeley, CA 94703
Tonight's meetup will feature a critical discussion of the Twelve Virtues of Rationality, an essay that is cited in the ... |
c5b8fc5e-3b3f-4bff-ad1c-a94b059d3404 | trentmkelly/LessWrong-43k | LessWrong | [Linkpost] Remarks on the Convergence in Distribution of Random Neural Networks to Gaussian Processes in the Infinite Width Limit
The linked note is something I "noticed" while going through different versions of this result in the literature. I think that this sort of mathematical work on neural networks is worthwhi... |
e029fc2d-913f-498e-88f1-ea6a954803fa | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Logical uncertainty reading list
This was originally part of a post I wrote on logical uncertainty, but it turned out to be post-sized itself, so I'm splitting it off.
Daniel Garber's article [Old Evidence and Logical Omniscience in Bayesian Confirmation Theory](http://fitelson.org/probability/garber.pdf). Wonderful... |
69cd3dd1-ea46-463f-836a-ca4c89d8dd81 | StampyAI/alignment-research-dataset/eaforum | Effective Altruism Forum | I'm interviewing Nova Das Sarma about AI safety and information security. What shouId I ask her?
Next week for [The 80,000 Hours Podcast](https://80000hours.org/podcast/) I'll be interviewing Nova Das Sarma.
She works to improve computer and information security at [Anthropic](https://www.anthropic.com/news/announce... |
e0a30e89-ef63-4322-8142-64cff340ddea | trentmkelly/LessWrong-43k | LessWrong | Character alignment II
In character alignment I lay out a view on alignment that focuses on behavioral priors aka character traits as opposed to values, goals or behaviors. The idea being that values and behaviors are complex, high dimensional, and not necessarily consistent and therefore hard to pin down. Behavioral ... |
f1af53dd-424d-46ee-8a92-cd2532d1f102 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Decision Theory Paradox: PD with Three Implies Chaos?
**Prerequisites:** Familiarity with decision theories (in particular, [Eliezer's Timeless Decision Theory](/lw/15z/ingredients_of_timeless_decision_theory/)) and of course the [Prisoner's Dilemma](http://wiki.lesswrong.com/wiki/Prisoner%27s_dilemma).
***Summary:*... |
45e0bc52-f165-4cc1-9201-158004debb22 | trentmkelly/LessWrong-43k | LessWrong | Colonialism in space: Does a collection of minds have exactly two attractors?
Depending on the estimates of parameters, the Drake equation produces drastically different amounts of contactable civilisations in the Milky Way. Some estimates imply that the reason is the extreme rarity of life, while others suggest that ... |
2267bb6d-b006-4a02-98df-0fc7393c7813 | trentmkelly/LessWrong-43k | LessWrong | My Thoughts on the ML Safety Course
This summary was written as part of Refine. The ML Safety Course is created by Dan Hendrycks at the Center for AI Safety. Thanks to Adam Shimi and Thomas Woodside for helpful feedback.
Overview
Background
I recently completed the ML Safety Course by watching the videos and brow... |
5d569485-7f3c-4928-aa6e-b5124050003c | trentmkelly/LessWrong-43k | LessWrong | A TAI which kills all humans might also doom itself
or TAI Murder-Suicide
> Losing a conflict with a high-powered cognitive system looks at least as deadly as "everybody on the face of the Earth suddenly falls over dead within the same second."
> - Eliezer Yudkowsky, AGI Ruin: A List of Lethalities (2022)
This is a p... |
5fcc3516-00b8-48ab-a1b1-401f98cc9ef8 | trentmkelly/LessWrong-43k | LessWrong | [LINK] Levels of Ethics
I've resumed blogging For Real This Time™, starting with an introductory overview of the distinction between metaethics and normative ethics.
