id stringlengths 36 36 | source stringclasses 15
values | formatted_source stringclasses 13
values | text stringlengths 2 7.55M |
|---|---|---|---|
86b10cf9-89eb-4168-b0bc-e9534703009c | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Sparse Autoencoders Find Highly Interpretable Directions in Language Models
*This is a linkpost for* [*Sparse Autoencoders Find Highly Interpretable Directions in Language Models*](https://arxiv.org/abs/2309.08600)
We use a *scalable* and *unsupervised* method called Sparse Autoencoders to find *interpretable*, *mono... |
d2cb73d4-3d54-4797-b9df-9b6a4b66aca5 | trentmkelly/LessWrong-43k | LessWrong | Link: Chessboxing could help train automatic emotion regulation
EDIT: Argh, I really failed to read this closely. Rewriting...
Just saw this over at Not Exactly Rocket Science. Chessboxing (or similar games) could help train automatic emotion regulation. Obviously this should generalize. Has this - by which I mean ... |
e540b596-9b15-4796-878f-4c9d37219ece | trentmkelly/LessWrong-43k | LessWrong | All AGI Safety questions welcome (especially basic ones) [~monthly thread]
tl;dr: Ask questions about AGI Safety as comments on this post, including ones you might otherwise worry seem dumb!
Asking beginner-level questions can be intimidating, but everyone starts out not knowing anything. If we want more people in th... |
e3aef336-60f9-4cb8-a9fa-56ecf81ec05e | StampyAI/alignment-research-dataset/special_docs | Other | AI Research with the Potential for Malicious Use: Publication Norms and Governance Considerations
June, 2021
Shanghai Institute for Science of Science
Website:www.aigovernancereview.com
The report editor can be reached at
globalaigovernance@gmail.comWe welcome any comments on this reportand any communication related t... |
03da40c8-4641-4c65-8822-c1976be715d9 | trentmkelly/LessWrong-43k | LessWrong | Where are the big questions?
There is no widely agreed upon list of the most important intellectual questions. Nor even a well-known attempted list, upon which people disagree. As far as I can tell there is hardly even such a list for any specific field. I don’t know of even a single person who has such a list. The cl... |
a698b97c-429f-4b1d-9bc8-f251f87094e6 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Instrumental Convergence To Offer Hope?
*TL;DR Could a superintelligence be motivated to preserve lower life forms because it fears an even greater superintelligence? This 'motivation' takes the form of something like timeless decision theory.*
This is a thought experiment about a superintelligent AI. No doubt the si... |
98379b29-5dff-4023-b20f-ee1b331e1e90 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | A test for symbol grounding methods: true zero-sum games
.mjx-chtml {display: inline-block; line-height: 0; text-indent: 0; text-align: left; text-transform: none; font-style: normal; font-weight: normal; font-size: 100%; font-size-adjust: none; letter-spacing: normal; word-wrap: normal; word-spacing: normal; white-sp... |
7866e3e3-1a76-4275-b265-245f7f2b8917 | trentmkelly/LessWrong-43k | LessWrong | Mech Interp Puzzle 2: Word2Vec Style Embeddings
Code can be found here. No prior knowledge of mech interp or language models is required to engage with this.
Language model embeddings are basically a massive lookup table. The model "knows" a vocabulary of 50,000 tokens, and each one has a separate learned embedding v... |
89b03b88-183c-4026-bd23-7a580ae9eac9 | trentmkelly/LessWrong-43k | LessWrong | Elicitation for Modeling Transformative AI Risks
This post is part 8 in our sequence on Modeling Transformative AI Risk. We are building a model to understand debates around existential risks from advanced AI. The model is made with Analytica software, and consists of nodes (representing key hypotheses and cruxes) and... |
781d67a7-5419-4f01-9ca7-b15fe2be9ab7 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Conditions for Mesa-Optimization
.mjx-chtml {display: inline-block; line-height: 0; text-indent: 0; text-align: left; text-transform: none; font-style: normal; font-weight: normal; font-size: 100%; font-size-adjust: none; letter-spacing: normal; word-wrap: normal; word-spacing: normal; white-space: nowrap; float: none... |
06eec5ad-8c00-473f-af08-8823fbfb6768 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Are language models close to the superhuman level in philosophy?
An alternative way to ask the question in the title of this post: what capacities do philosophers use that language models lack?
Reading OpenAI’s “[Self-critiquing models for assisting human evaluators](https://arxiv.org/pdf/2206.05802.pdf)” and Google’... |
403d9585-a5a0-488b-a6a7-bd4f592621f3 | trentmkelly/LessWrong-43k | LessWrong | Fermi Estimates
Just before the Trinity test, Enrico Fermi decided he wanted a rough estimate of the blast's power before the diagnostic data came in. So he dropped some pieces of paper from his hand as the blast wave passed him, and used this to estimate that the blast was equivalent to 10 kilotons of TNT. His guess ... |
bc621150-8ee8-4d48-9a87-790031c44746 | trentmkelly/LessWrong-43k | LessWrong | Beyond the moment of invention
Derek Thompson has a feature in The Atlantic this week on “Why the Age of American Progress Ended”—thoughtful and worth reading. Some comments and reactions.
