id stringlengths 36 36 | source stringclasses 15
values | formatted_source stringclasses 13
values | text stringlengths 2 7.55M |
|---|---|---|---|
f2ea077b-374e-4fc1-bbb4-b54308ca9e10 | trentmkelly/LessWrong-43k | LessWrong | Yes to Feature Requests, No to Entitlement
In No Free Features, Alan Gibson writes:
> I look forward to a day when asking anyone to do unpaid labor is considered unethical by our industry. That goes for feature requests on open source projects, on unpaid internships, and on unpaid 'take home' interview assignments.
>... |
45f1566d-d667-40a3-9e9a-215ed5347f6b | StampyAI/alignment-research-dataset/eaforum | Effective Altruism Forum | Predictions for future AI governance?
Humans have some history of banning dangerous techniques, like: human cloning, GMO food, CRISPR on human.(I'm thinking up of others). Also, I guess most people(non-EAs) will think AGI risks are as serious as biorisks. There has been some news(ex:Italy banning ChatGPT, Musk called ... |
f9a4dc1a-3aca-4d9f-8880-f6fcf9210df3 | trentmkelly/LessWrong-43k | LessWrong | Open thread, September 18 - September 24, 2017
IF IT'S WORTH SAYING, BUT NOT WORTH ITS OWN POST, 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; re... |
0829c72f-0380-46ed-b35d-1e64af9d6216 | trentmkelly/LessWrong-43k | LessWrong | You can just spontaneously call people you haven't met in years
Here's a recent conversation I had with a friend:
> Me: "I wish I had more friends. You guys are great, but I only get to hang out with you like once or twice a week. It's painful being holed up in my house the entire rest of the time."
>
> Friend: "You... |
8a183eb8-9257-48aa-8c36-9b44ed6217e1 | trentmkelly/LessWrong-43k | LessWrong | My (current) model of what an AI governance researcher does
Purpose of this post: We need more people working as AI governance researchers. This post establishes a basic framework for thinking about what this career path entails, and how you might test your fit. I hope to receive feedback on this framework.
Epistemic... |
1116ae95-8f79-4c85-ac33-b747989ea066 | trentmkelly/LessWrong-43k | LessWrong | Foma: Beliefs that Cause Themselves to be True
tl;dr: Sometimes it seems like in order to accomplish something, you need to hold a particular belief. However, the effect of your beliefs on what you accomplish can be screened off from what you actually do.
Also, thank you to Benquo for reading over a rough draft of th... |
d31a9a81-9a26-46af-b812-693e4fe0a9dd | trentmkelly/LessWrong-43k | LessWrong | Meetup : Pittsburgh - Past Wrongs
Discussion article for the meetup : Pittsburgh - Past Wrongs
WHEN: 14 June 2012 06:00:00PM (-0400)
WHERE: Eatunique, S. Craig street, Pittsburgh
What have you been most wrong about in the past? When have you changed your mind? Why were you wrong? How did you become right?
Bring a ... |
200cdb89-0092-4ca3-a97a-306df51621f3 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Weak HCH accesses EXP
*This post is a follow-up to my “[Alignment proposals and complexity classes](https://www.alignmentforum.org/posts/N64THGX7XNCqRtvPG/alignment-proposals-and-complexity-classes)” post. Thanks to Sam Eisenstat for helping with part of the proof here.*
[Previously](https://www.alignmentforum.org/p... |
19ab830f-aecb-45d3-a075-4a3e367e5747 | trentmkelly/LessWrong-43k | LessWrong | Possibilizing vs. actualizing
[Metadata: crossposted from https://tsvibt.blogspot.com/2022/12/possibilizing-vs-actualizing.html. First completed December 31, 2022. This essay is more like research notes than exposition, so context may be missing, the use of terms may change across essays, and the text might be revised... |
d2234977-be7d-4e1a-b793-e509aee487a2 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | [MLSN #2]: Adversarial Training
As part of a larger community building effort, I am writing a safety newsletter which is designed to cover empirical safety research and be palatable to the broader machine learning research community. You can [subscribe here](https://newsletter.mlsafety.org/) or follow the newsletter o... |
526e368e-9d61-4404-be84-98fea76b5a36 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Experimentally evaluating whether honesty generalizes
If we train our ML systems to answer questions honestly in cases where humans can check the answer, will they generalize to behave honestly on questions where we can’t check?
I think that we could learn a lot about this question by running experiments today. I th... |
975de5bb-4937-4924-a401-a790108260a9 | LDJnr/LessWrong-Amplify-Instruct | LessWrong | ""This plucked chicken has two legs and no feathers—therefore, by definition, it is a human!"
When people argue definitions, they usually start with some visible, known, or at least widely believed set of characteristics; then pull out a dictionary, and point out that these characteristics fit the dictionary definition... |
e3509eea-4176-40d8-ad24-8c9ccfe7ae68 | trentmkelly/LessWrong-43k | LessWrong | [link] Apostles' Creed = Tsuyoku Naritai???
