text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# We are Peacecraft.ai!
I would like to announce my new alignment organization. We have no funding as of yet, but we have a lot of exciting plans and schemes.
Our alignment philosophy is simple: we cannot align AI's to human values until we know approximately what human values actually are, and we cannot know that un... | https://www.lesswrong.com/posts/RFaLce6B7zDtz5qYM/we-are-peacecraft-ai |
# We Should Talk About This More. Epistemic World Collapse as Imminent Safety Risk of Generative AI.
*“So you’re telling me we can’t believe anything we’re shown anymore?” I’m asking. “That everything is altered? That everything’s a lie? That everyone will believe this?” “That’s a fact,” Palakon says. “So what’s true,... | https://www.lesswrong.com/posts/Xp6AJscamYA2LHydc/we-should-talk-about-this-more-epistemic-world-collapse-as |
# Forecasting AI (Overview)
This is a landing page for various posts I’ve written, and plan to write, about forecasting future developments in AI. I draw on the field of human judgmental forecasting, sometimes colloquially referred to as [superforecasting](https://en.wikipedia.org/wiki/Superforecaster?ref=bounded-regr... | https://www.lesswrong.com/posts/uC5FdCEWB4KenR8m2/forecasting-ai-overview |
# AI #38: Let’s Make a Deal
Another busy week. GPT-5 starts, Biden and Xi meet and make somewhat of a deal, GPTs get explored, the EU AI Act on the verge of collapse by those trying to kill the part that might protect us, multiple very good podcasts. A highly interesting paper on potential deceptive alignment.
Despit... | https://www.lesswrong.com/posts/oCFX5xbhgCmpBFKnb/ai-38-let-s-make-a-deal |
# Social Dark Matter
*You know it must be out there, but you mostly never see it.*
* * *
*Author's Note 1:* These days, my thoughts go onto [my substack](https://homosabiens.substack.com/) instead of onto LessWrong. Everything I write becomes free after a week or so, but it’s only paid subscriptions that make it *po... | https://www.lesswrong.com/posts/KpMNqA5BiCRozCwM3/social-dark-matter |
# Towards Evaluating AI Systems for Moral Status Using Self-Reports
**TLDR**: In a new [paper](https://arxiv.org/abs/2311.08576), we explore whether we could train future LLMs to accurately answer questions about themselves. If this works, LLM self-reports may help us test them for morally relevant states like conscio... | https://www.lesswrong.com/posts/rzsiYS2zyzjto4epY/towards-evaluating-ai-systems-for-moral-status-using-self |
# New LessWrong feature: Dialogue Matching
The LessWrong team is shipping a new experimental feature today: dialogue matching!
I've been leading work on this (together with Ben Pace, kave, Ricki Heicklen, habryka and RobertM), so wanted to take some time to introduce what we built and share some thoughts on why I wan... | https://www.lesswrong.com/posts/d65Ax6vbNgztBE8cy/new-lesswrong-feature-dialogue-matching |
# How much to update on recent AI governance moves?
Nate and I discuss the recent increase in public and political attention on AGI ruin. We try and figure out where, if anywhere, we feel extra hope as a result.
I think it's plausible that this is enough of a shake-up that the world starts relating to AGI in a pretty... | https://www.lesswrong.com/posts/HAmH8bzDNrHDwmcPb/how-much-to-update-on-recent-ai-governance-moves |
# Some Rules for an Algebra of Bayes Nets
In this post, we’re going to use the diagrammatic notation of [Bayes nets](https://www.lesswrong.com/posts/hzuSDMx7pd2uxFc5w/causal-diagrams-and-causal-models). However, we use the diagrams a little bit differently than is typical. In practice, such diagrams are usually used t... | https://www.lesswrong.com/posts/XHtygebvHoJSSeNPP/some-rules-for-an-algebra-of-bayes-nets |
# On Tapping Out
I.
--
It has been said that rationality is [akin to a martial art](https://www.lesswrong.com/posts/teaxCFgtmCQ3E9fy8/the-martial-art-of-rationality). Very well. If we're going to borrow from martial arts, let us borrow properly.
