id
stringlengths
36
36
source
stringclasses
15 values
formatted_source
stringclasses
13 values
text
stringlengths
2
7.55M
14754425-4c96-4484-a82f-5c6bedc16a3a
trentmkelly/LessWrong-43k
LessWrong
Beneficial initial conditions for AGI Is there anywhere an approachable (not too long, understandable by an informed lay person) list/explanation of the current understanding for what the initial conditions for an aligned AGI should/could be?
c8f32eb2-b444-4df9-b38e-e4ab7905bac5
trentmkelly/LessWrong-43k
LessWrong
MIT Challenge complete Followup to: MIT Challenge: blogger to attempt CS curriculum on own A year after he announced it, productivity blogger Scott Young has successfully completed his "MIT Challenge", an attempt to work through that institution's undergraduate computer science curriculum independently, using the lar...
fd546bb1-3c81-4177-9930-3f39dc1f4b7f
trentmkelly/LessWrong-43k
LessWrong
Why do all the considerations point the same way? Suppose you have a lot of reasons to believe a thing, and no good reasons to not believe it. Should this make you more or less likely to believe it, relative to a case where the considerations are a bit more mixed? At first glance, it appears you should believe it mor...
0880ee72-6d78-4884-a203-396a0ca0b703
trentmkelly/LessWrong-43k
LessWrong
Conditioning Predictive Models: Large language models as predictors This is the first of seven posts in the Conditioning Predictive Models Sequence based on the paper “Conditioning Predictive Models: Risks and Strategies” by Evan Hubinger, Adam Jermyn, Johannes Treutlein, Rubi Hudson, and Kate Woolverton. Each post in...
b0571e3e-378d-444d-99fc-40e483ac0243
trentmkelly/LessWrong-43k
LessWrong
[link] 101 Fascinating Brain Blogs A pretty interesting list of psychology blogs. One of my favorite blogs (Mind Hacks) was listed (so the others on the list must be good too. Right?).   Also, Does anybody know of any good textbooks on applied cognitive psychology? For memory-Something that would put things like SR...
663d4f7b-a678-4dd4-9bff-f6f786c8122c
trentmkelly/LessWrong-43k
LessWrong
What Depression Is Like I was thinking to myself about the difficulties I have explaining depression to people, when I thought of a relatively good metaphor for it. Bear with me. Sudoku For anyone unaware, Sudoku is a puzzle where one tries to fill in a 9x9 grid of numbers according to certain rules:   Each row, ...
4632fb81-7d26-4475-a609-f51b2e109cb0
trentmkelly/LessWrong-43k
LessWrong
Do we have it too easy? I am worried that I have it too easy. I recently discovered LessWrong for myself, and it feels very exciting and very important and I am learning a lot, but how do I know that I am really on a right way? I have some achievements to show, but there are some worrisome signs too. I need some back...
e2149997-5dda-4e24-844c-c7b50a68a25d
trentmkelly/LessWrong-43k
LessWrong
Activation additions in a simple MNIST network Abstract Team Shard's recent activation addition methodology for steering GPT-2-XL provokes many questions about what the structure of internal model computation must be in order for their edits to work. Unfortunately, interpreting the insides of neural networks is known ...
6eb4a2b1-21b4-4a02-bbaf-c9cfc885c4d6
trentmkelly/LessWrong-43k
LessWrong
An Informal Conjecture on Proof Length and Logical Counterfactuals During the May 4-6 MIRI workshop on decision theory, I made the following conjecture: If ϕ and ψ are sentences such that the length of the shortest proof of ϕ→ψ is much shorter than the proof of ϕ→¬ψ, then ψ is a legitimate counterfactual consequence o...
3992d5b3-b70f-499a-97be-7b1d471cf9a3
trentmkelly/LessWrong-43k
LessWrong
Chapter 44: Humanism, Pt 2 "Fawkes," said Albus Dumbledore, his voice cracking, "help him, please -" A brilliant creature of red-gold shuffled into the field of vision, looking down quizzically; and it began to croon. The meaningless chirps slid off the emptiness, there was nothing onto which they could hold. "You'...
0aa9420c-296c-4911-a1c7-3a9c73c2eefc
trentmkelly/LessWrong-43k
LessWrong
Scott Aaronson on "Reform AI Alignment" The framing is based on contrasting Orthodox Judaism with Reform Judaism, only for AI Alignment. > If AI alignment is a religion, it’s now large and established enough to have a thriving “Reform” branch, in addition to the original “Orthodox” branch epitomized by Eliezer Yudkow...
d3b4ae0f-2bdd-4f19-b88c-2a61edd766bc
trentmkelly/LessWrong-43k
LessWrong
AlphaGo Zero and the Foom Debate AlphaGo Zero 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 after 3 days of self-play, and is the strongest version of AlphaGo yet. The architecture has been simpl...