Background: Apostles' Creed, Tsuyoku Naritai
Related to: A Parable on Obsolete Ideologies
Just something I thought I might add to the annals of cases where someone tries to re-interpret an old religious text to mean something more acceptable to the modern ear, in contradic... |
3c3408f4-83f5-4407-a0c1-2d16b1547cbe | trentmkelly/LessWrong-43k | LessWrong | Improving parliamentary information processing
Crossposted from Equilibria Club
You can see organizations as information processing units: they eat and process information after which they create all of the products and services which we see around us. Parliaments seem to be especially constrained in their processing... |
7792314d-ad0e-48b4-803f-ca6656212b0b | trentmkelly/LessWrong-43k | LessWrong | AISN #38: Supreme Court Decision Could Limit Federal Ability to Regulate AI
Plus, “Circuit Breakers” for AI systems, and updates on China’s AI industry
Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required.
Listen to the AI Safet... |
c0716c30-d1e8-402a-a4de-9ff2bd30c474 | trentmkelly/LessWrong-43k | LessWrong | The One and a Half Gemini
Previously: I hit send on The Third Gemini, and within half an hour DeepMind announced Gemini 1.5.
So this covers Gemini 1.5. One million tokens, and we are promised overall Gemini Advanced or GPT-4 levels of performance on Gemini Pro levels of compute.
This post does not cover the issues w... |
74ffc719-7093-4227-a1b0-84315afecd73 | trentmkelly/LessWrong-43k | LessWrong | Choosing a book on causality
I'm looking to get a better understanding on causality, in particular with relation to AI. Judea Pearl has written at least 3 books on the subject.
I have already read The book of why - this one wasn't technical enough. Should I read Causality: Models, Reasoning and Inference or Causal In... |
e9142ad9-53eb-452f-b069-225c3ab7cc31 | trentmkelly/LessWrong-43k | LessWrong | Personally mitigating existential risk
Instrumental rationalilty is important. Existential risk is important. The average person will do most good in this world by helping himself. While only few people or all people at once can avoid a war, everyone can increase his and his loved ones likelihood to survive. In this s... |
76cc9f8b-834a-439f-967d-39b559aec124 | StampyAI/alignment-research-dataset/arbital | Arbital | Glossary (Value Alignment Theory)
The parent page for the definitions of words that are given a special or unusual meaning inside [value alignment theory](https://arbital.com/p/2v).
If a word or phrase could be mistaken for ordinary English (e.g. 'value' or 'utility function'), then [you should create a glossary page... |
a6f52de8-a296-464d-8bab-1050fc4d341e | trentmkelly/LessWrong-43k | LessWrong | Systems Biology for self study
If I self study biology by reading Systems Biology will I miss anything important? In this context the failure I'm worried about is "ended up learning a specialized field of biology, and would have to spend extra time learning biology in order to pass standardized tests if I go to collag... |
d47860c2-5268-4b35-8390-9a01f6fab475 | trentmkelly/LessWrong-43k | LessWrong | “Unsupervised” translation as an (intent) alignment problem
Suppose that we want to translate between English and an alien language (Klingon). We have plenty of Klingon text, and separately we have plenty of English text, but it’s not matched up and there are no bilingual speakers.
We train GPT on a mix of English an... |
1d0ab850-ece2-4084-8425-cfceb50591f8 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | P₂B: Plan to P₂B Better
***tl;dr:** Most good plans involve taking steps to make better plans. Making better plans is **the** convergent instrumental goal, of which all familiar convergent instrumental goals are an instance. This is key to understanding what agency is and why it is powerful.*
Planning means using a w... |
4c59a8e6-8dbc-4582-bcf2-c00defeb6b51 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Institute for Assured Autonomy (IAA) newsletter
Just a very brief link to a resource that people may not know about:
[Johns Hopkins Institute for Assured Autonomy newsletter](https://us7.campaign-archive.com/home/?u=607247891352c7db7d0b19bfb&id=704293ffb4)
For anybody interested in nearer-term AI safety / assured au... |
b8d843df-7fa5-42fe-8808-dd3ce35dce50 | trentmkelly/LessWrong-43k | LessWrong | Is Reality Ugly?
Yesterday I talked about the cubes {1, 8, 27, 64, 125, ...} and how their first differences {7, 19, 37, 61, ...} might at first seem to lack an obvious pattern, but taking the second differences {12, 18, 24, ...} takes you down to the simply related level. Taking the third differences {6, 6, ...} bri... |
c23ff0bb-67bf-4e25-aac0-843cc173174a | awestover/filtering-for-misalignment | Redwood Research: Alek's Filtering Results | id: 0e98c624a2e88e1f
Why it's hard to make settings for high-stakes control research Redwood Research blog Subscribe Sign in Share this post Redwood Research blog Why it's hard to make settings for high-stakes control research Copy link Facebook Email Notes More Why it's hard to make settings for high-stakes control r... |
8d7d2c68-3b67-4801-8a10-53bda89b7fc7 | awestover/filtering-for-misalignment | Redwood Research: Alek's Filtering Results | id: post2006
I hear a lot of different arguments floating around for exactly how mechanistically interpretability research will reduce x-risk. As an interpretability researcher, forming clearer thoughts on this is pretty important to me! As a preliminary step, I've compiled a list with a longlist of 19 different argum... |
0a4a0963-4019-472f-840f-a987cc0ed4a3 | trentmkelly/LessWrong-43k | LessWrong | Best videos inspiring first interest in rationality or the singularity
When faced with a decision that might be really important — if say, the life of a loved one may be at risk — many people, though unfortunately not all, are moved to a sense of responsibility whereby they suddenly care more about being right than ab... |
87a308c7-d79f-4bc7-a9fe-cc5a943869d3 | trentmkelly/LessWrong-43k | LessWrong | What is your personal experience with "having a meaningful life"?