One of the ideas explored in the article is that what matters for progress is not just the moment of invention, but what happens ... |
fa59d5c2-5462-4c52-9768-bc1b8403b9c7 | StampyAI/alignment-research-dataset/special_docs | Other | How feasible is the rapid development of artificial superintelligence?
Physica Scripta
INVITED COMMENT
How feasible is the rapid development of artificial superintelligence?
To cite this article: Kaj Sotala 2017 Phys. Scr. 92 113001
View the article online for updates and enhancements.
This content was downl... |
42e42d79-868a-4389-8cd1-b33954dda5e8 | trentmkelly/LessWrong-43k | LessWrong | Learn Bayes Nets!
It recently occurred to me that there are a lot of people in the rationalist community who want to deeply absorb intuitions about how Bayes' theorem works and how to think with it in practice, who have not been specifically told that learning inference algorithms for Bayesian networks is one of the b... |
f48d2d1b-1148-4d0d-9505-b9c00fec00ed | trentmkelly/LessWrong-43k | LessWrong | Upskilling, bridge-building, research on security/cryptography and AI safety
[This part 5 of a 5 part sequence on security and cryptography areas relevant for AI safety, published and linked here a few days apart.]
Summary
In this sequence, I highlighted a few relevant areas ripe for exploration to apply security an... |
f2d6d180-9f97-4631-89e4-fc27dd0a7c65 | trentmkelly/LessWrong-43k | LessWrong | Meetup : Washington DC Epistemic Spring Cleaning Meetup
Discussion article for the meetup : Washington DC Epistemic Spring Cleaning Meetup
WHEN: 29 September 2013 03:00:00PM (-0400)
WHERE: National Portrait Gallery, Washington, DC 20001, USA
Hi everyone,
We'll be meeting to find uncover cached beliefs, and then re... |
7d2ce826-fbaf-4663-8b8b-005c138c6626 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Specification gaming examples in AI
[Cross-posted from [personal blog](https://vkrakovna.wordpress.com/2018/04/02/specification-gaming-examples-in-ai/)]
Various examples (and [lists](https://www.gwern.net/Tanks#alternative-examples) [of](https://www.alexirpan.com/2018/02/14/rl-hard.html) [examples](https://arxiv.org/... |
cc5e31f2-0bbe-4ec0-af01-330a4e7c56cd | trentmkelly/LessWrong-43k | LessWrong | The Fusion Power Generator Scenario
Suppose, a few years from now, I prompt GPT-N to design a cheap, simple fusion power generator - something I could build in my garage and use to power my house. GPT-N succeeds. I build the fusion power generator, find that it works exactly as advertised, share the plans online, and ... |
3905b643-ad87-4e6d-8339-7725e441ab64 | trentmkelly/LessWrong-43k | LessWrong | Protect the seemingly useless
Advice I could have done with as a teenager, from G.K. Chesterton:
> In the matter of reforming things, as distinct from deforming them, there is one plain and simple principle; a principle which will probably be called a paradox. There exists in such a case a certain institution or law;... |
19ea4ddb-056a-41af-844f-df4ad9b176a1 | StampyAI/alignment-research-dataset/eaforum | Effective Altruism Forum | What are the top priorities in a slow-takeoff, multipolar world?
The model of AI risk I’ve mostly believed for years was that of fast-takeoff and therefore a unipolar world.[[1]](#fn-duaAoawcPwEieBXRo-1) This model allowed me to have some concrete models of what the EA community should do to make AI go better. Now, I ... |
9d32dcac-1b0e-48ff-b677-b4426fcdb185 | trentmkelly/LessWrong-43k | LessWrong | Is school good or bad?
Should society eliminate schools? Should we have more compulsory schooling? Should you send your kids to school? Should you prefer to hire job candidates who have received more schooling, beyond school's correlation with the g factor? Should we consider the spread of education requirements to be... |
a0c51b7e-0033-4143-b696-3c8d10f83964 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Four Motivations for Learning Normativity
I have been pretty satisfied with my [desiderata for learning normativity](https://www.lesswrong.com/s/Gmc7vtnpyKZRHWdt5/p/2JGu9yxiJkoGdQR4s#Summary_of_Desiderata), but I *haven't* been very satisfied with my explanation of why exactly these desiderata are important. I have a ... |
151ee422-e3e3-446b-beb3-e870e9281902 | trentmkelly/LessWrong-43k | LessWrong | The Onrushing Wave
There's a long article in this week's The Economist:
The onrushing wave
discussing the effect of changing technology upon the amount of employment available in different sectors of the economy.