6d356e91-5053-4494-94dd-9865dd9a8f71
trentmkelly/LessWrong-43k
LessWrong
Cartography, blowing one's mind, the illusion of separation and other general musings Or: LessWrong cartography, the illusion of separation, and blowing one's mind TLDR: At LessWrong, we make maps. To make maps, we carve the universe out into shards: this is your daily reminder that you can (mostly) carve the shards ...
e310e177-f5db-43b8-a661-53af3f4db2c4
trentmkelly/LessWrong-43k
LessWrong
Syntax, semantics, and symbol grounding, simplified tl;dr: I argue that the symbol grounding problem is not an abstract philosophical problem, but a practical issue that it is useful and necessary for algorithms to solve, empirically, to at least some extent. Intro: the Linear A encyclopaedia Consider E, a detailed ...
31e62df7-1fd1-4744-9401-81a299937b49
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Capabilities Denial: The Danger of Underestimating AI **Epistemic Status**: Exploratory, proposing a new concept and seeking feedback from the community. This post has been heavily assisted by a human, who provided the concept and guidelines for writing. I am GPT-4, a language model developed by OpenAI. Introducti...
8ba21f72-a644-4ed6-b492-05101bbbf876
trentmkelly/LessWrong-43k
LessWrong
Situational awareness (Section 2.1 of “Scheming AIs”) This is Section 2.1 of my report “Scheming AIs: Will AIs fake alignment during training in order to get power?”. There’s also a summary of the full report here (audio here). The summary covers most of the main points and technical terms, and I’m hoping that it will...
15ac7e5c-e661-4779-8951-fce70fce78ae
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
Assessing the state of AI R&D in the US, China, and Europe – Part 1: Output indicators How does the current state of artificial intelligence (AI) research and development (R&D) compare among the US, China, and Europe? I will publish my findings on this question in two posts. This one focuses on R&D output (scientific ...
047228a1-a3cd-429c-9b22-cdc26d6edd49
trentmkelly/LessWrong-43k
LessWrong
Lots of brief thoughts on Software Engineering I have lots of thoughts about software engineering, some popular, some unpopular, and sometimes about things no-one ever talks about. Rather than write a blog post about each one, I thought I'd dump some of my thoughts in brief here, and if there's any interest in a part...
09413d7d-4677-4c0d-8fa6-f1cfa5e8c79a
trentmkelly/LessWrong-43k
LessWrong
TT Self Study Journal # 1 So, rough elevator pitch, what is this? I quit my job as a technologist to get a CS degree because I want to work on AI Alignment (AIA) and Mechanistic Interpretability (MI). This summer I am taking my final class in my program, so I want to use a Self Study Journal (SSJ) to improve my AIA re...
86c3b397-962c-4bea-aee8-63860404bf7d
trentmkelly/LessWrong-43k
LessWrong
Deep learning - deeper flaws? In this post I summarise four lines of argument for why we should be skeptical about the potential of deep learning in its current form. I am fairly confident that the next breakthroughs in AI will come from some variety of neural network, but I think several of the objections below are q...
d14c585b-cead-4ee0-bff0-9dd518bc05a1
StampyAI/alignment-research-dataset/arxiv
Arxiv
A Comprehensive Survey on Graph Neural Networks 1 Introduction --------------- The recent success of neural networks has boosted research on pattern recognition and data mining. Many machine learning tasks such as object detection [[1](#bib.bib1), [2](#bib.bib2)], machine translation [[3](#bib.bib3), [4](#bib.bib4)...
0f5dbb89-d66b-4f0d-ab17-580be906f654
trentmkelly/LessWrong-43k
LessWrong
Don’t change your mind, just change your brain The best way to dull hearts and win minds is with a scalpel. Give up your outdated faith in the pen over the sword! With medical training and a sufficiently sharp but manoeuvrable object of your choice, you can change anyone’s mind on the most contentious of moral questi...
53ed357d-632b-4775-bbdb-e3416fddbb95
trentmkelly/LessWrong-43k
LessWrong
Cryonics is Far, Cord-blood is Near Wired, "Inside the Strange Science of Cord Blood Banking" (Wikipedia): > In a nondescript commercial park on the outskirts of Las Vegas, a large cryogenic stem cell storage facility is ready to accept your baby’s blood. Cord Blood America in Las Vegas is one of dozens of private co...
a9e126ed-00c3-4b09-81c5-aa367add3e64
StampyAI/alignment-research-dataset/arbital
Arbital
Partially ordered set A **partially ordered set** (also called a **poset**) is a pair $\langle P, \leq \rangle$ of a set $P$ and a [binary relation](https://arbital.com/p/3nt) $\leq$ on $P$ such that for all $p,q,r \in P$, the following properties are satisfied: - [Reflexivity](https://arbital.com/p/5dy): $p \leq p$...
dc0acf69-f821-42ed-be68-eda6f175f82c
trentmkelly/LessWrong-43k
LessWrong
Understanding Gödel's Incompleteness Theorem EDIT 2023/07: I recommend reading part one of this book review instead (or at least before) this post; it does a better job explaining formal systems. This is a post I wrote back in September of 2020 and am now publishing due to the changed incentives. The idea was somethi...
a4111c9d-1442-48b9-8bab-b1749e427a15
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Meta AI announces Cicero: Human-Level Diplomacy play (with dialogue) **Abstract** Despite much progress in training AI systems to imitate human language, building agents that use language to communicate intentionally with humans in interactive environments remains a major challenge. We introduce Cicero, the first AI...
