text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Nothing New: Productive Reframing
In a [comment](https://www.lesswrong.com/posts/qhxexho2mFjLnZ2nS/opportunity-cost-blackmail?commentId=mZALi6s3zbxQXwbsW) on [my last post](https://epistemologicalvigilance.substack.com/p/opportunity-cost-blackmail), someone deplored the tendency of LW bloggers to rediscover ideas of... | https://www.lesswrong.com/posts/ZZNM2JP6YFCYbNKWm/nothing-new-productive-reframing |
# David Krueger on AI Alignment in Academia, Coordination and Testing Intuitions
[David Krueger](https://www.davidscottkrueger.com/) is an assistant professor at the University of Cambridge and got his PhD from Mila. His research group focuses on aligning deep learning systems, but he is also interested in governance ... | https://www.lesswrong.com/posts/xph2djgMaGgdce4mz/david-krueger-on-ai-alignment-in-academia-coordination-and |
# Protectionism will Slow the Deployment of AI
I believe I’m more optimistic than the average LWer that regulation will be passed that slows down AI capabilities research. The current capabilities of language models threaten the interests of the politically salient professional managerial middle class, which will caus... | https://www.lesswrong.com/posts/apdXGcQJNuCSrgg4x/protectionism-will-slow-the-deployment-of-ai |
# Feature suggestion: add a 'clarity score' to posts
Add a voting mechanism for posts, beside the karma, that indicates whether you
* understood the post well enough to agree or disagree with it (upvote), or
* cannot understand the post with a reasonable time investment (downvote)
I suggest that this determines... | https://www.lesswrong.com/posts/punchyyf5vbcgbi7t/feature-suggestion-add-a-clarity-score-to-posts |
# Can Ads be GDPR Compliant?
I think the online ads ecosystem is most likely illegal in Europe, and as more decisions come out it will become clear that it can't be reworked to be within the bounds of the GDPR. This is a strong claim, but before I get into backing it up here's some background on me:
* I'm not a law... | https://www.lesswrong.com/posts/dFgfQTo4DRZG5t8Ap/can-ads-be-gdpr-compliant |
# AI psychology should ground the theories of AI consciousness and inform human-AI ethical interaction design
This post is a follow-up of Buck's [The case for becoming a black-box investigator of language models](https://www.lesswrong.com/posts/yGaw4NqRha8hgx5ny/the-case-for-becoming-a-black-box-investigator-of-langua... | https://www.lesswrong.com/posts/uHyZmfZKpXxo6uiEe/ai-psychology-should-ground-the-theories-of-ai-consciousness |
# Dangers of deference
*[Written September 02, 2022. Note: I'm likely to not respond to comments promptly.]*
Sometimes people defer to other people, e.g. by believing what they say, by following orders, or by adopting intents or stances. In many cases it makes sense to defer, since other people know more than you ab... | https://www.lesswrong.com/posts/ksBcnbfepc4HopHau/dangers-of-deference |
# Halifax Monthly Meetup: Moloch in the HRM
In the interest of growing the LW/EA/ACX community in Halifax, we are hosting meetups on the last Saturday of each month. Old hands and newcomers are welcome!
We'll begin by discussing Moloch in the HRM: Our ongoing coordination problems around the housing crisis/tax alloc... | https://www.lesswrong.com/events/dxsgZJrGiyMHsqWY9/halifax-monthly-meetup-moloch-in-the-hrm |
# 200 COP in MI: Image Model Interpretability
*This is the eighth post in a sequence called 200 Concrete Open Problems in Mechanistic Interpretability. *[*Start here*](https://www.alignmentforum.org/posts/LbrPTJ4fmABEdEnLf/200-concrete-open-problems-in-mechanistic-interpretability)*, then read in any order. If you wan... | https://www.lesswrong.com/posts/caMoe6yNfXcaCG2u3/200-cop-in-mi-image-model-interpretability |
# What specific thing would you do with AI Alignment Research Assistant GPT?
Why I think this question is important: I asked myself, "What would my AGI timelines be if some AI could summarize [Yudkowsky-Ngo debates on alignment difficulty](https://www.lesswrong.com/posts/7im8at9PmhbT4JHsW/ngo-and-yudkowsky-on-alignmen... | https://www.lesswrong.com/posts/cTdAXCb7WK6eR4fxf/what-specific-thing-would-you-do-with-ai-alignment-research |
# I tried to learn as much Deep Learning math as I could in 24 hours
**TL:DR** I designed an experiment where I committed to spend two 12 hour days trying to learn as much deep-learning math as possible, basically from scratch.
**Table of Contents**
1. [Origins and Motivations](https://www.lesswrong.com/posts/uFGD... | https://www.lesswrong.com/posts/uFGDrhzjGdELLajrK/i-tried-to-learn-as-much-deep-learning-math-as-i-could-in-24 |
# Citability of Lesswrong and the Alignment Forum
This post is short: as far as I know, there is no easy support to cite lesswrong or the alignment forum in academic papers. I think this should exist. This idea is related to the one on [putting Alignment Forum/Lesswrong posts on arxiv](https://www.alignmentforum.org/p... | https://www.lesswrong.com/posts/uEg5QisiRDsjqLrPj/citability-of-lesswrong-and-the-alignment-forum |
# Latent variable prediction markets mockup + designer request
[I've been trying to get places like Manifold or Metaculus to add latent variables to their prediction systems.](https://www.lesswrong.com/p... | https://www.lesswrong.com/posts/E8Zw4TPkRwWHNZ8ij/latent-variable-prediction-markets-mockup-designer-request |
# GPT learning from smarter texts?
(Epistemic status: I do not understand GPT deeply, so this is just a random idea.)