Should I cross-post it to LessWrong? Should I link or cross-post future blogging about metaethics and other LW-relevant topics? Is it rubbish? Inquiring... |
46b18158-1c94-4f51-9ab1-60379dbb6e62 | trentmkelly/LessWrong-43k | LessWrong | The Big Picture Of Alignment (Talk Part 2)
I recently gave a two-part talk on the big picture of alignment, as I see it. The talk is not-at-all polished, but contains a lot of stuff for which I don't currently know of any good writeup. Linkpost for the first part is here; this linkpost is for the second part.
Compar... |
3944bee4-18e0-4bde-baf0-b20f0d272c97 | trentmkelly/LessWrong-43k | LessWrong | David Chalmers on LessWrong and the rationalist community (from his reddit AMA)
|
6e2423e8-321d-40be-a991-1760dee15201 | awestover/filtering-for-misalignment | Redwood Research: Alek's Filtering Results | id: post1327
https://twitter.com/antimatter15/status/1602469101854564352 Currently, large language models ( ChatGPT , Constitutional AI ) are trained to refuse to follow user requests that are considered inappropriate or harmful. This can be done by training on example strings of the form “User: inappropriate request ... |
f2cfda9b-64f5-4e9b-9fdb-c6e21e065399 | trentmkelly/LessWrong-43k | LessWrong | LW authors: How many clusters of norms do you (personally) want?
note: this post is about gathering data for "what might the archipelago model look like if implemented on LW", which is s separate question from "should the archipelago model be implemented on LW?"
Periodically, I've argued for applying Archipelago-styl... |
eb4e5680-085e-459d-ae7c-ef1fa03c558a | trentmkelly/LessWrong-43k | LessWrong | AI Safety University Organizing: Early Takeaways from Thirteen Groups
TL;DR
* The number of AI Safety groups worldwide has grown rapidly, but there are many important uncertainties around strategy and tactics.
* This guide summarizes the opinions of the organizers of 13 AI safety university groups on 27 strategy-rel... |
a77b5693-c0d0-4f70-b107-07175b048237 | trentmkelly/LessWrong-43k | LessWrong | Giving Now Currently Seems to Beat Giving Later
> Abstract: There is a debate between either donating now or donating later (investing, realizing the returns of one's investment, and donating in a lumpsum upon death). While donating later may be appropriate in some circumstances, right now we have the ability to dona... |
f5fb7177-7156-46de-8512-5cc1d4c5fbd6 | trentmkelly/LessWrong-43k | LessWrong | Evaluating hidden directions on the utility dataset: classification, steering and removal
Produced as part of the SERI ML Alignment Theory Scholars Program - Summer 2023 Cohort, under the mentorship of Dan Hendrycks
We demonstrate different techniques for finding concept directions in hidden layers of an LLM. We prop... |
9b45cab2-d959-4837-8fba-12ebdd61eaf1 | trentmkelly/LessWrong-43k | LessWrong | "The Heart of Gaming is the Power Fantasy", and Cohabitive Games
I quite liked @mako yass' Peacewagers/Cohabitive Games so Far. But I found myself wanting to link to this essay by Jeff Vogel, an indie game developer and blogger. I think it says something important about the constraints of how to develop Cohabitive gam... |
310dcad7-21fd-4f90-9291-8272db8649dc | trentmkelly/LessWrong-43k | LessWrong | Your abstraction isn't wrong, it's just really bad
Epistemtic status: pretty sure
For a programming language to qualify as such, the only thing it needs is Turing Completeness. That means, at least in principle, it can be used to solve any (solvable) computational problem.
There are features besides Turing Completen... |
f6bfc84b-b2f8-4404-8def-47107a8d23c2 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | A Quick List of Some Problems in AI Alignment As A Field
