text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Failure modes in a shard theory alignment plan
*Thanks to David Udell, Alex Turner, and others for conversations that led to this*
Recently in a conversation with David Udell, Nate Soares said:
> \[David's summary of shard theory\] doesn't yet convince me that you know something i don't about the hopefullness of s... | https://www.lesswrong.com/posts/xsieF8SXw4J5LkzEg/failure-modes-in-a-shard-theory-alignment-plan |
# 7 traps that (we think) new alignment researchers often fall into
We've noticed that new alignment researchers (and sometimes experienced alignment researchers) often fall into similar traps.
Here are 7 traps that we often see in new alignment researchers:
### 1\. They think they need to be up-to-date on the liter... | https://www.lesswrong.com/posts/h5CGM5qwivGk2f5T9/7-traps-that-we-think-new-alignment-researchers-often-fall |
# Threat-Resistant Bargaining Megapost: Introducing the ROSE Value
$\newcommand{\eps}{\epsilon}\newcommand{\x}{\times}\newcommand{\bol}[1]{\textbf{#1}}\newcommand{\sur}{\mathcal{S}}$Post Status: Original research, mathy, high-context. Make sure you've read the last three posts in the sequence first, this is going to b... | https://www.lesswrong.com/posts/vJ7ggyjuP4u2yHNcP/threat-resistant-bargaining-megapost-introducing-the-rose |
# [Linkpost] "Intensity and frequency of extreme novel epidemics" by Mariani et al. (2021)
This is a useful paper for those wanting to better understand the risks posed by pandemics and, more generally, global catastrophic biological risk.
I have not seen this paper mentioned on LW or EAF and it does not appear on LW... | https://www.lesswrong.com/posts/8Xn2BFviydmgEKgMC/linkpost-intensity-and-frequency-of-extreme-novel-epidemics |
# About Q Home
*(I didn't know how to name the post.)*
I have an internet friend in Ukraine. My love is in Belarus. I have a non-binary friend in Russia who is a former conscript. I'm a young male in Russia. I'm at risk, but might be relatively safe.
Writing this post to highlight my main ideas (I would like to dis... | https://www.lesswrong.com/posts/iGa8Ecrsdirjdn5aN/about-q-home |
# Why I think strong general AI is coming soon
I think there is little time left before someone builds AGI (median ~2030). Once upon a time, I didn't think this.
This post attempts to walk through some of the observations and insights that collapsed my estimates.
The core ideas are as follows:
1. We've already cap... | https://www.lesswrong.com/posts/K4urTDkBbtNuLivJx/why-i-think-strong-general-ai-is-coming-soon |
# Strange Loops - Self-Reference from Number Theory to AI
Summary
=======
I see this post as essentially trying to make clear the link between self-reference as discussed in Hofstadter’s “[Gödel, Escher, Bach](https://www.amazon.com/G%C3%B6del-Escher-Bach-Eternal-Golden/dp/0465026567)” and self-reference as considere... | https://www.lesswrong.com/posts/gvXAoH9gR4FSzyeCa/strange-loops-self-reference-from-number-theory-to-ai |
# QAPR 3: interpretability-guided training of neural nets
Introduction
------------
This is week 3 of Quintin's Alignment Papers Roundup. This week, I'm focusing on papers that use interpretability to guide a neural network's training process. A [lot of alignment proposals](https://www.lesswrong.com/posts/fRsjBseRuvR... | https://www.lesswrong.com/posts/rgh4tdNrQyJYXyNs8/qapr-3-interpretability-guided-training-of-neural-nets |
# Georgism in Space
[Georgism](https://en.wikipedia.org/wiki/Georgism) is cool and all, but it was invented in 1879, and I prefer my ideologies to be more futuristic. Unlike the [Network State](https://thenetworkstate.com/), Georgism doesn’t require blockchains, and unlike Scientology, it lacks aliens.
So lets make... | https://www.lesswrong.com/posts/oBoCNcrJNdsQwhBd7/georgism-in-space |
# Will Values and Competition Decouple?
There are a great many forces shaping the evolution of the universe. Among them, the values of agents -- systems which attempt to optimize, or steer the future towards certain configurations over others -- seem likely to have a dominant influence on the long-term future. The val... | https://www.lesswrong.com/posts/2XLFyhuKP7m4xpgea/will-values-and-competition-decouple-1 |
# AI Safety Endgame Stories
Assume you are in the set of possible worlds where [AI takeover](https://www.alignmentforum.org/posts/pRkFkzwKZ2zfa3R6H/without-specific-countermeasures-the-easiest-path-to) happens by default. If you do nothing, then at some point in the 21st century the AI lab Magma develops a transformat... | https://www.lesswrong.com/posts/AsNjqggQQ4yJcbsWn/ai-safety-endgame-stories |
# A Library and Tutorial for Factored Cognition with Language Models
We want to advance [process-based supervision for language models](https://www.lesswrong.com/posts/pYcFPMBtQveAjcSfH/supervise-process-not-outcomes). To make it easier for others to contribute to that goal, we're sharing code for writing compositiona... | https://www.lesswrong.com/posts/X5L9g4fXmhPdQrBCA/a-library-and-tutorial-for-factored-cognition-with-language |
# LOVE in a simbox is all you need
***L**earning **O**ther's **V**alues or **E**mpowerment in simulation sandboxes is all you need*
TL;DR: We can develop self-aligning DL based AGI by improving on the brain's dynamic alignment mechanisms (empathy/altruism/love) via safe test iteration in simulation sandboxes.