If I understand it correctly, GPT learns from existing texts. Lots of texts.
Would it be possible to make GPT smarter by simply giving it *smarter text*? Of course, writing tons of smarter text would... | https://www.lesswrong.com/posts/LdEMkP2Ajn4SvhHwd/gpt-learning-from-smarter-texts |
# Simulacra are Things
This is a note I wrote about a year ago. It's fairly self-contained, so I decided to make a post out of it after [Vladimir_Nesov's comment](https://www.lesswrong.com/posts/TTn6vTcZ3szBctvgb/simulators-seminar-sequence-2-semiotic-physics?commentId=c7PpsY8euCPHb2eDe) caused me to dig up this text ... | https://www.lesswrong.com/posts/3BDqZMNSJDBg2oyvW/simulacra-are-things |
# On Blogging and Podcasting
_A cover of and response to [‘Why and how to write things on the internet’](https://www.benkuhn.net/writing/), by [Ben Kuhn](https://www.benkuhn.net/)._
Blogging
--------
One thing you could be doing with your limited number of hours on this earth is writing a blog. This basically involv... | https://www.lesswrong.com/posts/7imycCH6H95jCKdWN/on-blogging-and-podcasting |
# You're Not One "You" - How Decision Theories Are Talking Past Each Other
(Crossposted from [MidWittgenstein](https://midwittgenstein.substack.com/p/youre-not-one-you-how-decision-theories?r=ylvm6&utm_campaign=post&utm_medium=web) (as part of my lame resolution to write more this year))
*This is a post about what I ... | https://www.lesswrong.com/posts/4gaGgGs5mEara9rea/you-re-not-one-you-how-decision-theories-are-talking-past |
# Wolf Incident Postmortem
Incident #210
-------------
### Status
Complete, one action item outstanding.
### Summary
Sentinel consumed by wolf after repeated false alarms.
### Impact
Loss of sentinel. No flock impact.
### Root causes
Sentinel generated noisy alerts due to premature deployment, incomplete train... | https://www.lesswrong.com/posts/aRxDLju75KXD6PCpB/wolf-incident-postmortem |
# Big list of AI safety videos
This is a [Google doc](https://docs.google.com/document/d/1vDWLKPvJNo0k9IwMPvduO23_CXOdMpxn0aRDnvrL0yw/edit?usp=sharing) containing links to:
* (all?) AI safety-related YouTube channels (intended to be comprehensive, at least for active channels with videos that seem worth sharing in ... | https://www.lesswrong.com/posts/H2BPqnvv7YyjiEHam/big-list-of-ai-safety-videos |
# Reification bias
> “All right," said Susan. "I'm not stupid. You're saying humans need... *fantasies* to make life bearable."
>
> REALLY? AS IF IT WAS SOME KIND OF PINK PILL? NO. HUMANS NEED FANTASY TO BE HUMAN. TO BE THE PLACE WHERE THE FALLING ANGEL MEETS THE RISING APE.
>
> "Tooth fairies? Hogfathers? Little—"
... | https://www.lesswrong.com/posts/g2ZAH8mSrtSuZnHEi/reification-bias |
# How to Bounded Distrust
[Scott Alexander points out](https://astralcodexten.substack.com/p/the-media-very-rarely-lies) that the media, from The New York Times to Infowars, very rarely lies *explicitly and directly.*
[Alas, the media often misleads](https://astralcodexten.substack.com/p/sorry-i-still-think-i-am-righ... | https://www.lesswrong.com/posts/rvpEF2mBLeZE9j53n/how-to-bounded-distrust |
# Evidence under Adversarial Conditions
I recently noticed similarities between how I decide what stock market
evidence to look at, and how the legal system decides what lawyers are
allowed to tell juries.
This post will elaborate on Eliezer\'s [Scientific Evidence, Legal
Evidence, Rational
Evidence](https://ww... | https://www.lesswrong.com/posts/fZtAr9JznFPR5LsHP/evidence-under-adversarial-conditions |
# The special nature of special relativity
I honestly thought that both theories of relativity were theories of gravitation. Special and general relativity, the latter being the generalized version of the latter. That makes sense, right?