1. MIRI as central point of failure for... a few things...
----------------------------------------------------------
For the past decade or more, if you read an article saying "AI safety is important", and you thought, "I need to donate or apply to work somew... |
c9919e47-6bcf-4200-95f6-a26d4056079d | trentmkelly/LessWrong-43k | LessWrong | Suggest short Sequence readings for my college stat class
I'm teaching one section of an intro-to-stat-for-non-math-folks class this spring as an adjunct. I'm planning to supplement the dry, fire-hose-y textbook with some outside readings (data journalism written by me, where students can reconstruct the analysis; exc... |
ed4f4ac8-e428-42d4-845a-2fee5e812d3e | StampyAI/alignment-research-dataset/blogs | Blogs | Your Policy Regulariser is Secretly an Adversary
[](/?source=post_page-----14684c743d45--------------------------------)Listen
Share
*By Rob Brekelmans, Tim Genewein, Jordi Grau-Moya, Grégoire Delétang, Markus Kunes... |
aff1f1f1-25f9-4686-b738-70e2a8118b79 | StampyAI/alignment-research-dataset/special_docs | Other | AI governance research agenda
AI Governance:
A Research Agenda
Allan Dafoe
Centre for the Governance of AI
Future of Humanity Institute
University of Oxford
First draft July 2017
v1.0 August 27 2018
Visit fhi.ox.ac.uk/govaiagenda to check for the most recent version of this paper. ... |
ba667127-5221-428d-8d1a-4bcf26047a18 | trentmkelly/LessWrong-43k | LessWrong | FAI and the Information Theory of Pleasure
Previously, I talked about the mystery of pain and pleasure, and how little we know about what sorts of arrangements of particles intrinsically produce them.
Up now: should FAI researchers care about this topic? Is research into the information theory of pain and pleasure... |
9cc4d40f-6cad-4479-8a9c-475954b0e5c8 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | 100 Dinners And A Workshop: Information Preservation And Goals
*Produced on a grant from the LTFF. Thanks to Justis for proofreading and Paul Colognese, David Udell, Alex Altair, Nicky Pochinkov, and Jessica Rumbelow for discussion and feedback.*
I would like to reason about the emergence and impact of an AGI on ... |
43a169cf-6b58-42d8-88b7-eb81953036b0 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | The Indexing Problem
*Meta:* [*this project*](https://www.lesswrong.com/s/ehnG4mseKF6xALmQy) *is wrapping up for now. This is the first of probably several posts dumping my thought-state as of this week.*
Suppose we have a simple causal model of a system:
![](https://docs.google.com/drawings/u/1/d/sq3gbnSq7Z43VXgkrj... |
870f9f93-1db8-47c4-a73f-52f3c0d41a93 | trentmkelly/LessWrong-43k | LessWrong | How Should We Measure Intelligence Models: Why Use Frequency of Elemental Information Operations
Introduction
When I’m in the lab staring at the buzzing GPU servers, I often wonder: am I smarter, or are these machines smarter? At least when it comes to rendering tasks, it’s clear that I can’t compete with these machin... |
77d4135d-4cfa-43fc-b4c5-71dbcead0245 | trentmkelly/LessWrong-43k | LessWrong | Thomas C. Schelling's "Strategy of Conflict"
It's an old book, I know, and one that many of us have already read. But if you haven't, you should.
If there's anything in the world that deserves to be called a martial art of rationality, this book is the closest approximation yet. Forget rationalist Judo: this is ratio... |
6d64e5fc-980b-4aaf-94bc-834d1b3ae0ad | trentmkelly/LessWrong-43k | LessWrong | Most people should probably feel safe most of the time
There is an idea that I’ve sometimes heard around rationalist and EA circles, that goes something like “you shouldn’t ever feel safe, because nobody is actually ever safe”. I think there are at least two major variations of this:
1. You shouldn’t ever feel safe,... |
ff0ee35a-82f0-4cba-81c2-e8f484785bf8 | trentmkelly/LessWrong-43k | LessWrong | Improving Group Decision Making
This was a talk at Effective Altruism Global (London 2019) by Mahendra Prasad. It's a non-technical introduction to the ideas in his working paper I previously posted. It also covers some additional ground. For instance, from Republican polling data, we can see the difference voting met... |
594b1d21-8596-4d0e-91b3-7878e54459f8 | trentmkelly/LessWrong-43k | LessWrong | Why Not Just Outsource Alignment Research To An AI?