AGI i... | https://www.lesswrong.com/posts/WKGZBCYAbZ6WGsKHc/love-in-a-simbox-is-all-you-need |
# EA & LW Forums Weekly Summary (19 - 25 Sep 22')
*Supported by Rethink Priorities*
Also posted on the [EA forum.](https://forum.effectivealtruism.org/posts/tokGikSg3fSJun4Lw/ea-and-lw-forums-weekly-summary-19-25-sep-22)
This is part of a weekly series - you can see the full collection [here.](https://www.les... | https://www.lesswrong.com/posts/WdAazEFdouyyhhrFw/ea-and-lw-forums-weekly-summary-19-25-sep-22 |
# Progress links and tweets, 2022-09-28
**Announcements**
-----------------
* [“Vast is developing artificial gravity space stations to expand humanity across the solar system”](https://www.vast.space/)
* [Derek Thompson and Ezra Klein are writing a book about abundance](https://twitter.com/DKThomp/status/1571940... | https://www.lesswrong.com/posts/iXk6wKF27W6SgmxGj/progress-links-and-tweets-2022-09-28 |
# Estimating the Current and Future Number of AI Safety Researchers
Summary
=======
I estimate that there are about **300** full-time technical AI safety researchers, **100** full-time non-technical AI safety researchers, and **400** AI safety researchers in total today. I also show that the number of technical AI sa... | https://www.lesswrong.com/posts/mC3oeq62DWeqxiNBx/estimating-the-current-and-future-number-of-ai-safety |
# Petrov Day Retrospective: 2022
*This is a follow-up to the* [*Petrov Day commemoration of 2022*](https://www.lesswrong.com/posts/KTEciTeFwL2tTujZk/lw-petrov-day-2022-monday-9-26)
* * *
We died.
Twice.
Maybe three times.
* * *
*This is how it went down...*
**September, 21, 5:43 PM Pacific Time:** I [write the ... | https://www.lesswrong.com/posts/ytuLbHbdQweAwGk9L/petrov-day-retrospective-2022 |
# Better Construction Cost Estimates?
When I wrote about how Somerville's affordable housing "bonus density" rules made it likely [profitable to build 100% affordable housing](https://www.jefftk.com/p/come-build-affordable-housing), [Chris pointed out](https://www.facebook.com/jefftk/posts/pfbid0hDCioDSdoQTZHrq4415cXk... | https://www.lesswrong.com/posts/jDmqqJ2epCML75C7y/better-construction-cost-estimates |
# Reflection on a Consulting Workshop
Epistemic status: This is a summary / review / critique of a workshop I attended today. The workshop's materials are not publically available, so I don't expect any productive discussion regarding the workshop itself. This is mostly for myself to process the new information.
### ... | https://www.lesswrong.com/posts/kLKcraakKKrnsuQcs/reflection-on-a-consulting-workshop |
# Should reasonably healthy people still take Paxlovid?
Should a non-immunocompromised, fairly healthy person in their late 30s take Paxlovid if they get one of the currently circulating strains of Omicron?
How would that answer change if:
* Their symptoms are fairly mild (bad cold)
* They have a high enough BMI... | https://www.lesswrong.com/posts/MxxEStGrDB8ud8uyb/should-reasonably-healthy-people-still-take-paxlovid |
# Open application to become an AI safety project mentor
EffiSciences is a French organization committed to a more involved research to address the pressing problems of this century.
We are organizing bootcamps for French university students in AI safety. After teaching an intensive curriculum (PyTorch, ML, interpret... | https://www.lesswrong.com/posts/EAbuRJeaJbQLbMqf3/open-application-to-become-an-ai-safety-project-mentor |
# [Link] "Improper Nouns" by siderea
The post:
- [siderea | Improper Nouns [p/a/s, Patreon]](https://siderea.dreamwidth.org/1773806.html)
Interestingly, 'improper nouns' – as described by the post – is itself an example of the phenomena it describes (as 'improper noun' is itself an existing term):
> There's a thing... | https://www.lesswrong.com/posts/HsfFhXAeT6uNYDQgK/link-improper-nouns-by-siderea |
# FDT is not directly comparable to CDT and EDT
In [*Functional Decision Theory: A New Theory of Instrumental Rationality*](https://arxiv.org/pdf/1710.05060.pdf) and [*Cheating Death in Damascus*](https://intelligence.org/files/DeathInDamascus.pdf), the authors go through a multitude of decision problems and compare ... | https://www.lesswrong.com/posts/dmjvJwCjXWE2jFbRN/fdt-is-not-directly-comparable-to-cdt-and-edt |