I hear a lot of different stories about how meaning should fit into one's life
> "What's all this meaning bullshit? Just focus on doing your job well and providing for your family."
^my grandparents
> "Wanting meaning is wanting a simple narrative to... |
0d6d47d9-5863-4307-b180-2585d817dda9 | trentmkelly/LessWrong-43k | LessWrong | New SI publications design
The Singularity Institute's publications are badly and inconsistently formatted.
Our research associate Daniel Dewey has made us a nice-looking LaTeX template for them (example), but we need some help moving them from the original files into LaTeX. If you have a little experience with LaTeX... |
aade4142-27dd-47dc-8ae0-3c383cc57ad0 | StampyAI/alignment-research-dataset/eaforum | Effective Altruism Forum | Truthful AI
This post contains the abstract and executive summary of a new 96-page [paper](https://arxiv.org/abs/2110.06674) (subtitle: *Developing and governing AI that does not lie*) from authors at the Future of Humanity Institute, Global Priorities Institute, and OpenAI. Also [posted on LessWrong](https://www.less... |
496febf4-5b77-4c46-a3a3-388bb7fd27cb | trentmkelly/LessWrong-43k | LessWrong | Making it harder for an AGI to "trick" us, with STVs
Summary / Preamble
AI Alignment has various sub-areas. The area I focus on here is ways we might use a superintelligent AGI-system to help with creating an aligned AGI-system, even if the AGI we start out with isn’t fully aligned.
Imagine a superintelligence that “... |
a02ffe36-34a7-46c1-ac96-6be2ad12da4e | trentmkelly/LessWrong-43k | LessWrong | LessWrong Poll on AGI
This is just a short post to promote this Pol.is poll to loosely aggregate opinions about AGI / AGI x-risk of the LessWrong community. This poll is less structured than a questionnaire and anyone can add statements that others can agree/disagree with. Pol.is then automatically generates a nice vi... |
4b0d649e-d755-4547-9d81-c0c8d3539657 | StampyAI/alignment-research-dataset/blogs | Blogs | AI Risk and the Security Mindset
In 2008, security expert Bruce Schneier wrote about [the security mindset](http://www.schneier.com/blog/archives/2008/03/the_security_mi_1.html):
> Security requires a particular mindset. Security professionals… see the world differently. They can’t walk into a store without noticin... |
a3fad655-4423-41e2-aba3-da166e1ce548 | trentmkelly/LessWrong-43k | LessWrong | Counterfactual Induction
,
One of the first attempts at defining counterfactuals was the conjecture that ϕ is a counterfactual consequence of ψ when there is a short proof of ψ→ϕ.
This post will present a variant of logical induction that implements these proof-length-based counterfactuals, so there are well-defined ... |
9682afd2-07d3-4617-a459-e0a9ad7e0369 | trentmkelly/LessWrong-43k | LessWrong | Therapy and materialistic accounts of behavior
I used to see a therapist. Every week they would give me actions to take in the coming week. For instance, fill out a sheet about panic attacks immediately after any panic attacks that I had. Or, do a mindfulness exercise about once per day, at a time when I was anxious. ... |
363e48ea-eb8c-42dd-9683-9991a894f4cd | trentmkelly/LessWrong-43k | LessWrong | What's going on with AI progress and trends? (As of 5/2025)
AI progress is driven by improved algorithms and additional compute for training runs. Understanding what is going on with these trends and how they are currently driving progress is helpful for understanding the future of AI. In this post, I'll share a wide ... |
c13ec778-6587-4955-8c10-7e0b2a71d629 | trentmkelly/LessWrong-43k | LessWrong | Humans can be assigned any values whatsoever...
Crossposted at LessWrong 2.0.
Humans have no values... nor do any agent. Unless you make strong assumptions about their rationality. And depending on those assumptions, you get humans to have any values.
An agent with no clear preferences
There are three buttons in... |
0dceffd7-9476-453e-a384-872c16ccf273 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Anti-squatted AI x-risk domains index
I heard about a recent case of a relevant domain being squatted, and figured that the people who buy up domains then sell them at a markup might start to notice the growth of interest in AI x-risk and claim some domains which would otherwise host good projects. I've bought the fol... |
f5a819a2-a454-4cf9-9a12-32946bfec16d | trentmkelly/LessWrong-43k | LessWrong | Gas and Water
In general I think more people would be better off in the direction of preparing more for disasters. But here's a case where it didn't work out very well.