Warmup: The Expert
If you haven’t seen “The Expert” before, I recommend it as a warmup for this post:
> The Client: “We need you to draw seven red lines, all strictly perpendicular. Some with green ink, some with transparent. Can you do that?”
>
> (... a minute of... |
d0ffaeed-8fd2-4dde-a236-40075506d9e9 | StampyAI/alignment-research-dataset/arxiv | Arxiv | Fixing Weight Decay Regularization in Adam
1 Introduction
---------------
Adaptive gradient methods, such as AdaGrad (Duchi et al., [2011](#bib.bib4)), RMSProp (Tieleman & Hinton, [2012](#bib.bib18)), and Adam (Kingma & Ba, [2014](#bib.bib12)) have become a default method of choice for training feed-forward and rec... |
b720a00f-68b9-4bd9-9c00-bf2921ec4493 | trentmkelly/LessWrong-43k | LessWrong | Bandwidth Rules Everything Around Me: Oliver Habryka on OpenPhil and GoodVentures
In this episode of our podcast, Timothy Telleen-Lawton and I talk to Oliver Habryka of Lightcone Infrastructure about his thoughts on the Open Philanthropy Project, which he believes has become stifled by the PR demands of its primary fu... |
f5c70527-918c-430a-9696-00bd7f23c9ba | trentmkelly/LessWrong-43k | LessWrong | Symbiotic self-alignment of AIs.
Artificial intelligence (AI) is one of the most powerful and transformative technologies of our time. It has the potential to enhance human capabilities, solve complex problems, and create new opportunities for innovation and progress. However, it also poses significant challenges and ... |
1a892caf-a23a-43e9-ae58-54f45e8b8007 | trentmkelly/LessWrong-43k | LessWrong | What is Bayesianism?
This article is an attempt to summarize basic material, and thus probably won't have anything new for the hard core posting crowd. It'd be interesting to know whether you think there's anything essential I missed, though.
You've probably seen the word 'Bayesian' used a lot on this site, but may b... |
20379ff2-83bf-4d8a-b4a3-754f50b6b34a | StampyAI/alignment-research-dataset/aisafety.info | AI Safety Info | Wouldn't it be a good thing for humanity to die out?
In the words of [Nate Soares](https://mindingourway.com/a-torch-in-darkness/):
I don’t expect humanity to survive much longer.
Often, when someone learns this, they say:
"Eh, I think that would be all right."
So allow me to make this very clear: it would not be "... |
ccce1a11-6605-4c8a-a815-620db4f734cf | trentmkelly/LessWrong-43k | LessWrong | Would this be Progress in Solving Embedded Agency?
Would it be progress if one could figure out how to construct an embedded system that can have a complete model of a highly compressible world, such that the system can correctly generate a plan that when executed would put the world into a particular target state (mo... |
67c8cedf-f28d-408d-8ebf-d85d425d744a | StampyAI/alignment-research-dataset/arxiv | Arxiv | Was ist eine Professur fuer Kuenstliche Intelligenz?
Was ist eine Professur für Künstliche Intelligenz? Prof. Dr. Kristian Kerstingac, Prof. Jan Peters, PhD. ac, Prof. Rothkopf, PhD.bc kersting@cs.tu-darmstadt.de, peters@ias.informatik.tu-darmstadt.de, rothkopf@psychologie.tu-darmstadt.de aFachbereich Informatik, b... |
0df1cf34-60fe-40cd-9abf-32dd2d5b32b1 | LDJnr/LessWrong-Amplify-Instruct | LessWrong | "This post is prompted by the multitude of posts and comments here using quantum this and that in an argument (quantum dice, quantum immortality, quantum many worlds...). But how does one know if they understand the concept they use? In school a student would have to write a test and get graded. It strikes me as a reas... |
797a4db7-24d5-4b7e-813c-d227eb42137e | trentmkelly/LessWrong-43k | LessWrong | Superintelligence and physical law
It's been a few years since I read http://lesswrong.com/lw/qj/einsteins_speed/ and the rest of the quantum physics sequence, but I recently learned about the company Nutonian, http://www.nutonian.com/. Basically it's a narrow AI system that looks at unstructured data and tries out bi... |
52dc3d0d-5bc1-4ed6-ac05-bdb46d8de0b8 | StampyAI/alignment-research-dataset/special_docs | Other | Scaling Out-of-Distribution Detection for Real-World Settings.