# Resources to find/register the rationalists that specialize in a given topic?
As far as I can tell:
1. It's not possible for me (or most/all other people, for that matter) to individually evaluate claims every topic I hear about, as it takes too long time and too much effort to learn the foundations needed to eval... | https://www.lesswrong.com/posts/da2noz6x8JCRqGkwG/resources-to-find-register-the-rationalists-that-specialize |
# Builder/Breaker for Deconfusion
*This is something of a grab-bag of thoughts I've had about the Builder/Breaker game.*
[The ELK document](https://docs.google.com/document/d/1WwsnJQstPq91_Yh-Ch2XRL8H_EpsnjrC1dwZXR37PC8/edit) had a really nice explanation of its research methodology in terms of an imaginary dialogue ... | https://www.lesswrong.com/posts/yLTpo828duFQqPJfy/builder-breaker-for-deconfusion |
# Unit Test Everything
**Inspiration**
* [Specializing in Problems We Don't Understand](https://www.lesswrong.com/posts/CSZnj2YNMKGfsMbZA/specializing-in-problems-we-don-t-understand)
* [You Are Not Measuring What You Think You Are Measuring](https://www.lesswrong.com/posts/9kNxhKWvixtKW5anS/you-are-not-measuring... | https://www.lesswrong.com/posts/Kkqrx8hytBJCwhJo3/unit-test-everything |
# Covid 9/29/22: The Jones Act Waver
The pandemic is over. [Long live the Public Health Emergency.](https://sharylattkisson.com/2022/09/hhs-expected-to-renew-covid-19-pubic-health-emergency-for-11th-time/)
> HHS is set to extend the Covid-19 public health emergency by its standing deadline of Oct. 13.
>
> If renewe... | https://www.lesswrong.com/posts/YcqxizJpunJvMDbrj/covid-9-29-22-the-jones-act-waver |
# It matters when the first sharp left turn happens
*Thanks to Evan Hubinger for comments on these ideas.*
Introduction
============
A “sharp left turn” is a point where [capabilities generalize beyond alignment](https://www.alignmentforum.org/posts/GNhMPAWcfBCASy8e6/a-central-ai-alignment-problem-capabilities-gener... | https://www.lesswrong.com/posts/BCyK2GQKiiuYdvkst/it-matters-when-the-first-sharp-left-turn-happens |
# Where I currently disagree with Ryan Greenblatt’s version of the ELK approach
Context: This post is my attempt to make sense of Ryan Greenblatt's research agenda, as of April 2022. I understand Ryan to be heavily inspired by Paul Christiano, and Paul left some comments on early versions of these notes.
Two separate... | https://www.lesswrong.com/posts/LBzFCPbG5s95mf43M/where-i-currently-disagree-with-ryan-greenblatt-s-version-of |
# Clarifying the Agent-Like Structure Problem
A few months ago, Thomas Kwa put together [a statement](https://www.lesswrong.com/posts/voLHQgNncnjjgAPH7/utility-maximization-description-length-minimization?commentId=cPMXSrnnsbX9qLHEB) of a potential “holy grail” of [Selection Theorems](https://www.lesswrong.com/posts/G... | https://www.lesswrong.com/posts/moi3cFY2wpeKGu9TT/clarifying-the-agent-like-structure-problem |
# Distribution Shifts and The Importance of AI Safety
Preface
=======
The following text is my submission for the [AI Safety Public Materials contest](https://www.lesswrong.com/posts/gWM8cgZgZ9GQAYTqF/usd20k-in-bounties-for-ai-safety-public-materials). In it, I try to lay out the importance of AI Safety Research to p... | https://www.lesswrong.com/posts/TRKF9g65nhPBQoxJu/distribution-shifts-and-the-importance-of-ai-safety |
# Bruce Wayne and the Cost of Inaction
*My brother (DivineMango) and I (TurnTrout) wrote this first arc of a Batman fanfiction in the summer of 2020. *
1: Don't Look Away
------------------
“The man’s starving. That doesn’t change if you look away, Bruce.”
Hand wrapped firmly around his son’s shoulder, Thomas Wayne... | https://www.lesswrong.com/posts/9KF28JSTJBqqkmkJZ/bruce-wayne-and-the-cost-of-inaction |
# Prioritizing Parental Sleep
Overall I've really enjoyed being a parent, but poor sleep has been the hardest part. At times we've both been so tired that we weren't able to think clearly about how to fix the problem, which can be very tricky. We've figured this out more over time, however, and sleep has improved with... | https://www.lesswrong.com/posts/tmqoo5BekbmxyAuHp/prioritizing-parental-sleep |
# How To Observe Abstract Objects
**Opening Thoughts**
--------------------
**What is this thing and what is the point of it?**
I’m trying to build a branch of rationality that is about looking at ideas and problems “for real”, thinking about them “for real”, “as a whole person”, and “without all the bullshit in the... | https://www.lesswrong.com/posts/EgPE2xxKGYqkPfpnz/how-to-observe-abstract-objects |
# The Importance-Avoidance Effect
My current working theory is that one of the worst things someone with a big project can do is to make that project feel extremely important. (This is doubly worse for someone with ADHD.) The more important the project, the more daunting it becomes, and the less it gets worked on.