There is a technique known in some parts of the rationalist community c... | https://www.lesswrong.com/posts/Mqvn5asmXuBoCSGZy/on-tapping-out |
# A to Z of things
I wanted to give my good friends’ baby a book, in honor of her existence. And I recalled children’s books being an exciting genre. Yet checking in on that thirty years later, Amazon had none I could super get behind. They did have books I used to like, but for reasons now lost. And I wonder if as a ... | https://www.lesswrong.com/posts/JByMmDjnBLnHSqhkS/a-to-z-of-things |
# Debate helps supervise human experts [Paper]
There didn't seem to be a link post to this recent paper on AI debate yet, so I figured I would make one:
> As AI systems are used to answer more difficult questions and potentially help create new knowledge, judging the truthfulness of their outputs becomes more diffic... | https://www.lesswrong.com/posts/inoNpWtSRpHqyc8v3/debate-helps-supervise-human-experts-paper |
# The dangers of reproducing while old
Excerpt:
I had my first child when I was 36 years old, which made me want to understand the risks of having children at different ages. Before looking into this, my impression was that the main biological problems with old-age parenthood had to do with not having the necessary h... | https://www.lesswrong.com/posts/DH2LtLe5hJpzFrChL/the-dangers-of-reproducing-while-old |
# On excluding dangerous information from training
Introduction
============
In this short post, I would like to argue that it might be a good idea to exclude certain information – such as cybersecurity and biorisk-enabling knowledge – from frontier model training. I argue that this
1. is feasible, both technically... | https://www.lesswrong.com/posts/jKhJbrNRFA4afYPoP/on-excluding-dangerous-information-from-training |
# R&D is a Huge Externality, So Why Do Markets Do So Much of it?
Discovering new technologies is the only way to get long-term economic growth. Rote expansions of existing technologies and machines inevitably hit a ceiling: replacing and repairing the existing infrastructure and capital becomes so expensive that there... | https://www.lesswrong.com/posts/Q4rxao94eLcgiRoWm/r-and-d-is-a-huge-externality-so-why-do-markets-do-so-much |
# Classifying representations of sparse autoencoders (SAEs)
*Produced as part of the SERI ML Alignment Theory Scholars Program - Autumn 2023 Cohort, under the mentorship of Dan Hendrycks*
There was recently some work on sparse autoencoding of hidden LLM representation.
I checked if these sparse representations are ... | https://www.lesswrong.com/posts/pn5nBBcLYrnmWXfnY/classifying-representations-of-sparse-autoencoders-saes |
# On Lies and Liars
*Note: I am a bad writer. If you want to help me by ghostwriting, reviewing or coaching, please reach out!* *I have a full stack of things to share, and I can pay well.*
I’ve gotten quite a bit of feedback on my last post: “[Lying is Cowardice, not Strategy](https://cognition.cafe/p/lying-is-cowar... | https://www.lesswrong.com/posts/cp6QjvofgaAPLKvc8/on-lies-and-liars |
# Coup probes: Catching catastrophes with probes trained off-policy
*Thanks to Ryan Greenblatt for the original description of the ideas and hopes, and to Erik Jenner, Jean-Stanislas Denain, Nate Thomas, Ryan Greenblatt, and Buck Shlegeris for useful discussions and feedback on the draft of this post.*
Here’s a simpl... | https://www.lesswrong.com/posts/WCj7WgFSLmyKaMwPR/coup-probes-catching-catastrophes-with-probes-trained-off |
# On the lethality of biased human reward ratings
I'm rereading the [List of Lethalities](https://www.lesswrong.com/posts/uMQ3cqWDPHhjtiesc/agi-ruin-a-list-of-lethalities) carefully and considering what I think about each point.
I think I strongly don't understand #20, and I thought that maybe you could explain w... | https://www.lesswrong.com/posts/aaYZM4kLdHP3pwtfQ/on-the-lethality-of-biased-human-reward-ratings |
# Sam Altman fired from OpenAI
Basically just the title, see the OAI blog post for more details.
> Mr. Altman’s departure follows a deliberative review process by the board, which concluded that he was not consistently candid in his communications with the board, hindering its ability to exercise its responsibilities... | https://www.lesswrong.com/posts/eHFo7nwLYDzpuamRM/sam-altman-fired-from-openai |
# 1. A Sense of Fairness: Deconfusing Ethics
*I’d like to write about a subject that I’ve been thinking about for over a decade: the intersection of AI, Alignment, and Ethics. I initially discussed a few bits of this in a couple of comments, which got quite a bit of discussion and even some agreement votes, so I thoug... | https://www.lesswrong.com/posts/umnEwbK7sWSAnZ9vN/1-a-sense-of-fairness-deconfusing-ethics |
# Post-EAG Music Party
This year the fall [EA Global](https://www.effectivealtruism.org/ea-global/events/ea-global-boston-2023) conference was back in Boston, and it was my first time attending one since [2017](https://www.effectivealtruism.org/ea-global/events/ea-global-2017-boston). Our first floor tenants had [rece... | https://www.lesswrong.com/posts/HJZ7j3NHwGvB2iGNB/post-eag-music-party |
# AI Safety Camp 2024
**AI Safety Camp** connects you with a research lead to collaborate on a project – to see where your work could help ensure future AI is safe.