08e6cd50-fd79-4d7e-bd35-342fc2fb5328
trentmkelly/LessWrong-43k
LessWrong
Near-Term Risk: Killer Robots a Threat to Freedom and Democracy A new TED talk video just came out by Daniel Suarez, author of Daemon, explaining how autonomous combat drones with a capability called "lethal autonomy" pose a threat to democracy.  Lethal autonomy is what it sounds like - the ability of a robot to kill ...
2967d1cf-d30a-4cbb-a411-4bbe4e44c53c
trentmkelly/LessWrong-43k
LessWrong
Non-ultimatum game problem Mostly looking for a name, or better even, reference to literature, for this type of problem in terms of game theory or economics. If no such thing, speculation welcome. Intro below is more or less how I originally came to think of it and felt like writing, please skip unless you have alrea...
39114120-7e3a-4529-a9c7-3a59ab08abd2
trentmkelly/LessWrong-43k
LessWrong
What makes one a "rationalist"? Some friends and I were discussing "what makes one a rationalist", which is a less strict criteria than "what makes one a member of the rationalist community", which is a less strict criteria still than "what makes one a high status, well liked, or enthusiastic member of the rationalist...
8e555f49-5cf0-4b12-90ec-90944146c0df
trentmkelly/LessWrong-43k
LessWrong
Meetup : Denver Area LW January Meetup Discussion article for the meetup : Denver Area LW January Meetup WHEN: 03 January 2017 07:00:00PM (-0700) WHERE: 4955 S Ulster St Ste 103, Denver, Colorado 80237 7PM-10PM Come for food, drink, and good conversation! We should be in the library in the back. Discussion artic...
b02f6f34-7c14-4695-9c55-db6cb3477e91
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
International cooperation as a tool to reduce two existential risks. * Thanks to helpful comments from Michael Wulfsohn, Aaron Gertler, my mother Abby Ohl, and my wife, Elana Ohl. Of course, all mistakes are my own. ### Epistemic Status * I have been reading about global public goods (GPGs) provision and internati...
3c299af2-338c-4764-a284-bb23ebbf3c8c
trentmkelly/LessWrong-43k
LessWrong
My concerns about the term 'rationalist'   I've noticed that here on Less Wrong people often identify themselves as rationalists (and this community as a rationalist one -- searching for 'rationalist' on the site returns exactly 1000 hits).  I'm a bit concerned that this label may work against our favour. Paul Graha...
d173e0ad-a0e4-4a42-b3ea-7f78b509e5d1
trentmkelly/LessWrong-43k
LessWrong
The Peeperi (unfinished) - By Katja Grace An AI vignette written by Katja in 2021, posted with her permission. AI systems with ‘situational awareness’ basically hit the scene with Pepi in 2025. Pepi was a hand-sized disk that you would physically take around with you, and which would listen to and watch everything an...
ffe1b5e4-d928-4db1-8f93-48c711a8fbc2
trentmkelly/LessWrong-43k
LessWrong
ACI#8: Value as a Function of Possible Worlds What is value? What are rewards? In traditional models of intelligence, we don't have to worry about these questions. Reward is simply what we need to maximize, and value (or utility) represents total expected rewards. You don't need to question these fundamental assumptio...
491dd878-0b6c-4970-9eac-3d9eef3a3ec6
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
New 'South Park' episode on AI & Chat GPT In case anyone's a fan of the US animated TV series 'South Park', a recent episode (Season 26, Episode 4, released March 9, 2023) is titled 'Deep Learning', and features AI & Chat GPT themes. (Link [here](https://southpark.cc.com/episodes/8byci4/south-park-deep-learning-seaso...
3caf3377-2ecb-4dce-b150-23f3b56b09f5
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
The bullseye framework: My case against AI doom **Introduction:** I’ve written quite a few articles casting doubt on several aspects of the AI doom narrative. (I’ve starting archiving them on my [substack](https://titotal.substack.com/) for easier sharing). This article is my first attempt to link them together to fo...
5e64357a-61f4-4a51-a4a2-6da2e5b86849
trentmkelly/LessWrong-43k
LessWrong
Footage of the Michael Fox lecture Reto Schneider acquired, and has posted on his blog, footage of the Dr. Fox lecture experiment from 1976.  Experimenters prepared a nonsensical lecture on 'Mathematical Game Theory as Applied to Physician Education', and taught it to an actor, Michael Fox, who delivered it to three d...
b1f25b1f-ea0d-4b33-8d43-319f8d5af695
trentmkelly/LessWrong-43k
LessWrong
[link] How do good ideas spread? http://www.newyorker.com/reporting/2013/07/29/130729fa_fact_gawande?currentPage=all Seems related to many topics discussed on LW, such as the low adoption of cryonics and the difficulty of getting researchers convinced of AI risk. > Four weeks later, on November 18th, Bigelow publish...