Wh... | https://www.lesswrong.com/posts/yQAdnY3WqjWGx8x6s/the-importance-avoidance-effect |
# (Structural) Stability of Coupled Optimizers
*This post is part of my* [*hypothesis subspace*](https://www.lesswrong.com/s/H3xEgE7bPGKvucfQk) *sequence, a living collection of proposals I'm exploring at* [*Refine.*](https://www.lesswrong.com/posts/5uiQkyKdejX3aEHLM/how-to-diversify-conceptual-alignment-the-model-beh... | https://www.lesswrong.com/posts/zgQSsA2o6avsEsyMa/structural-stability-of-coupled-optimizers |
# Eli's review of "Is power-seeking AI an existential risk?"
See also Joe Carlsmith's report ["Is power-seeking AI an existential risk"](https://www.lesswrong.com/posts/HduCjmXTBD4xYTegv/draft-report-on-existential-risk-from-power-seeking-ai) and [previous reviews](https://www.lesswrong.com/posts/qRSgHLb8yLXzDg4nf/rev... | https://www.lesswrong.com/posts/NLqAQzAhE9u87TvNz/eli-s-review-of-is-power-seeking-ai-an-existential-risk-1 |
# A Few Terrifying Facts About The Russo-Ukrainian War
*Epistemic status: trying to summarize the news and predict, post is under revision, too lazy to citation everything*
I wanted to collect a few observations I've made, as best I understand them. This [PBS article](https://www.pbs.org/newshour/world/how-real-are-p... | https://www.lesswrong.com/posts/gyMXuhcYyMRpwEjGq/a-few-terrifying-facts-about-the-russo-ukrainian-war |
# Announcing the AI Safety Nudge Competition to Help Beat Procrastination
**Have you achieved your goal? Fill in **[**this form**](https://docs.google.com/forms/d/e/1FAIpQLSeXq1S5Qwdozhkk9YUbIxez34OW29Hgi_unYZBuFDl1TUKgBQ/viewform?usp=sharing) **to be added to the prize draw!**
Have you been postponing doing alignmen... | https://www.lesswrong.com/posts/FWCqBJcmbQYRSymZL/announcing-the-ai-safety-nudge-competition-to-help-beat |
# Google could build a conscious AI in three months
_Summary: Theories of consciousness do not present significant technical hurdles to building conscious AI systems. Recent advances in AI relate to capacities that aren't obviously relevant to consciousness. Satisfying major theories with current technology has been a... | https://www.lesswrong.com/posts/Dq9LaL4KzLcvCWa5v/google-could-build-a-conscious-ai-in-three-months |
# What's the actual evidence that AI marketing tools are changing preferences in a way that makes them easier to predict?
I've encountered this claim multiple times over the years (most recently on [this AXRP episode](https://axrp.net/episode/2021/02/17/episode-4-risks-from-learned-optimization-evan-hubinger.html)), b... | https://www.lesswrong.com/posts/vwnb8TncgyK6ZwFyt/what-s-the-actual-evidence-that-ai-marketing-tools-are |
# What sorts of preparations ought I do in case of further escalation in Ukraine?
I guess the Russian invasion of Ukraine can go in many ways. It might just fizzle out, or continue to be a fairly conventional war with extent limited to Ukraine, or it might escalate with a Russian nuclear strike against Ukraine followe... | https://www.lesswrong.com/posts/kA3HhDKamTeCoEkbC/what-sorts-of-preparations-ought-i-do-in-case-of-further |
# Do anthropic considerations undercut the evolution anchor from the Bio Anchors report?
The [Bio Anchors report](https://www.lesswrong.com/posts/KrJfoZzpSDpnrv9va/draft-report-on-ai-timelines) by Ajeya Cotra tries to forecast AGI timelines by getting estimates of how much compute might be needed to build or train a m... | https://www.lesswrong.com/posts/NHvspuLiirJwiLtfg/do-anthropic-considerations-undercut-the-evolution-anchor |
# Are c-sections underrated?
**Summary**
I think so. Correlations tend to show worse outcomes for c-sections, but the guidelines take this evidence too literally. Careful causal evidence finds that it’s more of a mixed bag. Moderate negative effects on *subsequent* births seem more robust. An elective c-section migh... | https://www.lesswrong.com/posts/wgzPxFX4BoJoLJPaK/are-c-sections-underrated |
# How I Make Diagrams
Someone asked me how I make diagrams like the ones in my [two loft](https://www.jefftk.com/p/planning-a-loft-bed) [bed posts](https://www.jefftk.com/p/another-loft-bed), so here's how. Note, though, that these aren't perspective drawings: lines that are parallel in reality remain parallel on the ... | https://www.lesswrong.com/posts/bAtRGhezL6TFZBx3d/how-i-make-diagrams |
# Clarifying Your Principles
*Following up on:*
* [Integrity and Accountability are Core Parts of Rationality](https://www.lesswrong.com/posts/xhE4TriBSPywGuhqi/integrity-and-accountability-are-core-parts-of-rationality)
* [Robust Agency for People and Organizations](https://www.lesswrong.com/posts/3yXRgbsDu38AGH... | https://www.lesswrong.com/posts/5GCeFw6CLzgm7CEs4/clarifying-your-principles |
# Company leadership and tactical decisions
The other day at work, I received an email from our leadership regarding "new principles for our organization", linking out to a web page listing the new principles. This ended up being partially a culture document, but mostly a direction document - "we want customers to be... | https://www.lesswrong.com/posts/wY2MCuHqjd4giiEYr/company-leadership-and-tactical-decisions |
# The computational complexity of progress
*Cross-posted from* [*Telescopic Turnip*](https://malmesbury.substack.com/p/the-computational-complexity-of-progress?sd=pf)*.*
Artem Kaznatcheev has a [curious model about the evolution of living things](https://doi.org/10.1534%2Fgenetics.119.302000). When a population of cr... | https://www.lesswrong.com/posts/GCRm9ysNYuGF9mhKd/the-computational-complexity-of-progress |
# Four usages of "loss" in AI
*Summary:* What does it *mean* for a loss function to be "aligned with" human goals? I perceive four different concepts which involve "loss function" in importantly different ways:
1. *Physical-loss:* The physical implementation of a loss function and the loss computations,
2. *Mathem... | https://www.lesswrong.com/posts/jnmG5jczvWbeRPcvG/four-usages-of-loss-in-ai |
# Paper: Large Language Models Can Self-improve [Linkpost]
This paper came out about a week ago. I am not the author - it was published anonymously.