Then I started actually digging into special relativity. I studied the original... | https://www.lesswrong.com/posts/hcsLuYxRnsf6WSPsk/the-special-nature-of-special-relativity |
# Trying to isolate objectives: approaches toward high-level interpretability
*Epistemic status: There’s a lot in this post and my general approach while working on it that, in retrospect, wasn’t thought out well enough. I’m posting it because I figure sharing flawed ideas is better than letting this languish in a doc... | https://www.lesswrong.com/posts/sDKi2pQ3fnTSbR7H8/trying-to-isolate-objectives-approaches-toward-high-level |
# [MLSN #7]: an example of an emergent internal optimizer
As part of a larger community building effort, [CAIS](https://safe.ai) is writing a safety newsletter that is designed to cover empirical safety research and be palatable to the broader machine learning research community. You can [subscribe here](https://newsl... | https://www.lesswrong.com/posts/BsbmsrpboN5jbESwD/mlsn-7-an-example-of-an-emergent-internal-optimizer |
# GWWC Should Require Public Charity Evaluations
One of the roles of [Giving What We Can](https://www.givingwhatwecan.org/) (GWWC) is to help its members and other interested people figure out where to give. If you go to their site and click " [start giving](https://www.givingwhatwecan.org/donate/organizations)" they ... | https://www.lesswrong.com/posts/HxoQddQKTejtTPg4k/gwwc-should-require-public-charity-evaluations |
# EA & LW Forum Summaries - Holiday Edition (19th Dec - 8th Jan)
*Supported by Rethink Priorities*
**Author's note:** this post summarizes all 150+ karma posts on the forums since 19th December. We'll be back to our regular schedule and karma requirements next week :)
**Usual preamble:** This is part of a weekly... | https://www.lesswrong.com/posts/kLxBj2K9cxhCrixD8/ea-and-lw-forum-summaries-holiday-edition-19th-dec-8th-jan |
# Wentworth and Larsen on buying time
*This post summarizes a discussion between Thomas Larsen and John Wentworth. They discuss efforts to buy time, outreach to AGI labs, models of institutional change, and other ideas relating to a post that Thomas, Olivia Jimenez, and I released: *[*Instead of technical research, mo... | https://www.lesswrong.com/posts/JdGuqg7ifRwPiirCe/wentworth-and-larsen-on-buying-time |
# Review AI Alignment posts to help figure out how to make a proper AI Alignment review
I've had many conversations over the last few years about the health of the AI Alignment field and one of the things that has come up most frequently (including in conversations with Rohin, Buck and various Open Phil people) is tha... | https://www.lesswrong.com/posts/Qprn8tMeZGLBfobu7/review-ai-alignment-posts-to-help-figure-out-how-to-make-a |
# Idea: Learning How To Move Towards The Metagame
I have just finished watching someone play a variant of Brick Breaker. A variant I became intimately familiar with, indeed, obsessed, as I strove to hit the top 40 all time score which scrolled past me each time I failed, mocking me.
And this reminded me of an intere... | https://www.lesswrong.com/posts/Mhyb4YkeHJkDy7J2e/idea-learning-how-to-move-towards-the-metagame |
# Notes on writing
Cross-posted from my [new blog](https://complexbubbletea.xyz/).
Recently, I gave a few people some advice on how to write well. Below is that advice.
What is it to write well?
1. Have a thought.
2. Get that thought from your head to someone else’s.
What is a thought?
In this community, ... | https://www.lesswrong.com/posts/7d6SJYzpWCqofp6MJ/notes-on-writing |
# Scaling laws vs individual differences
*Crossposted from my* [*personal blog*](https://www.beren.io/2022-09-18-Scaling-laws-vs-individual-differences/)
**Epistemic Status:** *This is a quick post on something I have been confused about for a while. If an answer to this is known, please reach out and let me know! *
... | https://www.lesswrong.com/posts/eYFscbv5BJ8Fezauj/scaling-laws-vs-individual-differences |
# Escape Velocity from Bullshit Jobs
Without speculating here on how likely this is to happen, suppose that GPT-4 (or some other LLM or AI) speeds up, streamlines or improves quite a lot of things. What then?
### The Dilemma
[Samo and Ben’s dilemma:](https://twitter.com/SamoBurja/status/1610392741665705985) To the e... | https://www.lesswrong.com/posts/Y5mZsBxH9ouMtNJ7v/escape-velocity-from-bullshit-jobs |
# Sorting Pebbles Into Correct Heaps: The Animation
This video is an animation of [Sorting Pebbles Into Correct Heaps](https://www.lesswrong.com/posts/mMBTPTjRbsrqbSkZE/sorting-pebbles-into-correct-heaps) by Eliezer Yudkowsky. The narration is a word-for-word reading of the story. Enjoy :)
Right after the video is ma... | https://www.lesswrong.com/posts/sQvGLgBuCiBJwNocj/sorting-pebbles-into-correct-heaps-the-animation |
# Against using stock prices to forecast AI timelines
*by* [*Trevor Chow*](https://tmychow.com/)*, *[*Basil Halperin*](https://basilhalperin.com/)*, and *[*J. Zachary Mazlish*](https://jzmazlish.substack.com/)
\[Note: This is an appendix to ["AGI and the EMH: markets are not expecting aligned or unaligned AI in th... | https://www.lesswrong.com/posts/yFkNYyspBBqfSeBx9/against-using-stock-prices-to-forecast-ai-timelines |
# The Alignment Problem from a Deep Learning Perspective (major rewrite)
We've recently uploaded a major rewrite of Richard Ngo's:
[The Alignment Problem from a Deep Learning Perspective](https://arxiv.org/abs/2209.00626)
We hope it can reach ML researchers by being more grounded in the deep learning literature and... | https://www.lesswrong.com/posts/5GxLiJJEzvqmTNyCK/the-alignment-problem-from-a-deep-learning-perspective-major |
# AGI and the EMH: markets are not expecting aligned or unaligned AI in the next 30 years
*by* [*Trevor Chow*](https://tmychow.com/)*, *[*Basil Halperin*](https://basilhalperin.com/)*, and *[*J. Zachary Mazlish*](https://jzmazlish.substack.com/)
In this post, we point out that short AI timelines would cause *real int... | https://www.lesswrong.com/posts/ngpC5PFAgxHJMhicM/agi-and-the-emh-markets-are-not-expecting-aligned-or-1 |
# 200 COP in MI: Interpreting Reinforcement Learning
*This is the ninth post in a sequence called 200 Concrete Open Problems in Mechanistic Interpretability. *[*Start here*](https://www.alignmentforum.org/posts/LbrPTJ4fmABEdEnLf/200-concrete-open-problems-in-mechanistic-interpretability)*, then read in any order. If y... | https://www.lesswrong.com/posts/eqvvDM25MXLGqumnf/200-cop-in-mi-interpreting-reinforcement-learning |
# We don’t trade with ants
When discussing advanced AI, sometimes the following exchanges happens:
“Perhaps advanced AI won’t kill us. Perhaps it will trade with us”
“We don’t trade with ants”
I think it’s interesting to get clear on exactly why we don’t trade with ants, and whether it is relevant to the AI situati... | https://www.lesswrong.com/posts/wB7hdo4LDdhZ7kwJw/we-don-t-trade-with-ants |
# Running With a Backpack
In the comments on my [running by default](https://www.jefftk.com/p/running-by-default) post, [one](https://www.lesswrong.com/posts/4y3qcbnhSmXrk6p34#rHwSsvRKR7S7jxh8v) [of](https://schelling.pt/@zebrask/109637179105823464) [the](https://www.facebook.com/jefftk/posts/pfbid02RjLVh3wnXp5BsjQpMz... | https://www.lesswrong.com/posts/CjJSqjuYK8XH7XkJJ/running-with-a-backpack |
# Compounding Resource X
*A concept I got from Anna Salamon, and maybe also from my coworker Jacob Lagerros, although I haven't run this by either of them and they may not endorse my phrasing. I think I probably run with it in a different direction than Anna would endorse.*
*Epistemic effort: thought about it for a f... | https://www.lesswrong.com/posts/uHRPJSHWwuori5Gx8/compounding-resource-x |
# Any significant updates on long covid risk analysis?
[ETA: This question is specifically about *risk analysis of long covid, given current knowledge.* Discussion should be about things such as:
- base rates
- priors and posteriors for LC risk
- evidence that would make one update in favor of certain new policies wi... | https://www.lesswrong.com/posts/GzjKoos4kgDcpcETB/any-significant-updates-on-long-covid-risk-analysis |
# How to write a big cartesian product symbol in MathJax
$$This short post is about how to write something like $\newcommand{\bigtimes}{\mathop{\Large\times\normalsize}} \bigtimes_{i\in I} V_i$ and $\displaystyle\bigtimes_{i\in I}V_i$
MathJax doesn't allow picture environments so we can't define a new command that d... | https://www.lesswrong.com/posts/wsgWALmFRhTMGLvi5/how-to-write-a-big-cartesian-product-symbol-in-mathjax |
# GWWC's Handling of Conflicting Funding Bars
When [recommending charities](https://www.givingwhatwecan.org/inclusion-criteria), [Giving What We Can](https://www.givingwhatwecan.org/) (GWWC) mirrors the recommendations of their [trusted evaluators](https://www.givingwhatwecan.org/trusted-evaluators). For example, they... | https://www.lesswrong.com/posts/dbJmWtuMdLn39AxbD/gwwc-s-handling-of-conflicting-funding-bars |
# Have we really forsaken natural selection?
Natural selection is often charged with having goals for humanity, and humanity is often charged with falling down on them. The big accusation, I think, is of sub-maximal procreation. If we cared at all about the genetic proliferation that natural selection wanted for us, t... | https://www.lesswrong.com/posts/T4Lfw2HZQNFjNX8Ya/have-we-really-forsaken-natural-selection |
# Any person/mind should have the right to suicide
I believe no person/mind may suffer without the right to suicide, so any person/mind should have the right to suicide. The perspective of forced immortality (when someone wants to die (because of some sufferings or doesn't really matter why), but is forced to exist in... | https://www.lesswrong.com/posts/FDYBborQvr8j7sveQ/any-person-mind-should-have-the-right-to-suicide |
# Categorical-measure-theoretic approach to optimal policies tending to seek power
The paper [Optimal Policies Tend to Seek Power](https://arxiv.org/abs/1912.01683) tries to justify the claim in the title with a mathematical argument, using a clever formal definition of "power". I like the general approach, but I thin... | https://www.lesswrong.com/posts/9dJgvGh4wNyPbQfrt/categorical-measure-theoretic-approach-to-optimal-policies |
# How it feels to have your mind hacked by an AI
Last week, while talking to an LLM (a large language model, which is the main talk of the town now) for several days, I went through an emotional rollercoaster I never have thought I could become susceptible to.
I went from snarkily condescending opinions of the recent... | https://www.lesswrong.com/posts/9kQFure4hdDmRBNdH/how-it-feels-to-have-your-mind-hacked-by-an-ai |
# Alignment is not enough
This post is a linkpost for my substack article.