Sample paragraph from it:
> The case for a highly disruptive period of economic growth is made by Erik... |
2a6984b9-5cf1-44af-be63-ae673751329a | StampyAI/alignment-research-dataset/special_docs | Other | Collective Action on Artificial Intelligence: A Primer and Review | Global Catastrophic Risk Institute
[View the paper “Collective Action on Artificial Intelligence: A Primer and Review”](https://gcrinstitute.org/papers/058\_c... |
3bf09464-d287-4134-bc4e-667e29224755 | trentmkelly/LessWrong-43k | LessWrong | Epistemic Spot Check: The Demon Under the Microscope (Thomas Hager)
Description
How much would it suck to be the guy who invented sulfa drugs? You dedicate your life to preventing a repeat of the horrors you saw in the war, succeed in that and so much more, and then 10 years later some idiot leaves a petri dish open a... |
656c28fb-153e-4bc3-b698-fb1b8347c4f1 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | We have promising alignment plans with low taxes
*Epistemic status: I’m sure these plans have advantages relative to other plans. I'm not sure they're adequate to actually work, but I think they might be.*
With good enough alignment plans, we might not need coordination to survive. If alignment taxes are low enough, ... |
91e8adbd-7921-472f-8fc3-b33fab8f237e | trentmkelly/LessWrong-43k | LessWrong | Malthusian copying: mass death of unhappy life-loving uploads
Robin Hanson has done a great job of describing the future world and economy, under the assumption that easily copied "uploads" (whole brain emulations), and the standard laws of economics continue to apply. To oversimplify the conclusion:
* There will be... |
78f69ce6-4e91-4019-b4c3-cc71b79c075f | trentmkelly/LessWrong-43k | LessWrong | AiPhone
Apple was for a while rumored to be planning launch for iPhone of AI assisted emails, texts, summaries and so on including via Siri, to be announced at WWDC 24.
It’s happening. Apple’s keynote announced the anticipated partnership with OpenAI.
The bottom line is that this is Siri as the AI assistant with ful... |
f6f13aa5-6fd0-4074-9b9d-60effb732377 | trentmkelly/LessWrong-43k | LessWrong | A First Attempt to Dissolve "Is Consciousness Reducible?"
Introduction
I recently made an attempt to dissolve the question: "is consciousness reducible"?
(This was borne out of a discussion with Rob Bensinger on Twitter.)
I've repackaged the thread for LessWrong.
Concepts Hierarchy
I see three relevant/particula... |
31c24dba-8576-4460-85ba-2dc207677e4a | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | [Interim research report] Taking features out of superposition with sparse autoencoders
*We're thankful for helpful comments from Trenton Bricken, Eric Winsor, Noa Nabeshima, and Sid Black.*
*This post is part of the work done at*[*Conjecture*](https://www.conjecture.dev/research)*.*
**TL;DR**: Recent results fr... |
5bacca1f-46ab-42f6-9047-d1ea214c19aa | LDJnr/LessWrong-Amplify-Instruct | LessWrong | "This is the introduction to a sequence on signing up for cryonics. In the coming posts I will lay out what you need to do, concretely and in detail. This sequence is intended for people who already think signing up for cryonics is a good idea but are putting it off because they're not sure what they actually need to d... |
33f553eb-91e7-43aa-a54e-9cb1dd9d1bd1 | trentmkelly/LessWrong-43k | LessWrong | Anyone working on D. Amodei's Bartlett show transcript?
(Don't know how to better coordinate on two lesswrongians not duplicating a future post, so I'll just put up this question) |
a069eed0-40ba-4eff-a184-e074c912559d | trentmkelly/LessWrong-43k | LessWrong | The Assassination of Trump's Ear is Evidence for Time-Travel
Epistemic Status: extreme, unhinged speculation. Possibly a small Basilisk-like effect from reading this that directionally seems neutral-to-positively aligned to the cause of human survival. Hopefully entertaining. Big if true.
Over the weekend Trump was v... |
65974220-3e60-4fd7-813f-8e1ab581d5d9 | trentmkelly/LessWrong-43k | LessWrong | The Book of The Machines (Chapter 8)
Part II : The Machines
In Part I of our saga, we visited with many of our orienting themes. We visited with several 20th Century thinkers, that, I think, helped set the stage for our current collective imaginaries about our lives, our societies, and our relationships with our ma... |
4e156d7a-8bd8-431d-91d9-7a925b2cef6c | trentmkelly/LessWrong-43k | LessWrong | Zen and Rationality: Just This Is It
This is post 4/? about the intersection of my decades of LW-style rationality practice and my several years of Zen practice.
In today's installment, I look at "just this is it" from a rationalist perspective.
When Dongshan, the co-founder of what would become the Soto Zen school ... |
edddc85b-6631-4181-8576-5ccefdfc2f1a | trentmkelly/LessWrong-43k | LessWrong | How to choose a country/city?
EDIT: I've found a very relevant indicator for my question, see "Quality of life" criteria below.
My main question is: which non-academic factors should I consider when moving to another country/city for a PhD? Further, I would also like to evaluate each country/city1 according to those... |
c5b9d825-13fe-4533-b61f-8b7bed4545d4 | trentmkelly/LessWrong-43k | LessWrong | Foom & Doom 2: Technical alignment is hard
2.1 Summary & Table of contents
This is the second of a two-post series on foom (previous post) and doom (this post).