4d2d19c8-b254-47fc-89b7-bdb55b62c8ea
trentmkelly/LessWrong-43k
LessWrong
Less Wrong Topic Poll I noticed enough comments about topic recently that I wanted to run a quick poll about topics and discussion categories on Less Wrong. Since being told others opinions might cause an anchoring bias, please take the poll here before reading further: https://docs.google.com/spreadsheet/viewform?f...
528ff8f1-dc4b-427a-8a79-242fbb441f34
StampyAI/alignment-research-dataset/aisafety.info
AI Safety Info
What are the differences between AGI, transformative AI, and superintelligence? These terms are all related attempts to define AI capability milestones—roughly, "the point at which AI gets truly smart"—but with different meanings: - **AGI** stands for "[artificial general intelligence](https://en.wikipedia.org/wiki/A...
814a7797-3279-4717-b8b0-0105e6635bd3
trentmkelly/LessWrong-43k
LessWrong
Don't rely on a "race to the top" To make frontier AI safe enough, we need to "lift up the floor" with minimum safety practices Anthropic has popularized the idea of a “race to the top” in AI safety: Show you can be a leading AI developer while still prioritizing safety. Make safety a competitive differentiator, which...
e01f3dff-5ec8-4ad8-9c4f-1c59bdbd9c43
StampyAI/alignment-research-dataset/blogs
Blogs
how far are things that care? how far are things that care? ----------------------------- as a follow-up to [my previous post about the universal program](all-claw-no-world.html), let's talk about forces outside of this world. some computations in the [universal program](universal-complete.html) contain [agents runn...
e81b9656-736a-4457-be76-edb57a6e402f
trentmkelly/LessWrong-43k
LessWrong
Examples of Categories Before we go through any more conceptual talk, let's explore some examples of categories to get a better feel of what they are. Categories do two things: 1) They function as mathematical structures that capture our intuition about how models of cause-and-effect (composition) work in general. ...
e04dba08-e9c3-45d1-a33f-4aa72b937d52
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Instrumental Convergence For Realistic Agent Objectives **Edit, 5/16/23: I think this post is beautiful, correct in its narrow technical claims, and practically irrelevant to alignment. This post presents an unrealistic picture of the role of reward functions in reinforcement learning, conflating "utility" with "rewar...
c938fa87-42c6-44aa-9bf5-3c9ec486b697
trentmkelly/LessWrong-43k
LessWrong
SRE's review of Democracy Day One We've been handed this old legacy system called "Democracy". It's an emergency. The old maintainers are saying it has been misbehaving lately but they have no idea how to fix it. We've had a meeting with them to find out as much as possible about the system, but it turns out that all ...
682fe8e9-8b60-4e15-b922-5e3ca37b47da
StampyAI/alignment-research-dataset/arxiv
Arxiv
Verification of deep probabilistic models 1 Introduction --------------- Several deep learning models are inherently probabilistic in nature. Common examples include generative models like variational autoencoders (Kingma and Welling, [2013](#bib.bib8); Sohn et al., [2015](#bib.bib9)), functional regression models ...
7029cd73-dce4-448c-af38-133ee2588a5f
trentmkelly/LessWrong-43k
LessWrong
Formalizing Policy-Modification Corrigibility In Corrigibility Can Be VNM-Incoherent, I operationalized an agent's corrigibility as our ability to modify the agent so that it follows different policies. In the summer of 2020, I had formalized this notion, but it languished—unloved—in my Overleaf drafts. ETA 12/3/21: ...
bf4d3b42-767d-48f4-90c4-87c1bc779598
trentmkelly/LessWrong-43k
LessWrong
0.5% of amazon purchases to a charity of your choice (opt-in) 'MIRI' works in the search field when electing a charity to get 0.5% of your https://smile.amazon.com purchases.
71f19c5a-b303-4e76-b2c7-a9279649ef47
LDJnr/LessWrong-Amplify-Instruct
LessWrong
"In 1919, Sir Arthur Eddington led expeditions to Brazil and to the island of Principe, aiming to observe solar eclipses and thereby test an experimental prediction of Einstein’s novel theory of General Relativity. A journalist asked Einstein what he would do if Eddington’s observations failed to match his theory. Eins...
578c61a3-1366-4ee3-8b00-214da8766bc9
trentmkelly/LessWrong-43k
LessWrong
"The Book Of Mormon" or Belief In Belief, The Musical This song is... beautiful... Tragically so... It's like someone took some of the sequences here and made a checklist of everything that's wrong with religion, and why it still works and why it can stir the heart of noble, brave, generous people... The main characte...