I learned about the paper when JJ Hepburn shared it on Slack. I just thought it seemed potentially really important and I hadn't seen it discussed on this forum yet:
* ... | https://www.lesswrong.com/posts/qwqowdhnMreKQvxLv/paper-large-language-models-can-self-improve-linkpost |
# Open & Welcome Thread - Oct 2022
If it’s worth saying, but not worth its own post, here's a place to put it.
If you are new to LessWrong, here's the place to introduce yourself. Personal stories, anecdotes, or just general comments on how you found us and what you hope to get from the site and community are invited... | https://www.lesswrong.com/posts/j9xxuJYiAiHtCBdeR/open-and-welcome-thread-oct-2022 |
# Monthly Shorts 9/22, and An Essay in Defense of Technodeterminism
Possibly all our fear over declining interstate migration is a …. [data artifact](https://www2.census.gov/ces/wp/2016/CES-WP-16-44R.pdf)? h/t [Evan Soltas](https://twitter.com/esoltas).
[
![Interstate migration rates as measures in several data sour... | https://www.lesswrong.com/posts/bF59t99pywtCnRzPj/monthly-shorts-9-22-and-an-essay-in-defense-of |
# Easy fixing Voting
\[I forgot to post this post during the elections, but still relevant\]
Context : Two round uninominal ballot
=====================================
At this moment in France, it's the election period. And between two existential anguishes about AGIs, I sometimes think about the presidential campa... | https://www.lesswrong.com/posts/aCCnkd4n7dLeziDHX/easy-fixing-voting |
# Against the weirdness heuristic
*Crossposted from the EA Forum:* [*https://forum.effectivealtruism.org/posts/DYmDGizxkcrZhmBCq/against-the-weirdness-heuristic*](https://forum.effectivealtruism.org/posts/DYmDGizxkcrZhmBCq/against-the-weirdness-heuristic)
Slightly more than a decade ago (approximately 2009), I was im... | https://www.lesswrong.com/posts/ZmKzbcx742mAy7xGt/against-the-weirdness-heuristic |
# What makes a probability question "well-defined"? (Part I)
**See also**: [Probability is in the Mind](https://www.lesswrong.com/posts/f6ZLxEWaankRZ2Crv/probability-is-in-the-mind), [You Are Never Entitled to Your Opinion](https://www.overcomingbias.com/2006/12/you_are_never_e.html), and [“I don’t know.”](https://www... | https://www.lesswrong.com/posts/TAtFRLmbNgZraBxcn/what-makes-a-probability-question-well-defined-part-i |
# Baby Monitor with Delay
Our youngest (15m) has recently started sleeping through the night on her own, without us needing to get up for feeding or settling. But she's still waking ( [one of](https://www.jefftk.com/p/baby-sleep-multiple-rooms)) us up, because she often cries a bit before falling back asleep. Turning ... | https://www.lesswrong.com/posts/NBpxwwzdFtFiupqJF/baby-monitor-with-delay |
# Is there a culture overhang?
Culture is adapted for learnability, transmissibility and recombinability for humans. To the extent AI and natural intelligence operate on similar principles, these adaptations should be expected to carry over when culture is used as a training dataset for AI. If so, an AI trained on cul... | https://www.lesswrong.com/posts/sjcQBQvassWqGEd5F/is-there-a-culture-overhang |
# Data for IRL: What is needed to learn human values?
*This post is a summary of my research. If you are interested in experime... | https://www.lesswrong.com/posts/2ZKLaqKLr8TkKAxRW/data-for-irl-what-is-needed-to-learn-human-values |
# A review of the Bio-Anchors report
This is a linkpost for a review of Ajeya Cotra's [Biological Anchors report](https://www.alignmentforum.org/posts/KrJfoZzpSDpnrv9va/draft-report-on-ai-timelines) (see also update [here](https://www.alignmentforum.org/posts/AfH2oPHCApdKicM4m/two-year-update-on-my-personal-ai-timelin... | https://www.lesswrong.com/posts/TMHWfRE7zZkzgFDSo/a-review-of-the-bio-anchors-report |
# If you want to learn technical AI safety, here's a list of AI safety courses, reading lists, and resources
If you're interested in technical AI safety, it's hard to know where to start. Lots of people have made courses and reading lists to help simplify things.
We [made a spreadsheet of such resources for learning... | https://www.lesswrong.com/posts/4aaJ8wvJ88okBxd4J/if-you-want-to-learn-technical-ai-safety-here-s-a-list-of-ai |
# Ivermectin: Much Less Than You Needed To Know
I've recently gotten in touch with Alexandros Marinos, which has been writing lengthy critiques of Scott's most famous post, his review of Ivermectin.