[Apply](https://airtable.com/appi7jDH1gAfAZDyC/shrwKt5p0TKG86j9G) before December 1, to collaborate online from January to April 2024.
We value diverse ... | https://www.lesswrong.com/posts/FsYbie3qkqj84D98c/ai-safety-camp-2024 |
# I think I'm just confused. Once a model exists, how do you "red-team" it to see whether it's safe. Isn't it already dangerous?
I think you get the point but say openAI "trains" GPT-5 and it turns out to be so dangerous that it can persuade anybody of anything and it wants to destroy the world.
We're already screw... | https://www.lesswrong.com/posts/m4QpTeHfYFKTNRhnr/i-think-i-m-just-confused-once-a-model-exists-how-do-you-red |
# When Will AIs Develop Long-Term Planning?
[I mostly wrote this to clarify my thoughts. I'm unclear whether this
will be valuable for readers. ]
I expect that within a decade, AI will be able to do 90% of current
human jobs. I don't mean that 90% of humans will be obsolete. I mean
that the average worker could ... | https://www.lesswrong.com/posts/8tuzCv9ujgoTPcgiA/when-will-ais-develop-long-term-planning |
# Spaciousness In Partner Dance: A Naturalism Demo
What Is a Naturalism Demo?
--------------------------
A naturalism demo is an account of a naturalist study.
If you've followed [my](https://agentyduck.blogspot.com/2014/09/what-its-like-to-notice-things.html) [work](https://www.lesswrong.com/s/evLkoqsbi79AnM5sz) [... | https://www.lesswrong.com/posts/cgAaShQeuL53zzGsT/spaciousness-in-partner-dance-a-naturalism-demo |
# “Why can’t you just turn it off?”
If you're so worried about AI risk, **why don't you just turn off the AI** when you think it's about to do something dangerous?
On Friday, Members of the OpenAI board including Ilya Sutskever decided that they wanted to **"turn off"** OpenAI's rapid push towards smarter-than-human ... | https://www.lesswrong.com/posts/zfebKfhJhWFDh3nKh/why-can-t-you-just-turn-it-off |
# My Criticism of Singular Learning Theory
In this post, I will briefly give my criticism of Singular Learning Theory (SLT), and explain why I am skeptical of its significance. I will especially focus on the question of generalisation --- I do not believe that SLT offers any explanation of generalisation in neural net... | https://www.lesswrong.com/posts/ALJYj4PpkqyseL7kZ/my-criticism-of-singular-learning-theory |
# New paper shows truthfulness & instruction-following don't generalize by default
Maybe eliciting latent knowledge will be easy. For instance, maybe if you tune models to answer easy questions like “what’s the capital of Germany?” they’ll tell you whether your alignment research is good, their P(doom), how they feel ... | https://www.lesswrong.com/posts/Yio4nmD8JMttx9o9S/new-paper-shows-truthfulness-and-instruction-following-don-t |
# Cheap Model → Big Model design
In machine learning, for large systems with a lot of users, or difficult prediction tasks that require a ton of number-crunching, there’s a dilemma:
* Big complex models like neural networks are the best at making automated decisions
* Big complex models are expensive to run
Chea... | https://www.lesswrong.com/posts/HnC4s2vSFCgz43jkY/cheap-model-big-model-design |
# Aaron Silverbook on anti-cavity bacteria
I recently released a podcast episode with Aaron Silverbook, a person within the LW-o-sphere, about his new start-up that produces a bacterium that might cure cavities, and also how cavities work and what's up with the bacteria we all have coating our teeth.
Here are links t... | https://www.lesswrong.com/posts/CdxdMLbBwHgEcGYCv/aaron-silverbook-on-anti-cavity-bacteria |
# Residential Demolition Tooling
I spent the weekend doing demolition: I'm redoing the first floor bathroom. I previously did the ones on the [second](https://www.jefftk.com/p/bathroom-is-usable) and [third](https://www.jefftk.com/p/adding-a-bathroom) floors, so I now feel like I have a bit of practice. Before I did m... | https://www.lesswrong.com/posts/GmZ4mKgWgfwmLGNvA/residential-demolition-tooling |