Substack link: [https://coordination.substack.com/p/alignment-is-not-enough](https://coordination.substack.com/p/alignment-is-not-enough)
* * *
[
___
# Excerpts
> If all goes according to OpenAI’s financial plans, Microsoft will close a $10 billion investment deal into the artificial intelligence startup before the end o... | https://www.lesswrong.com/posts/X7QbzyKWqnLmeZCnJ/microsoft-plans-to-invest-usd10b-in-openai-usd3b-invested-to |
# [Linkpost] DreamerV3: A General RL Architecture
Here's a link towards DreamerV3, a new model from DeepMind that can be trained on a bunch of different tasks (including a simplified version of Minecraft) and outperform more narrow models. Link: [https://arxiv.org/pdf/2301.04104v1.pdf](https://arxiv.org/pdf/2301.04104... | https://www.lesswrong.com/posts/kBJsot7RT7njnX2B5/linkpost-dreamerv3-a-general-rl-architecture |
# FFMI Gains: A List of Vitalities
*(This was mostly written as a part of a quick writing session at the* [*Lurkshop*](https://www.lesswrong.com/posts/r8K8ddYDYNrPJRFyC/lesswrong-readers-are-invited-to-apply-to-the-lurkshop)*. Thanks to Evgeny, whose handle I do not know, for some refinements and suggestions!)*
Most ... | https://www.lesswrong.com/posts/he9JtjoZJyteaMGeP/ffmi-gains-a-list-of-vitalities |
# Iron deficiencies are very bad and you should treat them
In brief
========
Recently I became interested in what kind of costs were inflicted by iron deficiency, so I looked up studies until I got tired. This was not an exhaustive search, but the results are so striking that even with wide error bars I found them c... | https://www.lesswrong.com/posts/6frs5xTkeLc9vZSRN/iron-deficiencies-are-very-bad-and-you-should-treat-them |
# VIRTUA: a novel about AI alignment
I’ve written a novel about AI alignment in German and translated it into English.
**Edit:** I'm going to publish the final version [on Amazon KDP](https://www.amazon.com/dp/B0C3CFD5VB/) on May 25. I have therefore removed the links to the previous version. If you want a free copy... | https://www.lesswrong.com/posts/rSiybWzeiG8agYtNr/virtua-a-novel-about-ai-alignment |
# ea.domains - Domains Free to a Good Home
**Tl;dr: I’ve set up **[**a database of domains at ea.domains**](https://ea.domains/) **which are**[**free to a good home**](https://ea.domains/#about)**, to prevent them being squatted and blocked for use. You can add domains you control to it using **[**this form**](https:/... | https://www.lesswrong.com/posts/sZetFH7cwxak6pLg5/ea-domains-domains-free-to-a-good-home |
# [Linkpost] Scaling Laws for Generative Mixed-Modal Language Models
In this paper authors explore the scaling properties of mixed-modal generative models, discovering new scaling laws that unify the contributions of individual modalities and the interactions between them. I find most interesting that they have found ... | https://www.lesswrong.com/posts/6w9uTPdJk52Nyknvm/linkpost-scaling-laws-for-generative-mixed-modal-language |
# Covid 1/12/23: Unexpected Spike in Deaths
Quite a lot of deaths got reported by quite a lot of different states this week. Case numbers did _not_ spike, and instead actively declined. What’s going on? It is right after Christmas and New Years, which means the least reliable reporting of the year, and no one cares mu... | https://www.lesswrong.com/posts/6vcfDvuRyE6DCoFCs/covid-1-12-23-unexpected-spike-in-deaths |
# ChatGPT struggles to respond to the real world
I wanted to test whether ChatGPT could respond to situations as they arise in the real world, so I asked it to guide me through making a turmeric latte. It had no problem giving me a nicely formatted recipe but it really struggled to guide me through the process of maki... | https://www.lesswrong.com/posts/H4bAKBwnFktvoLtp4/chatgpt-struggles-to-respond-to-the-real-world |
# Reward is not Necessary: How to Create a Compositional Self-Preserving Agent for Life-Long Learning
There is an idea in the AI safety community that [expected utility maximisation is the "limit" of agency](https://www.lesswrong.com/posts/XYDsYSbBjqgPAgcoQ/why-the-focus-on-expected-utility-maximisers?commentId=mgdud4... | https://www.lesswrong.com/posts/df4Jjg9cmJ7R2bkzR/reward-is-not-necessary-how-to-create-a-compositional-self-1 |
# What is a disagreement you have around AI safety?
I was once discussing Aumann's agreement theorem with a rationalist who wanted to use it to reduce disagreements about AI safety. We were thinking about why it doesn't work in that context when (I would argue) Aumann's agreement theorem applies robustly in lots of ot... | https://www.lesswrong.com/posts/8xBwd7cmwgEmFFgey/what-is-a-disagreement-you-have-around-ai-safety |
# Victoria Krakovna on AGI Ruin, The Sharp Left Turn and Paradigms of AI Alignment
Victoria Krakovna is a Research Scientist at DeepMind working on [AGI safety](https://deepmindsafetyresearch.medium.com/) and a co-founder of the [Future of Life Institute](https://futureoflife.org/), a non-profit organization working t... | https://www.lesswrong.com/posts/5SqSZazHjrzhvxmCE/victoria-krakovna-on-agi-ruin-the-sharp-left-turn-and |
# Announcing the 2023 PIBBSS Summer Research Fellowship
*What can we learn from biological and social systems about designing beneficial AI systems?*
We are, once again, pleased to announce the PIBBSS Research Fellowship program.