In some cases it would be helpful to have extra gasoline. Maybe something happens and you need to leave quickly, but your car is low on gas. Or there... |
bc116482-43b0-4266-a8ba-448b90589513 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | A comparison of causal scrubbing, causal abstractions, and related methods
**Summary:** We explain the similarities and differences between three recent approaches to testing interpretability hypotheses:[causal scrubbing](https://www.lesswrong.com/posts/JvZhhzycHu2Yd57RN/causal-scrubbing-a-method-for-rigorously-testin... |
ca5c63d7-33b6-416f-b4f8-1c23a3f4e9ed | StampyAI/alignment-research-dataset/blogs | Blogs | Transparency in Safety-Critical Systems
In this post, I aim to summarize one common view on AI transparency and AI reliability. It’s difficult to identify the field’s “consensus” on AI transparency and reliability, so instead I will present *a* common view so that I can use it to introduce a number of complications an... |
7b5f84c0-d9d9-4420-a161-33503235b5a7 | trentmkelly/LessWrong-43k | LessWrong | Value extrapolation, concept extrapolation, model splintering
Post written with Rebecca Gorman.
We've written before that model splintering, as we called it then, was a problem with almost all AI safety approaches.
There's a converse to this: solving the problem would help with almost all AI safety approaches. But s... |
098750cf-2e83-4064-9fe6-3607c63e57da | trentmkelly/LessWrong-43k | LessWrong | Training Regime Day 18: Negative Visualization
Introduction
There is an old riddle: "The inscription makes you sad when you're happy and happy when you're sad. What does it say?"
this too shall pass
Someone once told me that stoicism can be summarized as "don't worry about the stuff you can't change." Negative Visu... |
312eff77-65fa-40f0-a4d2-b300c82628d8 | trentmkelly/LessWrong-43k | LessWrong | [Link] [Fiction] Ted Chiang: The Lifecycle of Software Objects
http://www.subterraneanpress.com/index.php/magazine/fall-2010/fiction-the-lifecycle-of-software-objects-by-ted-chiang/
A well-written novella about a company that has a proto-AI engine they use to make teachable virtual pets for sale in a cyberspace envir... |
669b2a10-a363-4a38-86a8-4d498ac71d10 | trentmkelly/LessWrong-43k | LessWrong | Do Deep Neural Networks Have Brain-like Representations?: A Summary of Disagreements
Acknowledgments: Thanks to SPAR summer 2024 and the Promisingness of Automating AI Alignment group, particularly the supervisor for this project, Bogdan-Ionut Cirstea, for his feedback.
This post began as a write up of my review ... |
7164b07d-fba5-448d-9cbe-700a678cf901 | trentmkelly/LessWrong-43k | LessWrong | AGI is easier than robotaxis
[Epistemic status: Hot take I wrote in 1 hour. We'll see in the comments how well it holds up.]
[Update: OK, seems like people hate this. I guess it was probably too hot and sloppy in retrospect. I should have framed it as "People seem to have a prior that AGI is a grand challenge that's ... |
ac5d7f44-5b14-4e6c-8e0a-41f591e5e36c | StampyAI/alignment-research-dataset/blogs | Blogs | problems for formal alignment
problems for formal alignment
-----------------------------
here is a list of problems which i seek to either resolve or get around, in order to [implement](clarifying-formal-alignment-implementation.html) my [formal alignment](formal-alignment.html) plans, especially [QACI](qaci.html):... |
9d6c518c-b508-4d65-b601-9164ffd12909 | trentmkelly/LessWrong-43k | LessWrong | Reducing global AI competition through the Commerce Control List and Immigration reform: a dual-pronged approach
Summary: This essay proposes specific proposals for consolidating AI development within the US in order to ensure US has the power to unilaterally slow down AI without interference from race dynamics. It of... |
46142135-9eaa-4794-8693-577bdbf35126 | trentmkelly/LessWrong-43k | LessWrong | Paper Highlights, March '25
tl;dr
Paper of the month:
Alignment auditing tests methods for detecting hidden objectives in AI systems. Sparse autoencoders, data analysis, and prefills are most effective at finding hidden goals.
Research highlights:
* Circuit tracing extracts interpretable computational graphs from ... |
c3209c06-87a0-4df3-a57d-a0d88dddb02e | trentmkelly/LessWrong-43k | LessWrong | Seeking beta readers who are ignorant of biology but knowledgeable about AI safety
Seeking the biology-ignorant! Come and do your part!
I am working on a post with Beth Barnes that will explain meiotic drive to AI safety people so they can determine if there are any useful analogies to gradient hacking. If you don't ... |
1caf9dae-8c2f-4d91-b72a-68387f195dd6 | trentmkelly/LessWrong-43k | LessWrong | Eliezer passes 100,000 karma points
Eliezer Yudkowsky has passed an arbitrary milestone: 100,000 karma points on Less Wrong. Allow me just a moment to celebrate this like we celebrate other arbitrary milestones, like birthdays.