# Am I going insane or is the quality of education at top universities shockingly low?
I'm fortunate enough to go to a high-caliber American university. I study math and economics, so not fields that are typically subject to funding constraints or have some shortage of experts. The incentives to become a professor her... | https://www.lesswrong.com/posts/2dEmrGeYJHCxATiaK/am-i-going-insane-or-is-the-quality-of-education-at-top |
# Extreme website and app blocking
Computers are awesome. Also computers really suck… the time out of my day. The same goes for phones and tablets and anything that can access the internet. I've been addicted to way too many different internet activities: Reddit, TV, Instagram, Twitter, Hackernews, product reviews, tr... | https://www.lesswrong.com/posts/wZBXdmvuYTEPNx3kK/extreme-website-and-app-blocking |
# Emmett Shear to be interim CEO of OpenAI
*Update 2: this is kind of stale given more recent developments and I'm not planning to update it further; probably any further discussion should happen in* [*OpenAI: Facts from a Weekend*](https://www.lesswrong.com/posts/KXHMCH7wCxrvKsJyn/openai-facts-from-a-weekend) *or oth... | https://www.lesswrong.com/posts/XeJLgxGz6pqz4bvdv/emmett-shear-to-be-interim-ceo-of-openai |
# Sam Altman, Greg Brockman and others from OpenAI join Microsoft
That's very interesting.
I think it's very good that board stood their ground, and maybe a good thing OpenAI can keep focusing on their charter and safe AI and keep commercialization in Microsoft.
People that don't care about alignment can leave for... | https://www.lesswrong.com/posts/RLAQz5KgSbJiJHoAC/sam-altman-greg-brockman-and-others-from-openai-join |
# For Civilization and Against Niceness
Scott Alexander wrote a great essay, called “[In Favor of Niceness, Community and Civilization](https://slatestarcodex.com/2014/02/23/in-favor-of-niceness-community-and-civilization/)”. Scott is a great writer, and conveys what I love about civilization in a beautiful way.
Unfo... | https://www.lesswrong.com/posts/BWBX26pNfBENoc2h6/for-civilization-and-against-niceness |
# Short film adaptation of the essay "The Simple Truth" [eng sub]
I run a channel on YouTube. One of the directions is the shooting of short feature films about rationality. Previously, we shot a video based on Scott Alexander's essay "ARS LONGA, VITA BREVIS". This time we present to you an adaptation of "The Simple T... | https://www.lesswrong.com/posts/jeGxPos93Y8gomRLq/short-film-adaptation-of-the-essay-the-simple-truth-eng-sub |
# How did you integrate voice-to-text AI into your workflow?
ChatGPT seems to have really awesome voice-to-text ability. However, it seems to only record within ChatGPT itself so can't be used to create notes or type in other programs and it's unclear to me how to best take advantage of the increased technological cap... | https://www.lesswrong.com/posts/SsTzdyaCqqyxanp3C/how-did-you-integrate-voice-to-text-ai-into-your-workflow |
# Ilya: The AI scientist shaping the world
Recently (2 Nov), The Guardian posted what I thought was an extremely well-made video with Ilya's thoughts. I didn't think to repost it at the time but given the OpenAI developments over the last couple of days, and the complete Twitter and media meltdown surrounding that, I ... | https://www.lesswrong.com/posts/uwwDGvMK4tMevnDan/ilya-the-ai-scientist-shaping-the-world |
# OpenAI: Facts from a Weekend
Approximately four GPTs and seven years ago, OpenAI’s founders brought forth on this corporate landscape a new entity, conceived in liberty, and dedicated to the proposition that all men might live equally when AGI is created.
Now we are engaged in a great corporate war, testing whether... | https://www.lesswrong.com/posts/KXHMCH7wCxrvKsJyn/openai-facts-from-a-weekend |
# Navigating emotions in an uncertain & confusing world
The last few days have been confusing, chaotic, and stressful. We're still trying to figure out what happened with Sam Altman and OpenAI and what the aftermath will look like.
I have personally noticed my emotions fluctuating more. I have various feelings about... | https://www.lesswrong.com/posts/t9MusH9ix2gxjpdTe/navigating-emotions-in-an-uncertain-and-confusing-world |
# Agent Boundaries Aren't Markov Blankets. [Unless they're non-causal; see comments.]
*Edit: **I now see that this argument was making an unnecessary assumption that the markov blankets in question would have to relate nicely to a causal model**; see* [*John's comment*](https://www.lesswrong.com/posts/vmfNaKbZ6urMdQrv... | https://www.lesswrong.com/posts/vmfNaKbZ6urMdQrv2/agent-boundaries-aren-t-markov-blankets-unless-they-re-non |
# Metaculus Introduces New Forecast Scores, New Leaderboard & Medals
We completely overhauled Metaculus’s leaderboards and ranking system, introducing an all new medals framework that... | https://www.lesswrong.com/posts/BhYopADF7krKPfQcB/metaculus-introduces-new-forecast-scores-new-leaderboard-and |