After hearing his point of view, I must admit I've been swayed to the thinking that:
1. There are some obvious statis... | https://www.lesswrong.com/posts/WnZSHEC5mBjuiDSRD/ivermectin-much-less-than-you-needed-to-know-1 |
# Recall and Regurgitation in GPT2
*The first half of this post uses causal tracing to explore differences in how GPT2-XL handles completing cached phrases vs completing factual statements. The second half details my attempt to build intuitions about the high-level structure of GPT2-XL and is speculation heavy.*
*Som... | https://www.lesswrong.com/posts/qxvihKpFMuc4tvuf4/recall-and-regurgitation-in-gpt2 |
# No free lunch theorem is irrelevant
I often see people cite the "no free lunch theorem" as a counterargument to the existence of an AGI. I think this is a very bad argument. This theorem is formulated as a problem of **predicting random and uniformly distributed data**. Just open the [Wikipedia](https://en.wikipedia... | https://www.lesswrong.com/posts/ng4zA3qCyygFFtqSs/no-free-lunch-theorem-is-irrelevant-1 |
# Humans aren't fitness maximizers
Recently I've gotten a bunch of pushback when I claim that humans are not maximizers of inclusive genetic fitness (IGF).
I think that part of what's going on here is a conflation of a few claims.
One claim that is hopefully uncontroversial (but that I'll expand upon below anyway) i... | https://www.lesswrong.com/posts/CzrF5rsJWvccFdemb/humans-aren-t-fitness-maximizers |
# Adversarial vs Collaborative Contexts
You can often make pretty clear predictions about what sort of effect something would have in the world as it exists today, but when you start to take into account how the world might change in response it gets very difficult. One tool here is thinking about whether you mostly e... | https://www.lesswrong.com/posts/zZXmiBmxcKxa7roLH/adversarial-vs-collaborative-contexts |
# Paper+Summary: OMNIGROK: GROKKING BEYOND ALGORITHMIC DATA
There is a new paper by Liu et al. that claims to have understood the key mechanism underlying grokking (potentially even generalization more broadly).
**They argue: **
1\. Grokking can be explained via the norm of the weights. They claim that there is... | https://www.lesswrong.com/posts/jusq6kyZ6XSrtW3Bf/paper-summary-omnigrok-grokking-beyond-algorithmic-data |
# Does Google still hire people via their foobar challenge?
In years since past, the [foobar challenge](https://towardsdatascience.com/google-foobar-challenge-level-1-3487bb252780) first appeared. 5 levels of ever harder coding challenges, and if you bested the first three you'd be given the option of providing your d... | https://www.lesswrong.com/posts/ZzEzhpiRQqjaNzJfK/does-google-still-hire-people-via-their-foobar-challenge |
# Feature request: Filter by read/ upvoted
I am looking for a post that I read and probably upvoted. I can't find it. I am 90% sure you know which posts I've read and 99.9% sure you know which posts I've upvoted. Can I be able to filter by them.
That's it. That's the post.
It's that post about not having to have... | https://www.lesswrong.com/posts/ynqEaXsxDfpuCLi7R/feature-request-filter-by-read-upvoted |
# Quick notes on “mirror neurons”
1\. Two definitions of “mirror neurons”
---------------------------------------
I have noticed the term “mirror neurons” being used in two ways:
* **(A)** Any set of neurons that activates *both* when I am doing / feeling / relating to X *and* when I am watching (or imagining) som... | https://www.lesswrong.com/posts/9quAQcvBp6KbdA2xQ/quick-notes-on-mirror-neurons |
# Will you let your kid play football?
Sparked by a somewhat vitriolic discussion on a dadgroup I'm in. Would you let your kid play football, if so what restrictions? If not, what other sports are allowed, and with what restrictions?
https://publications.aap.org/pediatrics/article/144/5/e20192180/38225/Concussion-Inc... | https://www.lesswrong.com/posts/KDCe8yeKBo2EMoW4w/will-you-let-your-kid-play-football |
# Smoke without fire is scary
*This post was written in response to Evan Hubinger’s shortform prompt below and benefited from discussions with him. Thanks to Justis Mills for feedback on this post.*
> If we see precursors to deception (e.g. non-myopia, self-awareness, etc.) but suspiciously don’t see deception itself... | https://www.lesswrong.com/posts/PP2Lrpvhd3bBvR8Aj/smoke-without-fire-is-scary |
# How are you dealing with ontology identification?
*Not really novel, but I've made this point quite often in recent conversations, so I decided to make a short write-up.*
I think a wide range of alignment proposals will *somehow* have to deal with ontology identification, i.e. mapping an AI's beliefs about the worl... | https://www.lesswrong.com/posts/HjFAp4RiaaGqH6gNC/how-are-you-dealing-with-ontology-identification |
# Looping
*Remember to remember...*
* * *
**Part I **
Tell me if this sounds familiar...