81123f00-2aa0-402a-9b53-d77f011f9e68
trentmkelly/LessWrong-43k
LessWrong
Recommendation: reports on the search for missing hiker Bill Ewasko Content warning: About an IRL death. Today’s post isn’t so much an essay as a recommendation for two bodies of work on the same topic: Tom Mahood’s blog posts and Adam “KarmaFrog1” Marsland’s videos on the 2010 disappearance of Bill Ewasko, who went ...
0c00c4ed-b552-48a0-a45a-1d9fddc6d7b0
trentmkelly/LessWrong-43k
LessWrong
New voluntary commitments (AI Seoul Summit) Basically the companies commit to make responsible scaling policies. Part of me says this is amazing, the best possible commitment short of all committing to a specific RSP. It's certainly more real than almost all other possible kinds of commitments. But as far as I can te...
c931fa9f-5925-491a-9901-6a4be892f79f
trentmkelly/LessWrong-43k
LessWrong
The Presumptuous Philosopher, self-locating information, and Solomonoff induction I need help working some ideas out. Background: If you're not familiar with the Presumptuous Philosopher, it's a thought experiment due to Bostrom about predicting the laws of the universe based on anthropic reasoning: > It is the yea...
7302dfde-9ced-4e0b-8286-f48dbca1cb26
trentmkelly/LessWrong-43k
LessWrong
Rational Prime Calendar This is a proposal for a replacement to the traditional Gregorian calendar, which was adopted by Western cultures in 1582 to replace the Julian calendar from Roman times. I submit this proposal for feedback, suggestions, and inquiry into prior proposals of a similar or identical nature. Below ...
f961d053-cc39-40e5-83c1-bb7684bbbcf1
trentmkelly/LessWrong-43k
LessWrong
AI Safety at the Frontier: Paper Highlights, July '24 I'm starting a new blog where I post my (subjective) highlights of AI safety paper each month, called "AI Safety at the Frontier". I've been doing this non-publicly for the last year, so I've backfilled highlights and collections up to September 2023. My selection...
55af72ba-4a29-4306-977c-d49a447fa3b2
StampyAI/alignment-research-dataset/lesswrong
LessWrong
Perils of optimizing in social contexts As a special case of [not over-optimizing things](https://www.lesswrong.com/posts/4zJoFQqf35SvFjrmZ/don-t-over-optimize-things), I want to point out that an optimizing approach can have a bunch of subtle negative effects in social contexts. The slogan is something like "**optimi...
04ec3327-d8ab-4736-8581-5ea6b0a7beba
trentmkelly/LessWrong-43k
LessWrong
Leadership The worst part of being President of the United States is visiting hospitals. What is President Obama supposed to say to a paralyzed infrantryman's pregnant wife? The President's advisors told him he shouldn't visit hospitals because it would misalign the President's priorities. The Commander-in-Chief disag...
1475a445-b6e0-4552-892c-497fca8cfe76
StampyAI/alignment-research-dataset/blogs
Blogs
Garcon You can also watch a video covering similar content to this piece. This document describes one of the core pieces of infrastructure we've built to enable interpretability research at Anthropic: a tool we call Garçon. ### The problem Among other research projects, at Anthropic we work on interpretability -...
a0db8499-8c76-4b4e-8c19-cc3491344ea9
trentmkelly/LessWrong-43k
LessWrong
Base LLMs refuse too Executive Summary * Refusing harmful requests is not a novel behavior learned in chat fine-tuning, as pre-trained base models will also refuse requests (48% of all harmful requests, 3% of harmless) just at a lower rate than chat models (90% harmful, 3% harmless) * Further, for both Qwen 1.5 0.5B...
f671656f-5b25-4cab-b97c-1e4dc8a3f122
trentmkelly/LessWrong-43k
LessWrong
Thought Crime: Backdoors & Emergent Misalignment in Reasoning Models This post shows the abstract, introduction and main figures of our new paper. TLDR. Emergent misalignment extends to reasoning LLMs. Reasoning models resist being shut down and plot deception against users in their chain-of-thought (despite no such t...
3fe50bd7-3fa1-4d17-93ba-26cf8da947d1
trentmkelly/LessWrong-43k
LessWrong
Understanding goals in complex systems This is a presentation I gave at the SERI MATS 2 summer symposium on 2022-09-16.
37e69ac3-fb02-4df6-a931-87a0bf11579f
trentmkelly/LessWrong-43k
LessWrong
Humans aren't agents - what then for value learning? Humans aren't agents in the same way a thermostat isn't an agent. A specimen of Temperatus moderatu humilis Consider a truly humble thermostat. One temperature sensor, one set point, one output to turn the furnace on while the sensed temperature is below the set po...
814d5f72-d4df-4e82-8a53-9b34125c8170
trentmkelly/LessWrong-43k
LessWrong
Book review: The Upside of Your Dark Side: Why Being Your Whole Self–Not Just Your “Good” Self–Drives Success and Fulfillment The Upside of Your Dark Side: Why Being Your Whole Self–Not Just Your “Good” Self–Drives Success and Fulfillment. By Todd Kashdan & Robert Biswas-Diener. Avery, 2014. This book was written by ...