Scaling Out-of-Distribution Detection for Real-World Settings
Dan Hendrycks*Steven Basart*Mantas Mazeika Andy Zou Joe Kwon Mohammadreza Mostajabi
Jacob Steinhardt Dawn Song
Abstract
Detecting out-of-distribution examples is im-
portant for safety-critical ... |
bbb02071-9642-4d4d-9ffe-053c00fcde79 | trentmkelly/LessWrong-43k | LessWrong | Meetup : Washington, D.C.: What If Scenarios
Discussion article for the meetup : Washington, D.C.: What If Scenarios
WHEN: 01 March 2015 03:00:00PM (-0500)
WHERE: Reynolds Center
We will be meeting in the Kogod Courtyard of the Donald W. Reynolds Center for American Art and Portraiture (8th and F Sts or 8th and G S... |
eb4814eb-44db-4168-a5e3-d37ead0b5eda | trentmkelly/LessWrong-43k | LessWrong | College Selection Advice for Technical Alignment
Hi all, I'm a highschool senior trying to make some college-related decisions, and I'd like to ask for some advice.
My current situation is:
* I want to work on technical alignment. For exogenous reasons, not going to college (e.g., taking a year off, just being an a... |
600f6698-453b-4592-ad81-219bf212aa49 | trentmkelly/LessWrong-43k | LessWrong | Rationality versus Short Term Selves
Many of us are familiar with the marshmallow test.If you are not, here.
It is predictive of success, income, level of education, and several other correlated measures.
I'm here to argue for the marshmallow eaters, as a devil's advocate. Contra Ainslie, for instance. I do it out o... |
68b7d5d4-3510-43ac-9669-485febd5e20d | trentmkelly/LessWrong-43k | LessWrong | The 'Road Not Taken' in the Multiverse
Summary: The Many-Worlds interpretation of quantum mechanics can help us feel better about choosing between mutually exclusive options.
Previously I’ve suggested that when we think about the ethical implications of the many-worlds interpretation (MWI) of quantum mechanics, the k... |
f02b414e-d782-4e21-a3f2-96f8462a42ac | trentmkelly/LessWrong-43k | LessWrong | Jinnetic Engineering, by Richard Stallman
Thought the community might enjoy this:
Jinnetic Engineering |
ac83ec3f-b146-4d0a-a403-b697b44a6832 | trentmkelly/LessWrong-43k | LessWrong | Evolutionary prompt optimization for SAE feature visualization
TLDR:
* Fluent dreaming for language models is an algorithm based on the GCG method that can reliably find plain-text readable prompts for LLMs that maximize certain logits or residual stream directions by using gradients and genetic algorithms. Authors ... |
2c64e23a-c131-48c6-a3bb-7d64e2f9f9d2 | trentmkelly/LessWrong-43k | LessWrong | Locating My Eyes (Part 3 of "The Sense of Physical Necessity")
This is the third post in a sequence that demonstrates a complete naturalist study, specifically a study of query hugging (sort of), as described in The Nuts and Bolts of Naturalism. This one demos phases one and two: Locating Fulcrum Experiences and Getti... |
9c8a7599-d400-492d-9ac0-45c8e522f0c6 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | The AI Explosion Might Never Happen
[This is a crosspost from <https://amistrongeryet.substack.com/p/recursive-self-improvement-foom,> lightly edited for the LessWrong audience. This is my first LessWrong post; feedback greatly appreciated!]