I think that Eliezer's karma score vastly under-rates his relative contribution to this si... |
5bf800c5-93ee-4295-909e-71c243e43a5f | StampyAI/alignment-research-dataset/eaforum | Effective Altruism Forum | The Rival AI Deployment Problem: a Pre-deployment Agreement as the least-bad response
Introduction: The rival AI deployment problem
=============================================
Imagine an actor is faced with highly convincing evidence that with high probability (over 75%) a rival actor will be capable within two yea... |
6e76a1dc-dbdf-4b83-af2a-36652851050a | trentmkelly/LessWrong-43k | LessWrong | Starting a Rationalist Meetup during Lockdown
Authors: fkarg (my blog), crtltab, wilm
----------------------------------------
Backstory and Intentions
In October 2020, wilm moved to Karlsruhe and was disappointed by the lack of a local LessWrong community. In order to change this, he reached out to fkarg and toget... |
7cf5667e-3289-4d64-b44c-a417837d4a6f | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Finite Factored Sets
This is the edited transcript of a talk introducing finite factored sets. For most readers, it will probably be the best starting point for learning about factored sets.
Video:
(Lightly edited) slides: <https://intelligence.org/files/Factored-Set-Slides.pdf>
1. Short Combinatorics Talk
---... |
6c165de3-5aa3-43bf-b616-31a45bb05510 | trentmkelly/LessWrong-43k | LessWrong | Podcast: Shoshannah Tekofsky on skilling up in AI safety, visiting Berkeley, and developing novel research ideas
TLDR: I interviewed Shoshannah Tekofsky. You can listen to our conversation here or read some highlights below.
Some people in AI safety think that it’s important to have people coming up with fresh, new i... |
0f8cf32d-fb5b-4f52-9383-fc3326b0aa1f | trentmkelly/LessWrong-43k | LessWrong | Tool ideology
Follow-up to Journal article about politics and mindkilling. That post showed that people can be convinced that a view is correct by being told that their political party endorses it, even if their party actually opposes it. A similar, but stranger, effect, is that people can be convinced that a view i... |
af39e6fe-4728-423c-bc3a-f82f033556b1 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Microsoft and Google using LLMs for Cybersecurity
[Google](https://cloud.google.com/blog/products/identity-security/rsa-google-cloud-security-ai-workbench-generative-ai) and [Microsoft](https://blogs.microsoft.com/blog/2023/03/28/introducing-microsoft-security-copilot-empowering-defenders-at-the-speed-of-ai/) are inte... |
e78a37cd-fb45-4ef4-9496-2110e078da7a | trentmkelly/LessWrong-43k | LessWrong | Launching applications for AI Safety Careers Course India 2024
Announcing open applications for the AI Safety Careers Course India 2024!
Axiom Futures has launched its flagship AI Safety Careers Course 2024 to equip emerging talent working on India with foundational knowledge in AI safety. Spread out across 8-10 week... |
b1550a40-f922-4ecb-a55a-b5d27fcebc23 | trentmkelly/LessWrong-43k | LessWrong | From Conceptual Spaces to Quantum Concepts: Formalising and Learning Structured Conceptual Models
Submitted on 6 Nov 2023
Sean Tull, Razin A. Shaikh, Sara Sabrina Zemljic, Stephen Clark
Abstract
> In this article we present a new modelling framework for structured concepts using a category-theoretic generalisation ... |
4a575f2b-7f6e-4785-b2a2-116126d81060 | trentmkelly/LessWrong-43k | LessWrong | Status model
[Epistemic status: My best guess.]
Following a conversation on a previous post, I decided to do some research into the community’s thoughts on status. The community talks about status a lot and I spent a few happy hours sifting through everything I could find.
I came up with a status model based on the ... |
1aa078f5-85a7-4be3-b2d6-27e58c6cffde | trentmkelly/LessWrong-43k | LessWrong | Longtermist Consequences of a New Dark Age?
Inspired by a talk I had with James Faville.
Suppose that, sometime this decade, designer superbugs end up collapsing civilization as we know it, but not wiping out the entire species. Things are catastrophic, but enough people possess immunity to any particular act of biot... |
b899c187-1585-486a-9545-b04151a84be7 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Reframing inner alignment
The standard frame ([Evan Hubinger, 2021](https://www.alignmentforum.org/posts/FDJnZt8Ks2djouQTZ/how-do-we-become-confident-in-the-safety-of-a-machine)) is:
> * **Outer alignment** refers to the problem of finding a loss/reward function such that the training goal of “a model that optimizes... |
f4bbcdc2-d020-47ff-9adb-26460a840666 | trentmkelly/LessWrong-43k | LessWrong | Weekly LW Meetups
This summary was posted to LW Main on February 24th. The following week's summary is here.