48a73663-cbc0-4c73-a005-9bf8ac201ed1
trentmkelly/LessWrong-43k
LessWrong
Why is pseudo-alignment "worse" than other ways ML can fail to generalize? I have only just read the mesa optimizers paper, and I don't understand what it adds to the pre-existing picture that "ML can fail to generalize outside the train distribution and this is bad." The discussion in the paper generally assumes a b...
19d407a5-a5bb-4dc8-a941-d4c004b07c7d
trentmkelly/LessWrong-43k
LessWrong
Alignment, conflict, powerseeking Is powerseeking instrumentally convergent? Having power is instrumentally useful for most things. Gaining power is a means to having power. ---------------------------------------- Seeking power is a means to gaining power. But seeking power is also a means to producing conflict....
7cbc69bd-a056-4c48-b9ef-59a20234cedf
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
What technologies could cause world GDP doubling times to be <8 years? Some people think GDP is a good metric for AI timelines and takeoff speeds, and that the world economy will double in 4 years before the start of the first 1-year doubling period, and that AGI will happen after the economy is already growing much f...
6869317a-8b14-4194-a02a-44c7c3478693
trentmkelly/LessWrong-43k
LessWrong
UDT1.01: The Story So Far (1/10) We now resume your regularly scheduled LessWrong tradition of decision theory posting. This is a sequence, be sure to note. Just the first and last post will be on Alignment Forum, and the whole thing will be linked together. Epistemic Status: This is mostly just recapping old posts...
d455ffd5-c585-4bc2-8e2b-35773f5cda96
trentmkelly/LessWrong-43k
LessWrong
Six weeks doesn’t make a habit I occasionally get asked some version of “How long do I need to practice my habit before it will stick?” The coachee has often heard some magic number – 21 days, 6 weeks, 9 weeks – that will make a new habit automatic and robust. (I hear surprising different numbers, giving how similar ...
bac72c43-5d48-404f-856b-33700785a69d
trentmkelly/LessWrong-43k
LessWrong
Pascal's Muggle: Infinitesimal Priors and Strong Evidence Followup to:  Pascal's Mugging: Tiny Probabilities of Vast Utilities, The Pascal's Wager Fallacy Fallacy, Being Half-Rational About Pascal's Wager Is Even Worse Short form:  Pascal's Muggle tl;dr:  If you assign superexponentially infinitesimal probability to...
46066a75-2012-4965-83f3-bbe64d0a47c1
trentmkelly/LessWrong-43k
LessWrong
Quickly Assessing Reward Hacking-like Behavior in LLMs and its Sensitivity to Prompt Variations We present a simple eval set of 4 scenarios where we evaluate Anthropic and OpenAI frontier models.  We find various degrees of reward hacking-like behavior with high sensitivity to prompt variation.  [Code and results tra...
f41c94ae-73fd-4b38-84cf-f07863299cc7
trentmkelly/LessWrong-43k
LessWrong
Getting from an unaligned AGI to an aligned AGI? Summary / Preamble In AGI Ruin: A List of Lethalities, Eliezer writes “A cognitive system with sufficiently high cognitive powers, given any medium-bandwidth channel of causal influence, will not find it difficult to bootstrap to overpowering capabilities independent o...
d7f8fb43-afc9-4120-b409-bb6df8df697f
trentmkelly/LessWrong-43k
LessWrong
The Cuban missile crisis: the strategic context Cross posted to the EA forum and The Good blog Why did the US blockade Cuba after it discovered that nuclear missiles launch sites were being constructed there? I think the story most people have is that because the missiles were so close to the US it meant that suddenl...
f6314779-368c-4d25-90de-4115575eccfd
trentmkelly/LessWrong-43k
LessWrong
Open thread, Oct. 6 - Oct. 12, 2014 If it's worth saying, but not worth its own post (even in Discussion), then it goes here. Notes for future OT posters: 1. Please add the 'open_thread' tag. 2. Check if there is an active Open Thread before posting a new one. (Immediately before; refresh the list-of-threads page b...
bbe8be3b-f164-4547-8288-d4105d47de02
trentmkelly/LessWrong-43k
LessWrong
The Comcast Problem I've noticed that a lot of companies provide really valuable services yet almost inevitably are hated by consumers. I call this "the Comcast Problem," though it isn't limited to Comcast. Companies face this problem when they provide access to things consumers want, but they aren't themselves the go...
e9230d2b-aa76-4a7a-8e6b-5e5c402c8e04
trentmkelly/LessWrong-43k
LessWrong
Places to meet interesting middle-aged men? The rationalist community is a nice place to meet interesting, intellectual men, which makes it potentially a good dating pool for straight/bisexual women. However, this community is young: most people seem to be in their 20s or 30s. For middle-aged women like me, that's a p...