LessWrong readers will be familiar with the concept of **recursive self-impr... |
a96bdfc8-41b7-4347-adde-7d58c8ca31d4 | trentmkelly/LessWrong-43k | LessWrong | Orthogonality
TL;DR: I used to think the best way to get really good at skill si was to specialize by investing lots of time ti into si. I was wrong. Investing lots of time ti into si works only as a first-order approximation. Once ti becomes large, investing in some other tj≠i produces greater real-world performance ... |
3aa9e8a7-b6ba-446b-bfcf-3ff92de8a8a9 | trentmkelly/LessWrong-43k | LessWrong | Equivalent of Information Theory but for Computation?
A quick google search brings up "computation theory". This does not seem to be as precise as information theory, we cannot talk about "n units of computation" the same way we can talk about "m bits of information". In fact there does not seem to be a generally acce... |
e5d701ca-25d3-4060-81dc-5e0ccab2a70f | trentmkelly/LessWrong-43k | LessWrong | Real biosafety is socially inconvenient
Marc Lipsitch and Thomas V. Inglesby wrote back in 2014:
> However, research that aims to create new potential pandemic pathogens (PPP) (1)—novel microbes that combine likely human virulence with likely efficient transmission in humans—is an exception to that rule. While this r... |
340418b6-c5ef-4ba9-91b4-b6109502f0ec | trentmkelly/LessWrong-43k | LessWrong | Meetup : Helsinki Meetup
Discussion article for the meetup : Helsinki Meetup
WHEN: 08 September 2013 03:00:00PM (+0300)
WHERE: Jämeräntaival 3, 02150 Espoo, Finland
We’ll be trying a new location this time: Takkakabinetti in Otaniemi.
In this meetup, Pyry will give us a brief introduction to Lojban, which has been... |
4a38c1f2-2626-4774-9cb0-e64f2e3ba6d8 | trentmkelly/LessWrong-43k | LessWrong | The Benefits of Rationality?
Robin wrote how being rational can harm you. Let's look at the other side: what significant benefits does rationality give?
The community here seems to agree that rationality is beneficial. Well, obviously people need common sense to survive, but does an additional dose of LessWrong-style... |
9bb0176c-3700-43df-b333-39db134f1dec | trentmkelly/LessWrong-43k | LessWrong | [LINK] Popular press account of social benefits of motivated reasoning
The Monkey Cage: "The Not Quite As Depressing Psychological Theory That Explains Washington":
> In a landmark article in Behavioral and Brain Sciences, Hugo Mercier and Dan Sperber propose an “argumentative” theory of human reason which both direc... |
5916b2f2-be99-4eed-881d-bf5b3f0a11d3 | trentmkelly/LessWrong-43k | LessWrong | Nils Nilsson's AI History: The Quest for Artificial Intelligence
I just noticed that AI pioneer and former Association for the Advancement of Artificial Intelligence (AAAI) head Nils Nilsson, has published his history of AI, The Quest for Artificial Intelligence: A History of Ideas and Achievements. The book is availa... |
5ec8fbd0-2fe5-4c29-8799-a74d28d9f604 | StampyAI/alignment-research-dataset/arbital | Arbital | Interruptibility
"Interruptibility" is a subproblem of [corrigibility](https://arbital.com/p/45) (creating an advanced agent that allows us, its creators, to 'correct' what *we* see as our mistakes in constructing it), as seen from a machine learning paradigm. In particular, "interruptibility" says, "If you do interr... |
87c2b8fc-83d9-4d35-b234-f52c1cdb16d5 | trentmkelly/LessWrong-43k | LessWrong | 2014 Survey Results
Thanks to everyone who took the 2014 Less Wrong Census/Survey. Extra thanks to Ozy, who did a lot of the number crunching work.
This year's results are below. Some of them may make more sense in the context of the original survey questions, which can be seen here. Please do not try to take the sur... |
03f42c66-d9ad-4cff-abca-900e430cdde3 | trentmkelly/LessWrong-43k | LessWrong | January-February 2025 Progress in Guaranteed Safe AI
Ok this one got too big, I’m done grouping two months together after this.