You're walking to work one bright and caffeinated morning, and you find yourself rather engaged in a shiny train of thought. It feels *novel*, and *important*, and *relevant to what matters.* One might even go so far as to say... | https://www.lesswrong.com/posts/soF2Hnyc39STngK6f/looping |
# Sleep Training
Let's say you're a baby. You reach the end of a sleep cycle and become partially alert. What do you do? Ideally, if nothing were wrong you would settle back in for another sleep cycle. Chances are, however, you don't know how to do this yet, so you wake up more fully and start to cry. This seems to re... | https://www.lesswrong.com/posts/EAKjaLhEpphLj6vTK/sleep-training |
# Reflection Mechanisms as an Alignment target: A follow-up survey
This is the second of three posts ([part I](https://www.lesswrong.com/posts/XyBWkoaqfnuEyNWXi/reflection-mechanisms-as-an-alignment-target-a-survey-1)) about surveying moral sentiments related to AI alignment. This work was done by Marius Hobbhahn and ... | https://www.lesswrong.com/posts/i3pkxN43NgkLRaAGZ/reflection-mechanisms-as-an-alignment-target-a-follow-up |
# Paper: Discovering novel algorithms with AlphaTensor [Deepmind]
The authors apply an AlphaZero-like algorithm to discover new matrix multiplication algorithms. They do this by turning matrix multiplication into a one-player game, where the state represents how far from correct the current output is, moves are algori... | https://www.lesswrong.com/posts/5Zfyktwgz3rvAvZyL/paper-discovering-novel-algorithms-with-alphatensor-deepmind |
# Tracking Compute Stocks and Flows: Case Studies?
*Posted in my personal capacity*
The AGI governance community has recently converged on compute governance^[See, e.g., [this sequence](https://forum.effectivealtruism.org/s/4yLbeJ33fYrwnfDev) by Lennart Heim.] as a promising lever for reducing existential risks from ... | https://www.lesswrong.com/posts/h3mX4esebgXnMyZSM/tracking-compute-stocks-and-flows-case-studies |
# 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 tra... | https://www.lesswrong.com/posts/QzpKq92nXqp8NHM34/neural-tangent-kernel-distillation |
# A blog post is a very long and complex search query to find fascinating people and make them route interesting stuff to your inbox
I was born in July 1989, which means I am of the last generation who will remember the time before the internet. The cables and data centers and hyperlinks grew up around me; they grew w... | https://www.lesswrong.com/posts/vidXh2DJtnqH5ysrZ/a-blog-post-is-a-very-long-and-complex-search-query-to-find |
# Finding Great Tutors
Does anyone have tips for finding great tutors for technical subjects? I'm looking to speed up my self-studying for eventual alignment research. Money isn't much of an issue, I'd probably meet every few days to talk about stuff that confused me and generally grill my teacher. :)
I'm mostly stud... | https://www.lesswrong.com/posts/wh2cKLccQahTD5Fzr/finding-great-tutors |
# Dependency Tree For The Development Of Plate Tectonics
This post is really rough and mostly meant to refer back to when I’ve produced more work on the subject. Proceed at your own risk.
Introduction
============
As I [mentioned](https://acesounderglass.com/2022/08/31/impact-shares-for-speculative-projects/) a few ... | https://www.lesswrong.com/posts/ruzKko3siw3qgq6Tb/dependency-tree-for-the-development-of-plate-tectonics |
# Generative, Episodic Objectives for Safe AI
## Introduction
Current discourse around AI safety focuses on creating some new paradigm for training AI models to make them safe, such as [assistance games](http://people.eecs.berkeley.edu/~russell/papers/mi19book-hcai.pdf) (Russell 2021) or [corrigibility](https://intel... | https://www.lesswrong.com/posts/amz29pbNZ9nE6z4ry/generative-episodic-objectives-for-safe-ai |
# Do uncertainty/planning costs make convex hulls unrealistic?
It's almost a rule that as soon as you have a "utility of possible outcomes" plot like this:
You must then say "and by randomly choosing be... | https://www.lesswrong.com/posts/uGBi9FMbqv92bWCjr/do-uncertainty-planning-costs-make-convex-hulls-unrealistic |
# AI Timelines via Cumulative Optimization Power: Less Long, More Short
**TLDR**: *We can best predict the future by using simple models which best postdict the past (ala Bayes/Solomonoff). A simple model based on net training compute postdicts the relative performance of successful biological and artificial neural n... | https://www.lesswrong.com/posts/3nMpdmt8LrzxQnkGp/ai-timelines-via-cumulative-optimization-power-less-long |
# Against Arguments For Exploitation
[*Kind of a rambling collection of thoughts that I tried to organize a bit. Hopefully it's interesting to read*]
## Introduction
I've recently had several conversations in which a member of the rationalist community made the case that exploiting people from less economically dev... | https://www.lesswrong.com/posts/paqK2mD2Bvd5FsP83/against-arguments-for-exploitation |
# Notes on Notes on the Synthesis of Form
Christopher Alexander was an architect and designer; I've written about some of his ... | https://www.lesswrong.com/posts/aHiAhtDHmfec6vBYX/notes-on-notes-on-the-synthesis-of-form |
# Warning Shots Probably Wouldn't Change The Picture Much
One piece of advice I gave to EAs of various stripes in early 2021 was: do everything you can to make the government sane around biorisk, in the wake of the COVID pandemic, because this is a practice-run for AI.