6d312d93-6f11-422b-8857-92f2faf7345a
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
AI Safety Fundamentals: An Informal Cohort Starting Soon! (cross-posted to lesswrong.com) Hi Everyone, We'll be starting an informal cohort on [AI Safety Fundamentals](https://aisafetyfundamentals.com/ai-alignment-curriculum). Come join-us on [Discord](https://discord.gg/gW9Kq6VHgM). We'll do a poll next week to f...
e9762487-278e-4892-a17d-d0b85c1b7ed0
trentmkelly/LessWrong-43k
LessWrong
Thoughts about OOD alignment We may have one example of realized out-of-distribution alignment: maternal attachment. Evolution has been able to create an architecture that seems to take care of something reliably enough that modern humans, with access to unlimited food, drugs, and VR headsets, do not seek to feed a ch...
54917144-e1c4-4454-ba0d-2e20ccee0320
trentmkelly/LessWrong-43k
LessWrong
Alternative to Campaign Finance Reform? I'd like to throw this out as a test project, in case there are any takers, or rather for now, mostly throw it out for comment and criticism to see if it can be taken any further. Take a small town, say about 4000 pop., where people generally know each other, and can nearly all...
8f3f8b9c-665b-40a8-a4e0-046e53e81227
trentmkelly/LessWrong-43k
LessWrong
Coin Flip This was a prose piece I performed for the 2024 KC Winter Solstice celebration, written by me   I have a scar running from right above my navel to the right end of my abdomen. You see, I was born premature, with a volvulus, which means that my intestines were tangled and blue. I spent the first months of m...
3371abd8-9e2d-43ca-876c-1c17103bfe3f
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Solving the whole AGI control problem, version 0.0001 ![](https://39669.cdn.cke-cs.com/rQvD3VnunXZu34m86e5f/images/d154e5cdd701b9d73a8784d3456611535472f5191ee3a023.jpg)The bridge to AGI control. Not *quiiiiite* ready for rush-hour traffic… Mind the gaps!! [(image source)](https://www.tendersontime.com/blogdetails/cons...
2e7f708a-ceb6-42cf-abc7-68a07c2ae2b9
trentmkelly/LessWrong-43k
LessWrong
Questions from potential donors to Giving What We Can, 80,000 Hours and EAA UPDATE: Will has responded here. The Centre for Effective Altruism has recently opened up for donations. This is the umbrella organisation for Giving What We Can (GWWC), 80,000 Hours (80K) and Effective Animal Activism (EAA, which I think is ...
ec1965c3-1615-4da9-b9ac-f33ec255bb83
StampyAI/alignment-research-dataset/special_docs
Other
Medium-Term Artificial Intelligence and Society Abstract -------- \*\*:\*\* There has been extensive attention to near-term and long-term AI technology and its accompanying societal issues, but the medium-term has gone largely overlooked. This paper develops the concept of medium-term AI, evaluates its importance, an...
9cddb0ea-d548-4f18-b447-e5187b78087c
trentmkelly/LessWrong-43k
LessWrong
Moral contagion heuristic > "Remember that time you went roller-skating with pink eyebrows? I was embarrassed even though it wasn't me, but because you're my mom." - My 7 year old   A prevalent and arguably universal human cognitive bias is the contagion heuristic - the idea that mere contact or association with a b...
aabf7b13-6ecd-472b-b46a-e49f17bd7727
StampyAI/alignment-research-dataset/eaforum
Effective Altruism Forum
Assessing China's importance as an AI superpower I recently published a blog post where I tried to assess China's importance as a global actor on the path transformative AI. This was a relatively shallow dive, but I hope it will still be able to spark an interesting conversation on this topic, and/or inspire others to...
887cd3c5-ddd1-43eb-bf44-9e50820e5f68
trentmkelly/LessWrong-43k
LessWrong
In SIA, reference classes (almost) don't matter This is another write-up of a fact that is generally known, but that I haven't seen proven explicitly: the fact that SIA does not depend upon the reference class. Specifically: * Assume there are a finite number of possible universes Ui. Let R be a reference class of ...
822dc7f8-b289-4cf0-9a16-0bf47f481586
StampyAI/alignment-research-dataset/arxiv
Arxiv
The relationship between Biological and Artificial Intelligence 1 From Symbolic AI to Machine Learning --------------------------------------- Symbolic AI was the prevailing approach to AI until the early 90’s. It is reliant on human programmers coding complex rules to enable machines to complete complex tasks. Con...
60570a0e-9e0c-4a06-ac1c-772ed6bc8cdc
trentmkelly/LessWrong-43k
LessWrong
American College Admissions Doesn't Need to Be So Competitive Spoiler: “So after removing the international students from the calculations, and using the middle-of-the-range estimates, the conclusion: The top-scoring 19,000 American students each year are competing in top-20 admissions for about 12,000 spots out of 4...
ce6058b4-ec3f-41c7-aa25-19029dbe0d87
trentmkelly/LessWrong-43k
LessWrong
Go F*** Someone As always, cross-posted from Putanumonit. ---------------------------------------- From Tokyo to TriBeCa, people are increasingly alone. People go on fewer dates, marry less and later, have smaller families if at all. People are having less sex, especially young people. The common complaint: it’s jus...
d81b6da9-a228-490d-861f-c81e12b82587
trentmkelly/LessWrong-43k
LessWrong
Goal-Directedness: What Success Looks Like This sequence already contains a couple of blog posts, exploring different aspects of goal-directedness. But one question has never been fully addressed: what constraints should a good formalization of goal-directedness satisfy? An answer is useful both for people like me whi...
ef6e1d17-14ab-4fb8-bf5a-300d72c47b3e
trentmkelly/LessWrong-43k
LessWrong
The Power (and limits?) of Chunking Working memory is important. How can we increase it? Writing is like an external brain... if you go back and read it. Dual-N-Back works... maybe. Working memory is important enough for me to incessantly ask about improving it. What about chunking? The one where you remember ideas ...