BAIF wants to do user interviews to prospect formal verification acceleration projects, reach out if you’re shipping proofs but have pain points!
This edition has a lot of my takes, so I sh... |
6d935d43-6979-4288-8d05-848059aa12fc | trentmkelly/LessWrong-43k | LessWrong | Valuable Things to Know While Discussing Moral Philosophy
|
d37a6f4c-f20c-4c00-a62a-7140bfc66e0b | trentmkelly/LessWrong-43k | LessWrong | Under a week left to win $1,000! By questioning Oracle AIs.
That clickbait-y title is, for once, perfectly accurate: you have until the 31st of August to submit an entry to the Oracle design contest, either a counterfactual or a low bandwidth Oracle.
(A note: some of the submissions were done by people who clearly d... |
ef509d83-684b-4263-a07c-7352241fb593 | awestover/filtering-for-misalignment | Redwood Research: Alek's Filtering Results | id: post2301
Introduction Developing safe and beneficial reinforcement learning (RL) agents requires making them aligned with human preferences. An RL agent trained to fulfil any objective in the real world will probably have to learn human preferences in order to do well. This is because humans live in the real world... |
f2fdfef7-74c7-42b7-898e-18e90d228f86 | StampyAI/alignment-research-dataset/arxiv | Arxiv | Formal Language Constraints for Markov Decision Processes
1 Introduction
---------------
The ability to impose safety constraints on an agent is
key to the deployment of reinforcement learning (RL) systems in real-world environments (Amodei et al., [2016](#bib.bib3)). Controllers that are derived mathematically typ... |
8b991fda-066e-41b2-8d22-c5be3a26eb07 | StampyAI/alignment-research-dataset/arxiv | Arxiv | On the Geometry of Adversarial Examples.
1 Introduction
---------------
Deep learning at scale has led to breakthroughs on important problems
in computer vision (Krizhevsky et al. ([2012](#bib.bib25))), natural language processing (Wu et al. ([2016](#bib.bib45))),
and robotics (Levine et al. ([2015](#bib.bib27))). ... |
edd81aec-3126-4e19-a602-8370eb1bd4f7 | trentmkelly/LessWrong-43k | LessWrong | Amputation of Destiny
Followup to: Nonsentient Optimizers, Can't Unbirth a Child
From Consider Phlebas by Iain M. Banks:
> In practice as well as theory the Culture was beyond considerations of wealth or empire. The very concept of money—regarded by the Culture as a crude, over-complicated and inefficient form... |
c4f5a1e7-df03-4769-a0b3-6635af801e28 | trentmkelly/LessWrong-43k | LessWrong | I didn't have to avoid you; I was just insecure
I don't usually post stories on LessWrong so I'm curious to see how this is received.
The first time we spoke, you asked me some questions that felt really invasive. I didn’t want that to happen again, so I avoided you the entire following year.
So when you said “Hi” a... |
29a60314-1c47-4d3b-8b62-ba2e79c26564 | trentmkelly/LessWrong-43k | LessWrong | A Walkthrough of Interpretability in the Wild (w/ authors Kevin Wang, Arthur Conmy & Alexandre Variengien)
New paper walkthrough: Interpretability in the Wild: A Circuit for Indirect Object Identification In GPT-2 Small is a really exciting new mechanistic interpretability paper from Redwood Research. They reverse eng... |
0260c5d8-83df-4c10-962c-162f70636cb4 | trentmkelly/LessWrong-43k | LessWrong | Trying to understand Hanson's Cultural Drift argument
At 2024's Manifest, Robin Hanson gave a talk (in his usual sweeping polymathic style) on cultural drift - a phenomenon for which he thinks "there's going to be hell to pay", and about which he is "scared, because this is a really big fundamental problem".
Watch th... |
1476b614-c1a0-4b0d-8fb0-8d9f9a6b23ff | trentmkelly/LessWrong-43k | LessWrong | Covid 11/17/22: Slow Recovery
While the news has been fast and furious on many fronts, Covid is not one of them.