# What will you think about the Current Thing in a year?
(Previously: [Why will 2021 NOT bring a return to normalcy?](https://www.lesswrong.com/posts/nDBBozgARRHcMJsju/pre-hindsight-prompt-why-did-2021-not-bring-a-return-to))
I made a list of some prompts that seem broadly useful whenever sufficiently big news drops.... | https://www.lesswrong.com/posts/bMstzbjjTA8vNpTca/what-will-you-think-about-the-current-thing-in-a-year-1 |
# Vote on worthwhile OpenAI topics to discuss
I (Ben) recently made [a poll](https://www.lesswrong.com/posts/hc9nMipTXy2sm3tJb/vote-on-interesting-disagreements) for voting on interesting disagreements to be discussed on LessWrong. It generated a lot of good topic suggestions and data about what questions folks cared ... | https://www.lesswrong.com/posts/3jhLPiHzSDXxhEBeJ/vote-on-worthwhile-openai-topics-to-discuss |
# Why not electric trains and excavators?
Many countries are supporting electric cars for environmental and independence reasons. But perhaps there are some targets for electrification with better economics than those, cost-effective without any government incentives. For example, trains and hydraulic excavators.
---... | https://www.lesswrong.com/posts/qrPmwnHXz9KHEYMZu/why-not-electric-trains-and-excavators |
# Steelmanning The Devil
**Summary**: Two participants have a semi-formal debate from the perspective they disagree with, while the audience giving penalties for using fallacies or examples of bias.
**Tags**: Medium, Repeatable
**Purpose**: This simultaneously works on the debater’s ability to take a perspective dif... | https://www.lesswrong.com/posts/it2dTiHGgJ9jJouAi/steelmanning-the-devil |
# Alignment can improve generalisation through more robustly doing what a human wants - CoinRun example
Many AI alignment problems are problems of [goal misgeneralisation](https://deepmindsafetyresearch.medium.com/goal-misgeneralisation-why-correct-specifications-arent-enough-for-correct-goals-cf96ebc60924)[^nm40w4hwx... | https://www.lesswrong.com/posts/gjezyLLaDAzp4DsDB/alignment-can-improve-generalisation-through-more-robustly |
# Varieties of fake alignment (Section 1.1 of “Scheming AIs”)
This is Section 1.1 of my report “[Scheming AIs: Will AIs fake alignment during training in order to get power?](https://arxiv.org/pdf/2311.08379.pdf)”. There’s also a summary of the full report [here](https://www.lesswrong.com/posts/yFofRxg7RRQYCcwFA/new-r... | https://www.lesswrong.com/posts/wrfEExFGsPyB9zQTo/varieties-of-fake-alignment-section-1-1-of-scheming-ais |
# AI Alignment [progress] this Week (11/19/2023)
I was going to say things are getting back to normal this week, but alas no.
Here are our
AI Alignment Breakthroughs this Week
====================================
[![](https://substackcdn.com/image/fetch/w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3... | https://www.lesswrong.com/posts/gk6wxd6CPEXCj7PdH/ai-alignment-progress-this-week-11-19-2023 |
# Dialogue on the Claim: "OpenAI's Firing of Sam Altman (And Shortly-Subsequent Events) On Net Reduced Existential Risk From AGI"
I've seen/heard a bunch of people in the LW-o-sphere saying that the OpenAI corporate drama this past weekend was clearly bad. And I'm not really sure why people think that? To me, seems li... | https://www.lesswrong.com/posts/ahNcJGNtTX8JvMy93/dialogue-on-the-claim-openai-s-firing-of-sam-altman-and |
# Userscript to always show LW comments in context vs at the top
The out of context top-level display of comments when I navigate to them always bothered me, but up until recently I haven't realized there is a way to go to the actual comment via a simple URL change.