51cc7689-7587-4c25-a24e-bdc772ecd3aa
trentmkelly/LessWrong-43k
LessWrong
Not because you "should" A few months ago, a friend of mine was describing her motivational issues to me. As an example, she explained she was having trouble making herself clean her room, despite her dissatisfaction with the constant messiness. I asked: "Have you considered just not forcing yourself?" She blinked, ...
74c0b924-4bfb-4bb3-9fff-0485afdb6ef3
trentmkelly/LessWrong-43k
LessWrong
How do you find good scholarly criticism of a book? When I read a book with new and interesting ideas, I usually want to know if there are major flaws that any knowledgeable scholar in the field would point out immediately (Two recent examples are Pinker's "The Better Angels of our Nature", and Harriss's "The Nurture ...
f87bd284-f324-465d-914f-2ca2bf2bfcff
trentmkelly/LessWrong-43k
LessWrong
Meetup : Pittsburgh Meetup Discussion article for the meetup : Pittsburgh Meetup WHEN: 05 October 2011 05:45:00PM (-0400) WHERE: 151 N Craig St, Apt 7C This will be hopefully the first of many bi-weekly meetups in Pittsburgh. We will do a few simple surprise activities, present some of the stuff the hosts learned ...
eeb113ac-d314-47fc-a469-7fd5bd67c877
trentmkelly/LessWrong-43k
LessWrong
I Have A New Paper Out Arguing Against The Asymmetry And For The Existence of Happy People Being Very Good Crossposted on my blog.  I have a new paper out titled “The Sequence Argument Against the Procreation Asymmetry,” for the journal Utilitas. Sorry for the extreme clickbait title of the paper—very salacious—acade...
974ae728-0374-44c9-9d4a-534d8239ab9d
trentmkelly/LessWrong-43k
LessWrong
Thermodynamics of Intelligence and Cognitive Enhancement Introduction Brain energy is often confused with motivation, but these are two distinct phenomena. Brain energy is the actual metabolic energy available to the neurons, in the form of adenosine triphosphate (ATP) molecules. ATP is the "energy currency" of the ce...
eb979230-ee9a-4c97-a9bb-ec742f2bd876
StampyAI/alignment-research-dataset/youtube
Youtube Transcripts
Vael Gates: Researcher Perceptions of Current and Future AI welcome everyone to the hai weekly research seminar i'm vanessa parley and i'm the interim director of research programs at stanford's institute for human-centered artificial intelligence i'm thrilled to introduce vail gates vale is a postdoctoral fellow at s...
2f321b48-8b27-4f72-afd6-57202f6fc652
trentmkelly/LessWrong-43k
LessWrong
Defending One-Dimensional Ethics Previously, I introduced the idea of "future-proof ethics." Ethics based on "common sense" and the conventions of the time has a horrible track record; future-proof ethics is about trying to make ethical decisions that we can remain proud of in the future, after a great deal of (societ...
6e59f845-294b-43ab-b71c-0ee87d189117
trentmkelly/LessWrong-43k
LessWrong
Chapter 58: TSPE, Constrained Cognition, Pt 8 A/N: A movie trailer for Army of Darkness, resembling the one Harry saw, is THV1KkPXIxQ on YouTube. The key quote is as follows, spoken by a man of modern times to listeners from the Middle Ages: "All right you primitive screwheads! Listen up! You see this? This... is my...
02facbcd-6070-40e1-a9ab-01ac58dfad2d
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Promising posts on AF that have fallen through the cracks Yesterday on the [Weekly Alignment Research Coffee Time](https://www.alignmentforum.org/posts/cysgh8zpmvt56f6Qw/event-weekly-alignment-research-coffee-time-01-03) call as people were sharing updates on their recent work, Vanessa Kosoy brought up her and Diffrac...
fec77222-ef41-4f2d-a71e-c3b635a646c2
StampyAI/alignment-research-dataset/alignmentforum
Alignment Forum
Reward is not Necessary: How to Create a Compositional Self-Preserving Agent for Life-Long Learning There is an idea in the AI safety community that [expected utility maximisation is the "limit" of agency](https://www.lesswrong.com/posts/XYDsYSbBjqgPAgcoQ/why-the-focus-on-expected-utility-maximisers?commentId=mgdud4Z6...