The following meetups take place in cities with regular scheduling, but involve a change in time or location, special meeting content, or simply a helpful reminder about the meetup:
* Chicago Rationality Rea... |
cc9d2240-6ca9-4dd0-b362-bee6d941c590 | StampyAI/alignment-research-dataset/blogs | Blogs | December 2017 Newsletter
[Our annual fundraiser is live](https://intelligence.org/2017/12/01/miris-2017-fundraiser/). Discussed in the fundraiser post:
* [News](https://intelligence.org/2017/12/01/miris-2017-fundraiser/#1) — What MIRI’s researchers have been working on lately, and more.
* [Goals](https://intelligen... |
04c9272d-1365-4431-9e6a-5a2551b608d0 | StampyAI/alignment-research-dataset/blogs | Blogs | trading with superintelligence: a wonky proto-alignment scheme
trading with superintelligence: a wonky proto-alignment scheme
--------------------------------------------------------------
if we had as long as we want to figure out AI alignment, then we wouldn't worry as much — the problem is that [timelines are sho... |
aa34d2bb-e861-4dda-9bcb-98aa48be971a | trentmkelly/LessWrong-43k | LessWrong | Why economics is not a morality tale
Example nicked from this online Berkeley lecture.
Monopolies are bad (morality and economics agree here).
Firms that pollute are bad (morality and economics agree here).
What about monopolies that pollute?
What about strong monopolies that pollute and receive government subs... |
b92e3545-063b-425d-8f72-32d280806bf2 | trentmkelly/LessWrong-43k | LessWrong | The Bright Side of rationalization
Usually, people talk about rationalization ("the act, process, or result of rationalizing : a way of describing, interpreting, or explaining something (such as bad behavior) that makes it seem proper, more attractive, etc" https://www.merriam-webster.com/dictionary/rationalization) i... |
80dcb361-4f4f-4a61-9697-fa02e10350af | trentmkelly/LessWrong-43k | LessWrong | LW should go into mainstream academia ?
I feel that a lot of what's in LW (written by Eliezer or others) should be in mainstream academia. Not necessarily the most controversial views (the insistence on the MW hypothesis, cryonics, the FAI ...), but a lot of the work on overcoming biases should be there, be criticized... |
7e725263-2241-43ed-8621-6368f922acb2 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Aligned AI as a wrapper around an LLM
In my previous post [Are extrapolation-based AIs alignable?](https://www.lesswrong.com/posts/xQKHgEq9YrvzKiABA/are-extrapolation-based-ais-alignable) I argued that an AI trained only to extrapolate some dataset (like an LLM) can't really be aligned, because it wouldn't know what i... |
382f4c76-7a3d-405b-b0ab-108e5cab9e5f | trentmkelly/LessWrong-43k | LessWrong | GPT-3 + GAN
I am wondering if anyone tried to combine GPT-3 with GAN, basically trying to train network that would feed GPT-3 questions and then judge responses as correct/incorrect, thus providing GPT-3 opportunity to improve.
Does my question even make sense or I am far off base? |
dc570ecf-ff04-4a4e-bf09-b9a6c11bc679 | trentmkelly/LessWrong-43k | LessWrong | Refine: what helped me write more?
(previously: All the posts I will never write Refine Blogpost Day #3: The shortforms I did write )
Things that helped my writing
Key levers were (1) writing shortforms instead of posts (2) writing drafts on LW and letting them marinate for a while instead of writing everything in... |
06dcb163-556f-4eeb-8c06-fdaf8bd2dd06 | trentmkelly/LessWrong-43k | LessWrong | Open Communication in the Days of Malicious Online Actors
This article uses a thought experiment to probe at dangers of public online communication, then describes an ontology around “malicious supporters” and possible remedies.
It's arguably a more obvious fit for LW than the EA Forum, but I posted it there for reas... |
525f81f0-d995-496a-9db3-5d6e01115fbc | trentmkelly/LessWrong-43k | LessWrong | "Search" is dead. What is the new paradigm?
For over two decades internet search (which has been largely synonymous with Google for most of that time) has been the main way to augment the human brain with the digital information outside of it. First the virtual assistants and now LLMs are challenging and changing the ... |
7cb2885f-80ad-4549-9aa0-bcd695efe09f | trentmkelly/LessWrong-43k | LessWrong | Minneapolis Meetup: Saturday May 14, 3:00PM
Saturday May 14th, 3PMStarbucksCoffman Memorial Union300 Washington Avenue SE,University of Minnesota, Minneapolis, MN 55455
Goals:
To meet and network with likeminded Minnesotans.
I'll do my best to answer any questions people have about SIAI, rationality, x-risk str... |
80257403-88b4-4820-9661-93a7b5caacfe | trentmkelly/LessWrong-43k | LessWrong | Rationality Quotes from people associated with LessWrong
The other rationality quotes thread operates under the rule:
> Do not quote from Less Wrong itself, Overcoming Bias, or HPMoR.
Lately it seems that every MIRI or CFAR employee is excempt from being quoted.
As there are still interesting quotes that happen on ... |
85cbeb1c-6176-4b93-b4c7-6fe4dbce6ca4 | trentmkelly/LessWrong-43k | LessWrong | Predictable updating about AI risk
(Cross-posted from my website. Podcast version here, or search "Joe Carlsmith Audio" on your podcast app.)