I said things like: if you can't get the world t... | https://www.lesswrong.com/posts/idipkijjz5PoxAwju/warning-shots-probably-wouldn-t-change-the-picture-much |
# Research Deprioritizing External Communication
*This was originally written in December 2021. I think it's good practice to run posts by orgs before making them public, and I did in this case. This has some benefits: orgs aren't surprised, they can prepare a response in advance if they want to, they can point out er... | https://www.lesswrong.com/posts/Keik32r5EyuRfS6hD/research-deprioritizing-external-communication |
# More Recent Progress in the Theory of Neural Networks
*Thanks to Dan Roberts and Sho Yaida for comments on a draft of this post.*
In this post, I would like to draw attention to the book [Principles of Deep Learning Theory](https://deeplearningtheory.com/) (PDLT), which I think represents a significant advance in o... | https://www.lesswrong.com/posts/nRu92PXLrdwqdtQmn/more-recent-progress-in-the-theory-of-neural-networks-1 |
# American invention from the “heroic age” to the system-building era
*Part 1 of my review of American Genesis*
The most productive decades in American history were roughly 1870–1940: the era that saw the birth and growth of the electric light and power industry; the invention of the automobile and airplane, and the ... | https://www.lesswrong.com/posts/zzsxYXnKajYdsagkW/american-invention-from-the-heroic-age-to-the-system |
# A shot at the diamond-alignment problem
I think that relatively simple alignment techniques can go a long way. In particular, I want to tell a plausible-to-me story about how simple techniques can align a proto-AGI so that it makes lots of diamonds.
But why is it interesting to get an AI which makes lots of diamond... | https://www.lesswrong.com/posts/k4AQqboXz8iE5TNXK/a-shot-at-the-diamond-alignment-problem |
# Covid 10/6/22: Overreactions Aplenty
Thanks to Balsa Research, every weekday I now get to meet and talk to someone new, usually multiple new people, all of whom are interesting, know things I don’t know that they are eager to explain to me, and who share a passion for modeling the world with an eye towards making th... | https://www.lesswrong.com/posts/9xQwYpt3bqy9ySP9N/covid-10-6-22-overreactions-aplenty |
# So, geez there's a lot of AI content these days
Since [April](https://www.lesswrong.com/posts/j9Q8bRmwCgXRYAgcJ/miri-announces-new-death-with-dignity-strategy) this year, there's been a huge growth in the the number of posts about AI, while posts about [rationality](https://www.lesswrong.com/tag/rationality), [world... | https://www.lesswrong.com/posts/P32AuYu9MqM2ejKKY/so-geez-there-s-a-lot-of-ai-content-these-days |
# linkpost: neuro-symbolic hybrid ai
This video shows some really interesting work that's been done on factoring sensory perceptions (visual data, natural language) into concept space (they call it 'derendering' as an antonym to rendering), and then applying symbolic reasoning to the concepts. I think this is really p... | https://www.lesswrong.com/posts/pvrH8zZJxELt4yKDM/linkpost-neuro-symbolic-hybrid-ai |
# What does it mean for an AGI to be 'safe'?
*(Note: This post is probably old news for most readers here, but I find myself repeating this surprisingly often in conversation, so I decided to turn it into a post.)*
I don't usually go around saying that I care about AI "safety". I go around saying that I care about "a... | https://www.lesswrong.com/posts/BSee6LXg4adtrndwy/what-does-it-mean-for-an-agi-to-be-safe |
# Accrue Nuclear Dignity Points
*Content warning: nuclear doom*
*Previously on last week's episode:* [*A Few Terrifying Facts About The Russo-Ukrainian War*](https://www.lesswrong.com/posts/gyMXuhcYyMRpwEjGq/a-few-terrifying-facts-about-the-russo-ukrainian-war)
* * *
You have probably thought about prepping at som... | https://www.lesswrong.com/posts/ZMgxFwqabdGepm3e6/accrue-nuclear-dignity-points |
# Calibrate - New Chrome Extension for hiding numbers so you can guess
One thing I wish I did more often is checks to see if I understand the world - Fermis, landmark numbers, predictions about how things work, etc.
Along the way I started wishing I could cover up numbers on websites so I would have the chance to gue... | https://www.lesswrong.com/posts/BWjGELGiaeH2utrfA/calibrate-new-chrome-extension-for-hiding-numbers-so-you-can |
# More examples of goal misgeneralization
In our latest [paper](https://arxiv.org/abs/2210.01790) and [accompanying blog post](https://deepmindsafetyresearch.medium.com/goal-misgeneralisation-why-correct-specifications-arent-enough-for-correct-goals-cf96ebc60924), we provide several new examples of [goal misgeneraliza... | https://www.lesswrong.com/posts/Cfe2LMmQC4hHTDZ8r/more-examples-of-goal-misgeneralization |
# Alignment Might Never Be Solved, By Humans or AI
I think it's possible that neither we, nor any future AI system, will manage to solve alignment. By "solving alignment" I mean having a cheap easily-scalable method for aligning AIs of arbitrary capability level to arbitrary goals. Here I will briefly present some arg... | https://www.lesswrong.com/posts/BuaFZud9BwkiSCGpd/alignment-might-never-be-solved-by-humans-or-ai |
# Polysemanticity and Capacity in Neural Networks
Elhage et al at Anthropic recently published a paper, [Toy Models of Superposition](https://transformer-circuits.pub/2022/toy_model/index.html) (previous Alignment Forum discussion [here](https://www.alignmentforum.org/posts/CTh74TaWgvRiXnkS6/toy-models-of-superpositio... | https://www.lesswrong.com/posts/kWp4R9SYgKJFHAufB/polysemanticity-and-capacity-in-neural-networks |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.