*Principles of Intelligent Behavior in Biological and Social Systems* (PIBBSS) aims to... | https://www.lesswrong.com/posts/cJWipZCfAbg9tFX2P/announcing-the-2023-pibbss-summer-research-fellowship |
# Proposal for Inducing Steganography in LMs
This is my half-baked proposal for inducing steganography in a decoder-only model (like GPT), and I'd appreciate some feedback.
I've operationalized "steganography" to mean a model that has chain-of-thought (CoT) reasoning that doesn't appear to help the target task, but a... | https://www.lesswrong.com/posts/dC7SGJNLcRAHdCQcd/proposal-for-inducing-steganography-in-lms |
# The Alignment Problems
*Epistemic status: A core idea I've seen hinted at in many places, here presented and motivated explicitly.*
*This was produced during SERI MATS 3.0. Thanks to Vivek Hebbar for related discussion.*
There has recently been pushback on the inner/outer alignment conceptualization: it's [confuse... | https://www.lesswrong.com/posts/cHd6xLX6qeNYQACat/the-alignment-problems-1 |
# The AI Control Problem in a wider intellectual context
I’ve been thinking about the control problem lately. The control problem, also called the AI alignment problem is, per Wikipedia:
*\[A\]spects of how to build AI systems such that they will aid rather than harm their creators. One particular concern is that hum... | https://www.lesswrong.com/posts/Afs6FtptMSWcAetxR/the-ai-control-problem-in-a-wider-intellectual-context |
# AGISF adaptation for in-person groups
This past semester,[ HAIST](https://haist.ai/) and[ MAIA](https://www.mitalignment.org/) (the Harvard and MIT AI safety student groups) ran an adapted version of Richard Ngo's [AGI Safety Fundamentals alignment](https://www.agisafetyfundamentals.com/ai-alignment-curriculum) curr... | https://www.lesswrong.com/posts/jsy8t64Jp5xcuFrXX/agisf-adaptation-for-in-person-groups |
# Disentangling Shard Theory into Atomic Claims
Introduction
============
*Produced as part of the* [*SERI ML Alignment Theory Scholars Program*](https://serimats.org/) *\- Winter 2022 Cohort. Thanks to* [*Magdalena Wache*](https://www.lesswrong.com/users/magdalena-wache) *for giving feedback on a recent version, and... | https://www.lesswrong.com/posts/L4e7CqqpDxea2x4Gg/disentangling-shard-theory-into-atomic-claims |
# Burning Uptime: When your Sandbox of Empathy is Leaky and also an Hourglass
Epistemic status: idle musings
Some of us sometimes get so deeply caught up in a thought or understanding of someone else that it's difficult to switch back to our own context and experiences. This is what the mecha animation series Ge... | https://www.lesswrong.com/posts/v73A5tka3iKfebJmd/burning-uptime-when-your-sandbox-of-empathy-is-leaky-and |
# Could Simulating an AGI Taking Over the World Actually Lead to a LLM Taking Over the World?
As you know, large language models can be understood as [simulators](https://www.lesswrong.com/posts/vJFdjigzmcXMhNTsx/simulators) or more simply as models predicting what token would plausibly appear next after a first sequ... | https://www.lesswrong.com/posts/mR65cuuoG9prZnAvf/could-simulating-an-agi-taking-over-the-world-actually-lead |
# Some Arguments Against Strong Scaling
There are many people who believe that we will be able to get to AGI by basically just scaling up the techniques used in recent large language models, combined with some relatively minor additions and/or architectural changes. As a result, there are people in the AI safety commu... | https://www.lesswrong.com/posts/DvCLEkr9pXLnWikB8/some-arguments-against-strong-scaling |
# Beware safety-washing
Tl;dr: don’t be fooled into thinking that some groups working on AI are taking “safety” concerns seriously (enough).[^6zkt67gme6q]
**Outline**
* [Two non-AI examples](https://forum.effectivealtruism.org/posts/f2qojPr8NaMPo2KJC/beware-safety-washing#Two_non_AI_examples)
* [Greenwashing... | https://www.lesswrong.com/posts/PY3HEHc5fMQkTrQo4/beware-safety-washing |
# On Cooking With Gas
[Research meta-analysis says gas stoves give off unsafe N02 levels and increase risk of childhood asthma](https://twitter.com/curious_founder/status/1612533394109243397) by 34% and are responsible for 12.7% of childhood asthma.
None of this is new. The way safety panic works is that you focus on... | https://www.lesswrong.com/posts/deJHkRXqb8bv9Dnxs/on-cooking-with-gas |
# Robustness & Evolution [MLAISU W02]
Welcome to this week’s ML Safety Report where we talk about robustness in machine learning and the human-AI dichotomy. Stay until the end to check out several amazing competitions you can participate in today.