> "This present moment used to be the unimaginable future."
>
> - Stewart Brand
1. Introduction
Here’s a pattern you may have noticed. A new frontier AI, like GPT-4, gets re... |
e03706cb-2670-4fff-8b15-243840483e06 | trentmkelly/LessWrong-43k | LessWrong | "LLMs Don't Have a Coherent Model of the World" - What it Means, Why it Matters
There are a variety of problems with LLMs, and I want to argue that they are all somewhat related, and are related to the idea of having a model of the world. This seems useful both as a conceptual model for thinking about why AI is unsaf... |
06fe7b42-9e9c-4db5-8608-a2ca0fea4bb0 | trentmkelly/LessWrong-43k | LessWrong | Rationality Toughness Tests
(Epistemic) rationality has two major components:
* Smarts: An ability to, by attending, infer truth from info under ideal circumstances.
* Toughness: An ability to limit performance degradation as circumstances worsen.
Attending takes time, energy, quiet, etc. Circumstances where huma... |
a449c8af-4631-427e-aff1-4dbad279d5f2 | trentmkelly/LessWrong-43k | LessWrong | Meetup: Philadelphia, August 11, games
August 11, 1 PM, Giovanni Pizza at 1515 Chestnut. We'll be playing games, probably with an emphasis on games where making up rules is part of the play.
We have a google group and an interactive calendar.
|
24bba69f-d0f1-46fc-9c71-c692bbe9b6dd | trentmkelly/LessWrong-43k | LessWrong | Mind Projection Fallacy
In the dawn days of science fiction, alien invaders would occasionally kidnap a girl in a torn dress and carry her off for intended ravishing, as lovingly depicted on many ancient magazine covers. Oddly enough, the aliens never go after men in torn shirts.
Would a non-humanoid alien, with a d... |
99acc046-ea8f-430a-ac93-d0a1efae222e | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Oracles, Informers, and Controllers
*Epistemic Status: I think the axes are straightforward enough. I'm not particularly sure about the names, happy to hear other sorts of suggestions.*
*Scholarly Status: Push/pull systems have been discussed before. I'm sure there's other naming for this sort of thing in the enginee... |
812dc32e-3af8-4652-8e65-cf575ee6edf4 | trentmkelly/LessWrong-43k | LessWrong | Text Posts from the Kids Group: 2019
Another round of liberating kid posts from Facebook. For reference, in 2019 Lily turned 5 and Anna turned 3.
(Some of these were from me; some were from Julia. Ones saying "me" could mean either of us.)
2019-01-05
Anna: I was kicking Lily over and I tripped on a balloon.
Me: Y... |
3c2009e5-bfee-4a64-a32a-810397f95096 | StampyAI/alignment-research-dataset/blogs | Blogs | Singularity Fun Theory
*This page is now obsoleted by the [Fun Theory Sequence](http://lesswrong.com/lw/xy/the_fun_theory_sequence/) on [Less Wrong](http://lesswrong.com/) .*
------------------------------------------------------------------------------------------------------------------------------------------------... |
d2650ed7-2b86-40d2-9d33-4865eea8352e | StampyAI/alignment-research-dataset/arxiv | Arxiv | GALOIS: Boosting Deep Reinforcement Learning via Generalizable Logic Synthesis
1 Introduction
---------------
Deep reinforcement learning (DRL) has achieved great breakthroughs in various domains like robotics control Polydoros and Nalpantidis ([2017](#bib.bib1 "Survey of model-based reinforcement learning: applica... |
c71a7308-2eae-460f-a547-5a008fbbd1c4 | StampyAI/alignment-research-dataset/arbital | Arbital | The empiricist-theorist false dichotomy
No discussion here yet: See https://www.facebook.com/groups/674486385982694/permalink/784664101631588/ |
de4b9787-7070-43b9-9ab3-541d71024ebf | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Brief Notes on Transformers
*These are just some notes I wrote while reading about transformers which I thought might be a useful reference to others. Thanks to Aryan Bhatt for a correction to the attention normalization. Further corrections welcome.*
Overview of Transformers
========================
Many transforme... |
4c6a5241-e427-4dc4-9d03-1b5989c38ad9 | trentmkelly/LessWrong-43k | LessWrong | Thoughts on Dancing the Whole Dance: Positional Calling for Contra
After yesterday's post on my experience dancing to positional calling, Louise Siddons recommended I read her booklet, Dancing the Whole Dance: Positional Calling for Contra. Overall I like the it: it gives a good explanation of how to call positionally... |
a6f64eba-e586-4de7-a209-0224a12a9eb8 | trentmkelly/LessWrong-43k | LessWrong | A few analogies to illustrate key rationality points
Introduction
Due to long inferential distances it's often very difficult to use knowledge or understanding given by rationality in a discussion with someone who isn't versed in the Art (like, a poor folk who didn't read the Sequences, or maybe even not the Goedel, ... |
1bcf545c-9ce0-4075-99c6-e4eecd6cc254 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Report on modeling evidential cooperation in large worlds
I have published a [report](https://arxiv.org/pdf/2307.04879.pdf) on modeling [evidential cooperation in large worlds](https://longtermrisk.org/msr). I originally started working on the report during a CEA Summer Research Fellowship back in 2018 and have now fi... |
7ba4d5bc-09f5-47cd-a40b-08adae5a9faa | trentmkelly/LessWrong-43k | LessWrong | Post ridiculous munchkin ideas!