The last post talked about how I expect future AI to be different from present AI. This post will argue that, absent some future conceptual breakthrough, thi... |
ebaee6e3-a6ce-4fc6-b6dc-c10e7dfb9827 | trentmkelly/LessWrong-43k | LessWrong | SIA Is Just Being a Bayesian About the Fact That One Exists
Anthopics is often dense and technical, so I’ll begin this by presenting a captivating story. I was chatting with someone and she said “is SSA the better theory of anthropics?” So I launched into a lengthy digression of roughly the following form.
The Self ... |
1563c798-6737-4a75-a4dd-b78a7cb0d51f | trentmkelly/LessWrong-43k | LessWrong | Trust in Bayes
Followup to: Beautiful Probability, Trust in Math
In Trust in Math, I presented an algebraic proof that 1 = 2, which turned out to be - surprise surprise - flawed. Trusting that algebra, correctly used, will not carry you to an absurd result, is not a matter of blind faith. When we see apparent evid... |
fb94d7cf-b180-4357-88b6-a5c323644c13 | trentmkelly/LessWrong-43k | LessWrong | Analysis of transistion costs for changing habits
http://logicalwholesomehealth.blogspot.com/2011/05/how-easy-is-it-to-change-your-diet-or.html#comment-form
The article is from the angle of not blaming people who haven't made changes you think should be easy, but it also looks like a good way of allowing yourself app... |
6207eed7-7cae-47bc-b793-1e31e0fe892c | StampyAI/alignment-research-dataset/arbital | Arbital | Real-world domain
A 'domain', in Artificial Intelligence or machine learning, is a problem class on which an AI program can be targeted. Face recognition is a 'domain', as is playing chess, or driving a car. Each of these domains presents characteristic options and modes of action for accomplishing the task: a Go ga... |
a3a0bbd6-d825-4eb1-a9ab-7c5b659579f8 | trentmkelly/LessWrong-43k | LessWrong | Effective Effective Altruism Fundraising and Movement-Building
The title of this post isn't a typo—its purpose is to ask how we can effectively do fundraising and movement-building for the effective altruism movement. This is an important question, because the return on these activities is potentially very high. As Ro... |
d7b463cd-9fdb-4c1a-8e54-f189c8d2fc06 | StampyAI/alignment-research-dataset/arxiv | Arxiv | Simplifying Reward Design through Divide-and-Conquer
I Introduction
---------------
While significant advances have been made in planning and reinforcement learning for robots, these algorithms require access to a reward (or cost) function in order to be successful. Unfortunately, designing a good reward function b... |
884014d8-509b-49f2-8779-5e924b689846 | trentmkelly/LessWrong-43k | LessWrong | Apollo
Life is a response to being
----------------------------------------
Part II
Over the course of my twenties, I'd accumulated thousands of hastily jotted notes. Thoughts that had grown in the gaps between work and play I'd managed to save before lost off the cliff of forgetting.
Thoughts like, "I'm starting ... |
33368f6c-a5da-4073-9767-7962be5a7c55 | LDJnr/LessWrong-Amplify-Instruct | LessWrong | "So you're thinking, "April 1st... isn't that already supposed to be April Fool's Day?"
Yes—and that will provide the ideal cover for celebrating Amazing Breakthrough Day.
As I argued in "The Beauty of Settled Science", it is a major problem that media coverage of science focuses only on breaking news. Breaking news, i... |
808f8239-defc-493b-972a-236b811a7a20 | StampyAI/alignment-research-dataset/youtube | Youtube Transcripts | Owain Evans | Truthful language models and AI alignment
uh hi thank you all for joining us today
my name is Roger gross and I'm an
associate professor of computer science
at the University of Toronto a member of
the technical staff at anthropic a
founding member of the vector Institute
and the reason I'm here A Facult... |
27742b3d-31ee-4387-89fd-35f404751e19 | trentmkelly/LessWrong-43k | LessWrong | Sam Altman Q&A Notes - Aftermath
To me it was quite painful when I found out after the first one of Sam Altman's Q&As that there was no recording and no way for me to find out what had been said. Which is why this time around I took notes and shared them here to help out people in the same situation.
This was initial... |
adb472db-af4a-40bf-bba6-f7a76d3599de | trentmkelly/LessWrong-43k | LessWrong | Why are all these domains called from Less Wrong?
When I visit a Less Wrong page, the browser also attempts to load content from the following domains:
* algolia.net
* algolianet.com
* cloudflare.com
* cloudinary.com
* dl.drop
* dropbox.com
* dropboxusercontent.com
* google.com
* googleapis.com
* googletagmanager.com... |
36f4755a-c481-456b-9d37-167b4df2859b | trentmkelly/LessWrong-43k | LessWrong | Instrumental Rationality is a Chimera
Eliezer observes, “Among all self-identified "rationalist" communities that I know of, and Less Wrong in particular, there is an obvious gender imbalance - a male/female ratio tilted strongly toward males.” and provides us with a selection of hypotheses that attempt to explain thi... |
c3db0040-6ebd-4405-9e7a-774e3320483b | trentmkelly/LessWrong-43k | LessWrong | My Expected Value Approach to Newcomb's Problem
Related to: Newcomb's Problem and Regret of Rationality and Newcomb's Problem: A Problem for Casual Decision Theories.