Except that, this week, I got it.
On Monday evening, I started having chills. They seem better during the day, worse at night, slowly improving.
That night, I was almost entirely unable to sleep. I would... |
1b063017-84a1-47f5-b42f-ee27dc37b522 | StampyAI/alignment-research-dataset/arxiv | Arxiv | Relative Expressiveness of Defeasible Logics
Introduction
------------
Defeasible reasoning concerns reasoning where a chain of reasoning
can be defeated (that is, not considered the basis of an inference)
by another chain of reasoning (or, perhaps, several chains of reasoning).
Defeasible logics are a class of non... |
c561ccd6-45eb-48da-a64b-a0b54faa540e | StampyAI/alignment-research-dataset/aisafety.info | AI Safety Info | But why would misaligned AI pose a threat that we can’t deal with?
Human civilization is pretty robust. New technologies, cultural changes, and malign actors have sometimes caused great harm. However, in many (but not all) cases, we’ve been better able to adapt to their consequences than expected. Even the worst cases... |
a8e897fd-339c-4891-806e-85f094ce6021 | trentmkelly/LessWrong-43k | LessWrong | Could MMRPGs be used to test economic theories?
Most (all?) economic theories are based on models. Models make assumptions about the world. Sometimes these assumptions are reasonable and the model will make useful/accurate predictions. Sometimes these assumptions are unreasonable, and the model will make bad predictio... |
7bf54e62-b158-4d54-9d00-9374bc5da0e5 | StampyAI/alignment-research-dataset/arxiv | Arxiv | Clusterability in Neural Networks.
1 Clustering Neural Networks
-----------------------------
###
1.1 Definitions
We represent a neural network as a weighted, undirected graph G.
To do this for an MLP, we identify each neuron with any incoming or outgoing non-zero weights111When networks are pruned, often some ... |
63910b1b-2a58-43f6-957a-12b78c6cfd96 | trentmkelly/LessWrong-43k | LessWrong | Do I have what it takes to work on FAI?
Inspired by Eliezer's response to Maxwell_IV's question What's your advice for Less Wrong readers who want to help save the human race? I would like to find out whether I have the potential to help Singinst. I thought the first step to this would be reading the books Creating Fr... |
e6aad387-4365-4c4f-9eae-918c8e57c7b7 | trentmkelly/LessWrong-43k | LessWrong | Basic Question about LLMs: how do they know what task to perform
My understanding of LLMs:
I have some very basic understanding of LLM and the underlying transformer architecture. As I understand it, these LLMs are, as an analogy, basically a much-improved version of autocomplete. During training they are shown a word... |
81d82d79-e710-4698-95f2-85f045d17385 | StampyAI/alignment-research-dataset/eaforum | Effective Altruism Forum | Stampy's AI Safety Info - New Distillations #3 [May 2023]
Hey! This is another update from the distillers at the [AI Safety Info](https://aisafety.info/) website (and its more playful clone [Stampy](https://stampy.ai/)).
![](https://res.cloudinary.com/lesswrong-2-0/image/upload/f_auto,q_auto/v1/mirroredImages/EELddDm... |
74df9b65-7835-46b7-811f-9444f6423025 | trentmkelly/LessWrong-43k | LessWrong | Understanding and justifying Solomonoff induction
I've been trying to understand the uses and limitations of Solomonoff induction. Following the principle that in order to fully understand something you should explain it others, here's a try. I prefer to write such things in a form for dialogue, as that better reflect... |
4a8342de-ee34-4918-8627-34877395a06c | trentmkelly/LessWrong-43k | LessWrong | Irrationalism on Campus?
Since many LRers are fairly recent college graduates, it seems worthwhile to ask to what extent would people here agree with reports of rampant irrationalism such as this one: http://www.city-journal.org/2014/24_4_racial-microaggression.html from a right-leaning journalist known for her book T... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.