Watch this week's MLAISU on [YouTube](https://youtu.be/bT0VD0lGO_Q) or... | https://www.lesswrong.com/posts/RjeshF3iKaY2Bvkt9/robustness-and-evolution-mlaisu-w02 |
# How we could stumble into AI catastrophe
This post will lay out a couple of stylized stories about **how, if transformative AI is developed relatively soon, this could result in global catastroph... | https://www.lesswrong.com/posts/85DTWEmA25sTciHvy/how-we-could-stumble-into-ai-catastrophe |
# Tracr: Compiled Transformers as a Laboratory for Interpretability | DeepMind
## Abstract
Interpretability research aims to build tools for understanding machine learning (ML) models. However, such tools are inherently hard to evaluate because we do not have ground truth information about how ML models actually wo... | https://www.lesswrong.com/posts/2J6fFHQZkWxFcjL6c/tracr-compiled-transformers-as-a-laboratory-for-1 |
# Money is a way of thanking strangers
In Ann Arbor, I live with 121,000 strangers. Together, we live among 10 million strangers in the state of Michigan, 332 million strangers in the USA, and nearly 8 billion strangers on Earth.
I don't mind doing favors for my friends and loved ones, or even for one of the 8 billio... | https://www.lesswrong.com/posts/gA39uaqkcrcyYMurk/money-is-a-way-of-thanking-strangers |
# [ASoT] Simulators show us behavioural properties by default
*I’m trying out making a few posts with less polish and smaller scope, to iterate more quickly on my thoughts and write about some interesting ideas in isolation *[*before having fully figured them out*](https://www.alignmentforum.org/s/KgrG4cQdLtL9DvNr2)*.... | https://www.lesswrong.com/posts/mC2omdN4ekcsNkCmp/asot-simulators-show-us-behavioural-properties-by-default-1 |
# How does GPT-3 spend its 175B parameters?
*\[Target audience: Me from a week ago, and people who have some understanding of ML but want to understand transformers better on a technical level.\]*
Free advice for people learning new skills: ask yourself random questions. In answering them, you’ll strengthen your und... | https://www.lesswrong.com/posts/3duR8CrvcHywrnhLo/how-does-gpt-3-spend-its-175b-parameters |
# Smallpox vaccines are widely available, for now
In response to the 2022 monkeypox outbreak, public health departments throughout the USA and Europe are offering free smallpox vaccines to the public. (The diseases are closely related and the same vaccine protects against both.) Demand for these vaccines was high when... | https://www.lesswrong.com/posts/acocKCzCLZxhyQjyt/smallpox-vaccines-are-widely-available-for-now |
# Concrete Reasons for Hope about AI
Recent advances in machine learning—in reinforcement learning, language modeling, image and video generation, translation and transcription models, etc.—without similarly striking safety results, have rather dampened the mood in many AI Safety circles. If I was any less concerned ... | https://www.lesswrong.com/posts/BfN88BfZQ4XGeZkda/concrete-reasons-for-hope-about-ai |
# Abstractions as morphisms between (co)algebras
In a [previous post](https://www.lesswrong.com/posts/nLhHY2c8MWFcuWRLx/good-ontologies-induce-commutative-diagrams), I suggested we should think of abstractions/ontologies as maps that make a certain commutative diagram commute:
I wanted to share that I think it's not bad to think about the object level question of whether there are group differences in intelligence rooted in genetic differences. This is a... | https://www.lesswrong.com/posts/GqD9ZKeAbNWDqy8Mz/a-general-comment-on-discussions-of-genetic-group |
# Current AI Models Seem Sufficient for Low-Risk, Beneficial AI
*Note: Mostly opinion/speculation, hoping to encourage further discussion.*
**Summary:** Current AI models have demonstrated amazing capabilities. It seems feasible to turn them into usable products without much additional scaling. These models are pro... | https://www.lesswrong.com/posts/a2QkRM3C8kJnngLfS/current-ai-models-seem-sufficient-for-low-risk-beneficial-ai |
# World-Model Interpretability Is All We Need
**Summary, by sections**:
1. Perfect world-model interpretability seems both sufficient for robust alignment (via a decent variety of approaches) and realistically attainable (compared to "perfect interpretability" in general, i. e. insight into AIs' heuristics, goals, a... | https://www.lesswrong.com/posts/HaHcsrDSZ3ZC2b4fK/world-model-interpretability-is-all-we-need |
# When to mention irrelevant accusations?
I expect in the near-ish future to publish a thing. That thing references an idea that someone had and published on LW. I believe (having seen a credible accusation) that person has acted unethically and illegally, in ways irrelevant to the idea.