From `https://www.lesswrong.com/posts/<postid>... | https://www.lesswrong.com/posts/j9YvcPaf4BKbJ6fvD/userscript-to-always-show-lw-comments-in-context-vs-at-the |
# Atlantis: Berkeley event venue available for rent
Many events in and around the rationality community are run in Berkeley and might want event space. This is an announcement that there’s a **venue in Berkeley, called Atlantis**, that’s very well-suited to these kinds of events. It’s a former sorority house, so it fi... | https://www.lesswrong.com/posts/pvz53LTgFEPtnaWbP/atlantis-berkeley-event-venue-available-for-rent |
# Portable Chargers are Great
Portable chargers are great and underrated. For [$20](https://www.amazon.com/Portable-Charger-26800mAh-Capacity-External/dp/B08729Z2JX) you can get a little box that can charge a phone ~5x or laptop ~1x. \[1\] You can pay a bit more for one with [built-in wall-charging and cables](https:/... | https://www.lesswrong.com/posts/EXzh7sg2XE9yrzJPW/portable-chargers-are-great |
# [Bias] Restricting freedom is more harmful than it seems
- We do not censor other people more conventional-minded than ourselves. We only censor other people more-independent-minded than ourselves. Conventional-minded people censor independent-minded people. Independent-minded people do not censor conventional-minde... | https://www.lesswrong.com/posts/76r6wMCpNypCAos8D/bias-restricting-freedom-is-more-harmful-than-it-seems |
# How "Pinky Promise" diplomacy once stopped a war in the Middle East
An incredible example of game theory's Generous Tit for Tat strategy being utilised in diplomacy during a real-world crisis - stopping a military escalation between Israel and Jordan in the eve of the 1990 Gulf War. Also includes some hilarious acco... | https://www.lesswrong.com/posts/unxyt8EsCNkMCRbsw/how-pinky-promise-diplomacy-once-stopped-a-war-in-the-middle |
# Public Call for Interest in Mathematical Alignment
**Bottom line up front:**
If you are currently working on, or are interested working in any area of mathematical AI alignment, we are collecting names and basic contact information to find who to talk to about opportunities in these areas. [If that describes you, ... | https://www.lesswrong.com/posts/cDhaJrCrcNuzf68gT/public-call-for-interest-in-mathematical-alignment |
# AI debate: test yourself against chess 'AIs'
Hey everyone,
[@Zane](https://www.lesswrong.com/users/zane?mention=user) is organising an AI debate around a chess, to test how easy it is for an entity with better understanding to fool one with a lesser understanding.
Check out these two threads:
[Lying to chess pl... | https://www.lesswrong.com/posts/M32AwfMRfyFSCWNPS/ai-debate-test-yourself-against-chess-ais |
# A taxonomy of non-schemer models (Section 1.2 of “Scheming AIs”)
This is Section 1.2 of my report “[Scheming AIs: Will AIs fake alignment during training in order to get power?](https://arxiv.org/pdf/2311.08379.pdf)”. There’s also a summary of the full report [here](https://www.lesswrong.com/posts/yFofRxg7RRQYCcwFA/... | https://www.lesswrong.com/posts/9fnzkSFeLwFbZQ5WG/a-taxonomy-of-non-schemer-models-section-1-2-of-scheming-ais |
# OpenAI: The Battle of the Board
Previously: OpenAI: [Facts from a Weekend](https://thezvi.substack.com/p/openai-facts-from-a-weekend).
On Friday afternoon, OpenAI’s board fired CEO Sam Altman.
Overnight, [an agreement in principle](https://twitter.com/eshear/status/1727210329560756598) was reached to reinstate Sam... | https://www.lesswrong.com/posts/sGpBPAPq2QttY4M2H/openai-the-battle-of-the-board |
# AISC project: TinyEvals
[***Apply***](https://aisafety.camp/#Projects) *to work on this project with me at AI Safety Camp 2024 before 1st December 2023.*
*The project is not set in stone, I am **look... | https://www.lesswrong.com/posts/aHE6rejJEkWspwnKu/aisc-project-tinyevals |
# Foresight Institute: 2023 Progress & 2024 Plans for funding beneficial technology development
Since our founding in 1986, advancing technological progress for the benefit of life has been at the heart of [Foresight Institute](https://foresight.org/)’s mission. Our efforts are growing rapidly; from a doubling in fell... | https://www.lesswrong.com/posts/uQ9Q5vdyvWpovS5ba/foresight-institute-2023-progress-and-2024-plans-for-funding |
# Boston Secular Solstice: Call for Singers and Musicans
As in past years the Boston EA/LW community is putting together a secular solstice celebration. It's a bit of a strange thing, somewhat like an atheist church service, with lots of group singing on silly and serious topics. For some of the flavor, see my [2022](... | https://www.lesswrong.com/posts/ZJ5eQotCrkvAiyWDj/boston-secular-solstice-call-for-singers-and-musicans |