From the list of standard problems in rationality that have been talked to death but still don't have a strong consensus, allow me to re-present ... |
a6d527be-7eed-4f1e-99d6-87497716fdbd | trentmkelly/LessWrong-43k | LessWrong | Meetup : FallacyMania game in Kocherga club
Discussion article for the meetup : FallacyMania game in Kocherga club
WHEN: 25 November 2015 07:30:00PM (+0300)
WHERE: Moscow, B.Dorogomilovskaya, 5-2
Welcome to FallacyMania: it is a game where you guess logical fallacies in arguments, or practise using logical fallacie... |
181ab4ca-ce6c-4757-8e4f-c7c0b12a1b38 | trentmkelly/LessWrong-43k | LessWrong | AI box question
I believe that even through a text-only terminal, a superintelligence could do anything to a human. Persuade the human to let it out, inflict extreme pleasure or suffering with a word. However, can't you just... limit the output of the superintelligence? Just make it so that the human can say anything,... |
8f55b082-eb60-4cc4-83d4-3b7821f56d57 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Neural Tangent Kernel Distillation
*Produced As Part Of The SERI ML Alignment Theory Scholars Program 2022 Under* [*John Wentworth*](https://www.lesswrong.com/users/johnswentworth)*.*
Introduction
============
Consider the following example from [Goal Misgeneralization](https://arxiv.org/abs/2105.14111): you train a... |
f1046380-d077-4b22-8ed1-ce5c74ec1897 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | When to use quantilization
In 2015, Jessica introduced [quantilization](https://intelligence.org/files/QuantilizersSaferAlternative.pdf) as a countermeasure for [Goodhart's Law](https://arxiv.org/pdf/1803.04585.pdf) and specification-gaming. Since these are such central problems in AI safety, I consider quantilization... |
f91b3328-81cd-4be2-9a9b-4d26f8c5e3f0 | trentmkelly/LessWrong-43k | LessWrong | Book Review: On the Edge: The Gamblers
Previously: Book Review: On the Edge: The Fundamentals
As I said in the Introduction, I loved this part of the book. Let’s get to it.
POKER AND GAME THEORY
When people talk about game theory, they mostly talk solving for the equilibrium, and how to play your best game or strat... |
ef4eb8a0-011a-4168-b4c3-c9190a21e351 | trentmkelly/LessWrong-43k | LessWrong | [AN #56] Should ML researchers stop running experiments before making hypotheses?
Find all Alignment Newsletter resources here. In particular, you can sign up, or look through this spreadsheet of all summaries that have ever been in the newsletter.
Highlights
HARK Side of Deep Learning -- From Grad Student Descent t... |
6a566b46-cc04-4325-9a87-e94e9e2b3016 | trentmkelly/LessWrong-43k | LessWrong | Progress Report 2
Prev: Report 1
Next: Report 3
In looking at interpretability in transformers (1), I've been looking at the Causal Tracing notebook here: https://rome.baulab.info/ (2)
Expanding on this code somewhat, I made an attempt to increase the 'resolution' of the trace from layer-specific down to neuron-spe... |
6280b7b3-f874-4a58-af3a-d496cac43be1 | trentmkelly/LessWrong-43k | LessWrong | Potential Alignment mental tool:
Keeping track of the types
Epistemic status: Came up with what I was thinking of as a notation. Realised that there was a potentially useful conceptual tool behind it.
Consider this diagram.
1) A diagram showing a useless alignment scheme (ignore the computer, make the human do ever... |
93a1a627-9452-4a58-a7b8-ae57e60b4293 | trentmkelly/LessWrong-43k | LessWrong | Meetup : London meetup
Discussion article for the meetup : London meetup
WHEN: 24 May 2015 02:00:00PM (+0100)
WHERE: Lincoln's Inn Fields, London
The next LW London meetup will be on May 10th. Join us from 2pm to talk about the sorts of things that your other friends will look funny at you for talking about.
If th... |
bcfe0c77-1733-40b5-8b74-fdfed9c95a89 | trentmkelly/LessWrong-43k | LessWrong | GPT as an “Intelligence Forklift.”
[See my post with Edelman on AI takeover and Aaronson on AI scenarios. This is rough, with various fine print, caveats, and other discussions missing. Cross-posted on Windows on Theory.]
One challenge for considering the implications of “artificial intelligence,” especially of the... |
8b4b0f11-a165-450a-b4f9-737ee79f757e | StampyAI/alignment-research-dataset/aisafety.info | AI Safety Info | What is Conjecture's research strategy?