I currently intend to menti... | https://www.lesswrong.com/posts/rDuqaL4QGEvizBusE/when-to-mention-irrelevant-accusations |
# We Need Holistic AI Macrostrategy
Summary
=======
AI Macrostrategy is the study of high level questions having to do with prioritizing the use of resources on the current margin in order to achieve good AI outcomes. AI macrostrategy seems important if it is tractable. However, while few people are working on esti... | https://www.lesswrong.com/posts/Jh8Trhc89JDPjnk2J/we-need-holistic-ai-macrostrategy |
# Podcast with Divia Eden on operant conditioning
I interviewed Divia Eden for 2.5 hours about operant conditioning on my side podcast, [The Filan Cabinet](https://thefilancabinet.com/). I think it's both an interesting conversation in its own right, as well as being kinda related to human rationality as well as AI an... | https://www.lesswrong.com/posts/hcMy8saPnrbsBLax5/podcast-with-divia-eden-on-operant-conditioning |
# Consider paying for literature or book reviews using bounties and dominant assurance contracts
*Cross-posted to the EA Forum* [*here*](https://forum.effectivealtruism.org/posts/dMhscQ9sTXCaegYju/consider-paying-for-literature-or-book-reviews-using)*.*
In September 2021, a [LessWrong pilot program](https://www.lessw... | https://www.lesswrong.com/posts/9cvK8QX9ZRBxmEfcz/consider-paying-for-literature-or-book-reviews-using |
# Core Concept Conversations
There are various important, conceptual things that I just don't really understand. [For example](https://www.lesswrong.com/posts/SHuRvZxmCSC3hMHj9/core-concept-conversation-making-wealth-1), creating wealth. What exactly does that mean? I read Paul Graham's [essay](http://www.paulgraham.c... | https://www.lesswrong.com/posts/Gtf9aTjsW2eJZRMSe/core-concept-conversations |
# Core Concept Conversation: What is wealth?
This is a [Core Concept Conversation](https://www.lesswrong.com/posts/Gtf9aTjsW2eJZRMSe/core-concept-conversations) post. The topic is what wealth is and what it means to create it, largely based off of Paul Graham's essay [How to Make Wealth](http://www.paulgraham.com/weal... | https://www.lesswrong.com/posts/SHuRvZxmCSC3hMHj9/core-concept-conversation-what-is-wealth |
# Language Ex Machina
Reading the spoilered information before reading this post will alter your experience of it, possibly for the worse, but it might also save you a lot of confusion. If you generally dislike being confused, I recommend you read it now. If you choose not to read it now, read it later.
This is an es... | https://www.lesswrong.com/posts/vPsupipfyeDoSAirY/language-ex-machina |
# Non-directed conceptual founding
*[Metadata: crossposted from [https://tsvibt.blogspot.com/2022/08/non-directed-conceptual-founding.html](https://tsvibt.blogspot.com/2022/08/non-directed-conceptual-founding.html). Written 13 June 2022. I'm fairly likely to not respond to comments promptly. If you're especially inte... | https://www.lesswrong.com/posts/JFoQ3echrH2pfjKuP/non-directed-conceptual-founding |
# Deceptive failures short of full catastrophe.
### Epistemic status: trying to unpack a fuzzy mess of related concepts in my head into something a bit cleaner.
A lot of my concern about risks from advanced AI is because of the possibility of [deceptive alignment](https://www.alignmentforum.org/s/r9tYkB2a8Fp4DN8yB/p/... | https://www.lesswrong.com/posts/iNaB6GA6Seti3biTJ/deceptive-failures-short-of-full-catastrophe |
# How Does the Human Brain Compare to Deep Learning on Sample Efficiency?
I have an impression that within lifetime human learning is orders of magnitude more sample efficient than large language models, but there are numerous caveats to this:
1. We don't have "an ecological evaluation objective" for language models... | https://www.lesswrong.com/posts/t6MNtQzsdY66ayDf4/how-does-the-human-brain-compare-to-deep-learning-on-sample |
# Underspecification of Oracle AI
*Rubi and Johannes worked on this post as part of the SERI MATS program, with Evan Hubinger providing mentorship to both. Rubi also received mentorship from Leo Gao. Thanks to Paul Colognese and Nicholas Schiefer for discussions related to this post.*
An oracle is a type of AI system... | https://www.lesswrong.com/posts/aBRS3x4sPSJ9G6xkj/underspecification-of-oracle-ai |
# Speculation on Path-Dependance in Large Language Models.
*Epistemic Status: **Highly Speculative.** I spent less than a day thinking about this in particular, and though I have spent a few months study... | https://www.lesswrong.com/posts/pt8Sf2kvRZ8BBW5b5/speculation-on-path-dependance-in-large-language-models |
# Understanding the diffusion of large language models: summary
5-minute summary
================
| **Time from GPT-3’s** [**publication**](https://arxiv.org/abs/2005.14165)** (May 2020) to…** | **Time** | **Date** | **Model** | **Organizations involved** |
| --- | --- | --- | --- | --- |
| A better model is produced... | https://www.lesswrong.com/posts/2hxyLjtsaHWgdn4vm/understanding-the-diffusion-of-large-language-models-summary |
# GPT-3-like models are now much easier to access and deploy than to develop
*This post is one part of the sequence* [*Understanding the diffusion of large language models*](https://forum.effectivealtruism.org/s/8rYkpiFhbb4HsbzFc)*. As context for this post, I strongly recommend reading at least the* [*5-minute summa... | https://www.lesswrong.com/posts/WnvC5PwfLMhynPoBr/gpt-3-like-models-are-now-much-easier-to-access-and-deploy |
# Reflections on Trusting Trust & AI
**Reflections on Trusting Trust & AI**
======================================
The goal of the post is to highlight the necessity for mechanistic interpretability from an information security perspective:
### **Background**
In 1984, Ken Thompson authored a seminal paper on inform... | https://www.lesswrong.com/posts/BMnhDjJrix5BXE7yr/reflections-on-trusting-trust-and-ai |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.