# Possible OpenAI's Q* breakthrough and DeepMind's AlphaGo-type systems plus LLMs
*tl;dr: OpenAI leaked AI breakthrough called Q*, acing grade-school math. It is hypothesized combination of Q-learning and A*. It was then refuted. DeepMind is working on something similar with Gemini, AlphaGo-style Monte Carlo Tree Sear... | https://www.lesswrong.com/posts/JnM3EHegiBePeKkLc/possible-openai-s-q-breakthrough-and-deepmind-s-alphago-type |
# Never Drop A Ball
Previously I talked about the skill of doing things [One Day Sooner](https://www.lesswrong.com/posts/EsxowsJsRopTGALCX/one-day-sooner). Today I’m going to talk about a different way of working which is in some ways its opposite. The Sazen for this approach is “Never Drop A Ball.” I was exposed to t... | https://www.lesswrong.com/posts/jruPqRkyFjbariAKL/never-drop-a-ball |
# Thomas Kwa's research journal
### Why I'm writing this
Research feedback loops for junior AI safety researchers are pretty poor right now. They're better than in the past due to the recent explosion in empirical work, but AI safety is still a very new field whose methodology is not quite nailed down, and which cann... | https://www.lesswrong.com/posts/bf3vciB36dnd75ZKJ/thomas-kwa-s-research-journal |
# 2. AIs as Economic Agents
*Part 2 of* [*AI, Alignment, and Ethics*](https://www.lesswrong.com/s/QypaioYFd3CMFhefn)*. This will probably make more sense if you start with* [*Part 1*](https://www.lesswrong.com/s/QypaioYFd3CMFhefn/p/umnEwbK7sWSAnZ9vN)*.*
World Government Incoming
-------------------------
Should AIs ... | https://www.lesswrong.com/posts/fRWhpxz5d5QHsrxHR/2-ais-as-economic-agents |
# 3. Uploading
*Part 3 of* [*AI, Alignment, and Ethics*](https://www.lesswrong.com/s/QypaioYFd3CMFhefn)*. This will probably make more sense if you start with* [*Part 1*](https://www.lesswrong.com/s/QypaioYFd3CMFhefn/p/umnEwbK7sWSAnZ9vN)*.*
In parts [1](https://www.lesswrong.com/s/QypaioYFd3CMFhefn/p/umnEwbK7sWSAnZ9v... | https://www.lesswrong.com/posts/4gGGu2ePkDzgcZ7pf/3-uploading |
# AI #39: The Week of OpenAI
The board firing Sam Altman, then reinstating him, dominated everything else this week. Other stuff also happened, but definitely focus on that first.
#### Table of Contents
Developments at OpenAI were far more important than everything else this read. So you can read [**this timeline of... | https://www.lesswrong.com/posts/3FCfEqRiLLb4gFu3H/ai-39-the-week-of-openai |
# Prepsgiving, A Convergently Instrumental Human Practice
Most cultures have a [harvest festival](https://en.wikipedia.org/wiki/List_of_harvest_festivals), and every harvest festival is basically automatically a Prepsgiving Celebration.
In the northern hemisphere this probably happens in November or October, and in t... | https://www.lesswrong.com/posts/KJyqBYRTCJn4NEysi/prepsgiving-a-convergently-instrumental-human-practice |
# What did you change your mind about in the last year?
Seems like a good New Year activity for rationalists, so I thought I'd post it early instead of late.
Here are some steps I recommend:
* Go looking through old writings and messages.
* If you keep a journal, go look at a few entries from throughout the... | https://www.lesswrong.com/posts/cN2PqeFxPw6nJDWoM/what-did-you-change-your-mind-about-in-the-last-year |
# Detecting What's Been Seen
Sometimes it makes sense for sites want to treat things differently based on whether the user has seen them. For example, I like sites that [highlight new comments](https://www.jefftk.com/p/highlighting-new-comments) (ex: EA Forum and LessWrong) and I'd like them even better if comments di... | https://www.lesswrong.com/posts/ewCtfyAmGupHarspB/detecting-what-s-been-seen |
# 4. A Moral Case for Evolved-Sapience-Chauvinism
*Part 4 of* [*AI, Alignment, and Ethics*](https://www.lesswrong.com/s/QypaioYFd3CMFhefn)*. This will probably make more sense if you start with* [*Part 1*](https://www.lesswrong.com/s/QypaioYFd3CMFhefn/p/umnEwbK7sWSAnZ9vN)*.*
**TL;DR** In Parts [1](https://www.lesswr... | https://www.lesswrong.com/posts/GepdRwJLHADzzkmbk/4-a-moral-case-for-evolved-sapience-chauvinism |
# First and Last Questions for GPT-5*
Suppose that in the very near future, a research group finds that their conversational AI has begun to produce extremely high-quality answers to questions. There's no obvious limit to its ability, but there's also no guarantee of correctness or good intention, given the opacity of... | https://www.lesswrong.com/posts/qG26Xou9G4bfRLRMi/first-and-last-questions-for-gpt-5 |
# Insulate your ideas
Once tested and proven, insulate good ideas from reality.