[Conjecture](https://www.conjecture.dev/) is an AI research lab focused on "[building Cognitive Emulation - an AI architecture that bounds systems' capabilities and makes them reason in ways that humans can understand and control](https://www.conjecture.dev/research/)". Conjectu... |
1ec0f5ba-5be3-4889-8e9e-df4753c3b941 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | What a compute-centric framework says about AI takeoff speeds
As part of my work for Open Philanthropy I’ve written a [draft report](https://www.openphilanthropy.org/research/what-a-compute-centric-framework-says-about-takeoff-speeds/) on AI takeoff speeds, the question of how quickly AI capabilities might improve as ... |
0dc818df-4abb-48e4-9128-f4e659c4cccf | trentmkelly/LessWrong-43k | LessWrong | Spoiler-Free Review: Death and Taxes
Death and Taxes is a short Tier 3 game, meaning it is a Good game but not universally Worth It or a Must Play. It is a series of discrete choices and has some interesting questions to ask, and I likely will use it as a reference point in the future once I’ve given people a chance t... |
754ac2f9-b9f1-4970-8ea4-adf1c8ace17f | trentmkelly/LessWrong-43k | LessWrong | Meetup : Moscow: paranoid calibaration, EA criticism, deliberate practice dojo
Discussion article for the meetup : Moscow: paranoid calibaration, EA criticism, deliberate practice dojo
WHEN: 24 April 2016 02:00:00PM (+0300)
WHERE: Москва, Большая Дорогомиловская, 5к2
Note: most our members join meetups via other ch... |
50c0040d-75c0-4b32-a702-6276ca630eaa | LDJnr/LessWrong-Amplify-Instruct | LessWrong | "Photo by Danielle Cerullo on UnsplashOver the course of nearly two decades in the workplace, I’ve seen the inside of dozens of organizations and teams as an employee, consultant, or friendly collaborator. With rare exceptions, it seems, important decisions get made one of two ways. If an organization is particularly h... |
84587016-5c0d-423a-a378-4b891b552849 | StampyAI/alignment-research-dataset/youtube | Youtube Transcripts | The Windfall Clause: Sharing the benefits of advanced AI | Cullen O’Keefe
so you're in the right room if you'd
like to hear about the windfall clause
sharing benefits from AI our speaker is
Cullen O'Keefe since he graduated from
Harvard Law School Cullen has been
working part-time as a research
affiliate with the cent... |
c21364c7-b89d-4235-badf-207baf5959f1 | trentmkelly/LessWrong-43k | LessWrong | Mini-review: The Book of Why
Someone should probably write a real book review, but to make a brief recommendation: The Book of Why by Judea Pearl and Dana Mackenzie is probably the most interesting general-science book I've read since Thinking Fast and Slow.
Pearl's goal is to explain and promote causal inference, wh... |
dd2a3cd2-863e-4f73-b435-797fd1bb31f5 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Formalizing Objections against Surrogate Goals
*I recently had a chance to look into surrogate goals and safe Pareto improvements, two recent proposals for avoiding the realization of threats in multiagent scenarios. In this report, I formalize some objections to these proposals.*
*EDIT: by discussing objections to S... |
f69a8ada-9f07-4e41-9d3f-0095e4973432 | trentmkelly/LessWrong-43k | LessWrong | H+ Summit Meetup Harvard 6/12
Just realized I hadn't seen a post about this, and couldn't find mention of it by searching...
The H+ Summit is happening this Saturday and Sunday at Harvard. I assume there are quite a few Humanity+ fellow-travelers here who might be going (though it's kindof late notice if this is the... |
014ab629-5cc6-47a9-9fe4-f1699a84ec48 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Can you get AGI from a Transformer?
***UPDATE IN 2023: I wrote this a long time ago and you should NOT assume that I still agree with all or even most of what I wrote here. I’m keeping it posted as-is for historical interest.***
Introduction
============
I want to share my thoughts about the calculations that Transf... |
6b115fb5-6e99-4438-bdf8-2e2c894bde9f | trentmkelly/LessWrong-43k | LessWrong | June 2015 Media Thread
This is the monthly thread for posting media of various types that you've found that you enjoy. Post what you're reading, listening to, watching, and your opinion of it. Post recommendations to blogs. Post whatever media you feel like discussing! To see previous recommendations, check out the ol... |
befe19a8-8bc2-4a4c-b7df-1d65bff12f01 | trentmkelly/LessWrong-43k | LessWrong | Oxford Prioritisation Project Review
Cross-posted to Tom's website.
Short summary
The Oxford Prioritisation Project was a research group between January and May 2017. The team conducted research to allocate £10,000 in the most impactful way, and published all our work on our blog. Tom Sittler was the Project’s dire... |
28e80d34-e5a3-4e0a-a4ea-57bebc731799 | awestover/filtering-for-misalignment | Redwood Research: Alek's Filtering Results | id: post60
Right now, alignment seems easy – but that’s because models spill the beans when they are misaligned. Eventually, models might “fake alignment,” and we don’t know how to detect that yet. It might seem like there’s a swarming research field improving white box detectors – a new paper about probes drops on ar... |
b2aa1104-7c13-4ca6-8db3-2f9e8f485639 | trentmkelly/LessWrong-43k | LessWrong | Do men have more partners than women?