Thus spake Eliezer:
> A Munchkin is the sort of person who, faced with a role-playing game, reads through the rulebooks over and over until he finds a way to combine three innocuous-seeming magical items into a cycle of infinite wish spells. Or who, in real life, composes a surprising... |
c7cc53fd-0582-4217-bd62-df941cde1d3d | trentmkelly/LessWrong-43k | LessWrong | How To Use Bureaucracies
This is an excerpt from the draft of my upcoming book on great founder theory. It was originally published on SamoBurja.com. You can access the original here.
When we encounter unsavory features of reality, it can be tempting to look away. Instead, we should ask, “What purpose does this serve... |
6bfff3dd-be9c-4f7c-8e93-9278a0d08cc6 | trentmkelly/LessWrong-43k | LessWrong | Curious Inquiry and Rigorous Training
Epistemic status: personal opinion and theorizing, based on recent conversations and explorations of Anki and the use cases of flashcards for autodidacticism in STEM. Mainly of use to continue directing my own Curious Inquiry, and hoping others find it a valuable framework as well... |
920eed49-d918-4da0-b47a-bb845f0773b8 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | Transformer Architecture Choice for Resisting Prompt Injection and Jail-Breaking Attacks
It seems very likely to me that decoder-only transformer architectures such as GPT may be particularly vulnerable to prompt injection attacks, since these only have a single text context which is shared between the prompt, any blo... |
c03a67db-20ed-4223-ae45-ac81c2db4025 | trentmkelly/LessWrong-43k | LessWrong | Rationality Is Not Systematized Winning
> "Rationality is systematized winning" is a slogan that was adopted to patch a bug in human cognition. Namely our endless capacity to delude ourselves about how we did in an attempt to save face. The concept seems to have been absorbed, but I'm skeptical it's translated into mo... |
bcdc4538-023e-4b6c-badf-e0e55baee69f | trentmkelly/LessWrong-43k | LessWrong | Book review: Xenosystems
I've met a few Landians over the last couple years, and they generally recommend that I start with reading Nick Land's (now defunct) Xenosystems blog, or Xenosystems, a Passage Publishing book that compiles posts from the blog. While I've read some of Fanged Noumena in the past, I would agree ... |
f6dab7a2-adbd-473c-9c1a-8e7c859ba3b0 | trentmkelly/LessWrong-43k | LessWrong | Conditions for Mesa-Optimization
This is the second of five posts in the Risks from Learned Optimization Sequence based on the paper “Risks from Learned Optimization in Advanced Machine Learning Systems” by Evan Hubinger, Chris van Merwijk, Vladimir Mikulik, Joar Skalse, and Scott Garrabrant. Each post in the sequence... |
4d4cb1f9-eb26-4729-8a72-c79a402af503 | StampyAI/alignment-research-dataset/eaforum | Effective Altruism Forum | Ngo and Yudkowsky on scientific reasoning and pivotal acts
This is a transcript of a conversation between Richard Ngo and Eliezer Yudkowsky, facilitated by Nate Soares (and with some comments from Carl Shulman). This transcript continues the [Late 2021 MIRI Conversations](https://intelligence.org/late-2021-miri-conver... |
59504c93-91f8-444f-95a2-15ef10d6fd6b | trentmkelly/LessWrong-43k | LessWrong | China Covid #3
China has once again explicitly recommitted to its goal of Zero Covid.
Even the WHO has said this policy is unsustainable. China don’t care.
Or rather, Xi doesn’t know and who is going to tell him?
Premiere Li might be trying, but it’s not working.
Xi, whose name keeps coming up headlines during h... |
f8575554-c0fd-4f7d-a309-8559046cecb5 | StampyAI/alignment-research-dataset/arxiv | Arxiv | DisCor: Corrective Feedback in Reinforcement Learning via Distribution Correction
###
1 Introduction
Reinforcement learning (RL) algorithms, when combined with high-capacity deep neural net function approximators, have shown promise in domains ranging from robotic manipulation [[1](#bib.bib1)] to recommender syste... |
2a84fdde-3d76-4a10-b80c-a4769a18ed9d | trentmkelly/LessWrong-43k | LessWrong | Conversation with Paul Christiano
AI Impacts talked to AI safety researcher Paul Christiano about his views on AI risk. With his permission, we have transcribed this interview.
Participants
* Paul Christiano — OpenAI safety team
* Asya Bergal – AI Impacts
* Ronny Fernandez – AI Impacts
* Robert Long – AI Impacts... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.