Today I came across an article in the Telegraph that states that an average man has 9 sexual partners in his lifetime, whilst a woman has only 4.
Let's assume for a moment that in fact all these men and women are heterosexuals. In that case, each partnership contains one man and ... |
155ea740-9d91-4efc-81d8-9cfe30670e08 | trentmkelly/LessWrong-43k | LessWrong | What a practical plan for Friendly AI looks like
I have seen too many discussions of Friendly AI, here and elsewhere (e.g. in comments at Michael Anissimov's blog), detached from any concrete idea of how to do it. Sometimes the issue is the lack of code, demos, or a practical plan from SIAI. SIAI is seen as a source o... |
414cdf7d-0e14-4e8d-a263-5722b12d2558 | StampyAI/alignment-research-dataset/arbital | Arbital | Bulverism
[Bulverism](https://en.wikipedia.org/wiki/Bulverism) is when you [explain what goes so horribly wrong in people's minds](https://arbital.com/p/43h) when they believe X, before you've actually explained why X is wrong. It can be seen as a subspecies of the fallacy of "[poisoning the well](https://en.wikipedi... |
4d5301d1-72d0-4413-9178-700c0234ccb0 | StampyAI/alignment-research-dataset/arxiv | Arxiv | Approximate Causal Abstraction.
1 Introduction
---------------
Scientific models aim to provide a description of reality that offers
both an explanation of observed phenomena and
a basis for intervening on and manipulating the system to bring
about desired outcomes. Both of these aims lead to a consideration of
mod... |
36afc51c-73da-4b35-b78f-bb1c4a069140 | trentmkelly/LessWrong-43k | LessWrong | Looking Deeper at Deconfusion
Introduction
My PhD thesis probably wins the prize of weirdest ever defended at my research lab. Not only was it a work of theory of distributed computing in a formal methods lab, but it didn’t even conform to what theory of distributed computing is supposed to look like. With the excepti... |
36afadd6-f3e9-4458-bc6c-1a91bd7c0799 | trentmkelly/LessWrong-43k | LessWrong | Visualizing small Attention-only Transformers
Visualizing small Attention-only Transformers
Work done during an internship at MILES, Paris Dauphine University, under the supervision of Yann Chevaleyre. You can find the git page for the post here.
Research has indicated that in large Transformers, facts are primarily ... |
c7bb12d3-59f0-4262-9245-02d91e5f929a | trentmkelly/LessWrong-43k | LessWrong | Information-Theoretic Boxing of Superintelligences
Boxing an agent more intelligent than ourselves is daunting, but information theory, thermodynamics, and control theory provide us with tools that can fundamentally constrain agents independent of their intelligence. In particular, we may be able to contain an AI by l... |
288e4036-602b-486d-8004-bd12b8100daf | trentmkelly/LessWrong-43k | LessWrong | Impact Prizes as an alternative to Certificates of Impact
I posted this one to the EA Forum, because it seemed a bit more relevant there. I think that LessWrong users will also find it interesting.
TLDR Example
An EA donor puts up a $50k prize for distribution in 2022. In 2022, several projects that have started sin... |
3a3d88ac-86e7-4447-b18f-bced267c5dd9 | trentmkelly/LessWrong-43k | LessWrong | MIRIx Israel, Meeting summary
|
0b86a916-9759-44b8-9069-4533ddae1fa1 | trentmkelly/LessWrong-43k | LessWrong | [linkpost] Why Going to the Doctor Sucks (WaitButWhy)
Y'all heard about this already, right? |
efa224a5-1fce-4e58-83f4-75ced30404b4 | trentmkelly/LessWrong-43k | LessWrong | Dance Weekends: Tests not Masks
I'm really excited that contra dance weekends are starting up again, and I just got back from playing with Amy and Audrey for the first time in three years. I'm also seeing a lot of weekends struggling with registration, however, and choosing between going ahead with a small crowd or ca... |
762e461a-c331-49ee-b540-d7fdbbe7710c | trentmkelly/LessWrong-43k | LessWrong | Overcoming the negative signal of not attending college.
The signaling view of college holds that graduates of elite colleges earn high average salaries not because of what they learned in school but rather because top colleges select for students who have highly valued traits, the two most important probably being h... |
2766f25e-6ee5-4d14-af40-1cdaf9a40aef | StampyAI/alignment-research-dataset/blogs | Blogs | AI-written critiques help humans notice flaws
Self-critiquing models for assisting human evaluators
William SaundersCatherine YehJeff Wu
Steven Bills Long Ouyang Jonathan Ward Jan Leike
OpenAI
Abstract
We fine-tune large language models to write natural language critiques (natural
language critical comments) using b... |
0b113fef-a05b-4fd1-ab77-25db8e5b8c6d | trentmkelly/LessWrong-43k | LessWrong | ML Alignment Theory Program under Evan Hubinger
In the past six weeks, the Stanford Existential Risks Initiative (SERI) has been running a trial for the “ML Alignment Theory Scholars” (MATS) program. Our goal is to increase the number of people working on alignment theory, and to do this, we’re running a scholars prog... |
6acaf4a3-04e7-4ca8-938c-1a304e77a24b | trentmkelly/LessWrong-43k | LessWrong | The Steering Problem
Most AI research focuses on reproducing human abilities: to learn, infer, and reason; to perceive, plan, and predict. There is a complementary problem which (understandably) receives much less attention: if you had these abilities, what would you do with them?
The steering problem: Using black-bo... |
3d680fcc-3ab8-4e61-829e-36c7c5916d02 | trentmkelly/LessWrong-43k | LessWrong | Understanding Simpson's Paradox
An article by Judea Pearl, available here. It's quick at 8 pages, and worth reading if you enjoy statistics (though I think people who already are familiar with the math of causality1 will get more out of it than others2). I'll talk here about the part that I think is generally interest... |
34b9c945-dbee-40de-beec-cd84b90649ca | StampyAI/alignment-research-dataset/arbital | Arbital | Group isomorphism
A group isomorphism is a [https://arbital.com/p/-47t](https://arbital.com/p/-47t) which is [bijective](https://arbital.com/p/499).
We say that two groups are *isomorphic* if there is an isomorphism between them.
It turns out that isomorphism is a much more useful concept than true equality of [group... |
19b29b3c-7aeb-488d-ab41-69ee12d6f66f | trentmkelly/LessWrong-43k | LessWrong | self-improvement-executors are not goal-maximizers
Agents are adaptation-executors with adaptations that accomplish goals, not goal-maximizers. Understanding agents as maximizing goals is a simplification used by humans to make them easier to understand. This is as true when the goal is self-improvement as it is with ... |
a2dd59c2-c7dc-45cc-81ce-d4566292500c | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Some background for reasoning about dual-use alignment research
*This is pretty basic. But I still made a bunch of mistakes when writing this, so maybe it's worth writing. This is background to a specific case I'll put in the next post.*
It's like a a tech tree
=======================
If we're looking at the big pic... |
fc7748a1-127d-43a6-a93b-791cd4cb9be5 | trentmkelly/LessWrong-43k | LessWrong | When does introspection avoid the pitfalls of rumination?
A theme I see being explored in the LW-sphere is the importance of connecting to system 1 information in various ways.
There's lots of stuff about meditation, Focusing, connecting to sub-agents, the question "What would happen if [...] you introspected instead... |
450d0eb2-02fb-47b4-b912-a7de5dcb0a34 | trentmkelly/LessWrong-43k | LessWrong | Thinking Machines
Self understanding at a gears level
I think an AI which understands the source of its intelligence at a gears level, and self improves at the gears level, will be much better at keeping its future versions aligned to itself.
There's also more hope it'll keep its future versions aligned with humanity... |
9c752f5e-adf5-4541-8ce4-64b802b18915 | trentmkelly/LessWrong-43k | LessWrong | Is sitting in the sun much better than sitting in the shade?
How much worse is sitting outside in the shade compared to the sun, for the purpose of experiencing enough lux to feel alert?
Making the difference between shade and sun a bit more extreme than it probably is in practice, we can take "sun" to mean "10,000 ... |
f4aed68f-cc2a-47fd-a6f1-cfbb9d2d149b | trentmkelly/LessWrong-43k | LessWrong | [SEQ RERUN] Priming and Contamination
Today's post, Priming and Contamination was originally published on 10 October 2007. A summary (taken from the LW wiki):
> Contamination by Priming is a problem that relates to the process of implicitly introducing the facts in the attended data set. When you are primed with a... |
a7de349b-8328-4752-84d5-f993627d430f | trentmkelly/LessWrong-43k | LessWrong | Decomposing independent generalizations in neural networks via Hessian analysis
In our joint SERI MATS project, we came up with a series of equations and experiments to mechanistically understand and steer the generalization behavior of neural nets. The core conceit is to locate the circuits (which we call "modules") ... |
1ef0a360-e7f6-4d56-b9d1-7d97fc582308 | trentmkelly/LessWrong-43k | LessWrong | An illustrative model of backfire risks from pausing AI research
TLDR
The post contains a simple modelization of the maximum explosive takeoff speed (simplified as "takeoff speed"), using as the main parameters:
* overhangs[1] sizes (investment, hardware efficiency, software efficiency)
* overhangs consumption rat... |
893c0d37-e045-45e2-a687-9ff52f5ded3c | trentmkelly/LessWrong-43k | LessWrong | Why I write about the basics
In a thread on A Rationalist's Tale, lessdazed wrote:
> Being levels above in [rationality] means doing rationalist practice 101 much better than others, [just like] being a few levels above in fighting means executing a basic front-kick much better than others.
Eliezer replied:
> I reg... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.