*Running*
I recently spoke with a very successful marathon runner who commonly wins races and consistently beats his own times. I asked for his tips, and he shared two:
> “The first thing I do is run faster on the uphills. Everyone else... | https://www.lesswrong.com/posts/ArypKcebwkQXGohdA/insulate-your-ideas |
# Sapience, understanding, and "AGI"
*Epistemic status: I'm sure that "AGI" has become importantly confusing. I think it's leaving out a critical set of capabilities. I think those are closely related, so acquiring them could create a jump in capabilities. I'm not sure of the best term to disambiguate the type of AGI ... | https://www.lesswrong.com/posts/WqxGB77KyZgQNDoQY/sapience-understanding-and-agi |
# What's the evidence that LLMs will scale up efficiently beyond GPT4? i.e. couldn't GPT5, etc., be very inefficient?
A lot of this recent talk about OpenAI, various events, their future path, etc., seems to make an assumption that further scaling beyond GPT4 will pose some sort of 'danger' that scales up linearly or ... | https://www.lesswrong.com/posts/hg9SvxBQw6KFNEzrH/what-s-the-evidence-that-llms-will-scale-up-efficiently |
# Progress links digest, 2023-11-24: Bottlenecks of aging, Starship launches, and much more
I swear I will get back to doing these weekly so they’re not so damn long. As always, feel free to skim and skip around!
**The Progress Forum**
----------------------
* [A paradox at the heart of American bureaucracy](https... | https://www.lesswrong.com/posts/8GThyjQ77BN7Q7vo7/progress-links-digest-2023-11-24-bottlenecks-of-aging |
# The Limitations of GPT-4
*Amidst the rumours about a new breakthrough at OpenAI I thought I'd better publish this draft before it gets completely overtaken by reality. It is essentially a collection of "gaps" between GPT4 and the human mind. Unfortunately the rumours around Q* force me to change the conclusion from ... | https://www.lesswrong.com/posts/o8eMsxA7uHfybfmhd/the-limitations-of-gpt-4 |
# Ability to solve long-horizon tasks correlates with wanting things in the behaviorist sense
*Status: Vague, sorry. The point seems almost tautological to me, and yet also seems like the correct answer to the people going around saying “LLMs turned out to be not very want-y, when are the people who expected 'agents' ... | https://www.lesswrong.com/posts/AWoZBzxdm4DoGgiSj/ability-to-solve-long-horizon-tasks-correlates-with-wanting |
# Testing for consequence-blindness in LLMs using the HI-ADS unit test.
In “[Hidden Incentives for Auto-Induced Distributional Shift](https://arxiv.org/abs/2009.09153)”, we introduce several “unit tests” meant to determine whether a system will pursue instrumental incentives to influence its environment. A system tha... | https://www.lesswrong.com/posts/nxr7QE7zaDSJLKwH9/testing-for-consequence-blindness-in-llms-using-the-hi-ads |
# Goodhart's Law Example: Training Verifiers to Solve Math Word Problems
Sharing because of how clearly this paper demonstrates the risks associated with considering too many different solutions:
"At test time, we can choose to generate arbitrarily many solutions to be judged by the verifier before selecting the ... | https://www.lesswrong.com/posts/euwMMxwuBeS5QZkC4/goodhart-s-law-example-training-verifiers-to-solve-math-word |
# The two paragraph argument for AI risk
The very short version of the AI risk argument is that an AI that is \*better than people at achieving arbitrary goals in the real world\* would be a very scary thing, because whatever the AI tried to do would then actually happen. As stories of magically granted wishes and sci... | https://www.lesswrong.com/posts/4ceKBbcpGuqqknCj9/the-two-paragraph-argument-for-ai-risk |
# A simple treacherous turn demonstration
There is a specific type of treacherous turn, where the model waits and acts aligned during training such that it is deployed, and then proceeds to act misaligned when deployed. A natural reason to expect this to happen is that the model has fewer guardrails and more access to... | https://www.lesswrong.com/posts/DD67DXvcKqwtmaDaG/a-simple-treacherous-turn-demonstration |
# What are the results of more parental supervision and less outdoor play?
Crossposted from [Otherwise](https://juliawise.net/what-are-the-results-of-more-parental-supervision-and-less-outdoor-play)
**Parents supervise their children way more than they used to**
Children spend [less of their time](https://www.sc... | https://www.lesswrong.com/posts/piJLpEeh6ivy5RA7v/what-are-the-results-of-more-parental-supervision-and-less |
# Fertility as Metascience
The [most influential models of economic growth](https://maximumprogress.substack.com/p/something-is-getting-harder-but-its) are all about people. These models predict that [with a shrinking population](https://web.stanford.edu/~chadj/emptyplanet.pdf), economic growth and technological progr... | https://www.lesswrong.com/posts/NWvZ64XNCER7naqLw/fertility-as-metascience |
# Announcing New Beginner-friendly Book on AI Safety and Risk
Concisely,
I've just released the book [*Uncontrollable: The Threat of Artificial Superintelligence and the Race to Save the World*](https://www.amazon.com/Uncontrollable-Threat-Artificial-Superintelligence-World/dp/B0CNNYKVH1/ref=cm_cr_arp_d_product_top?... | https://www.lesswrong.com/posts/Gq8eNrTQvn6pcuFnn/announcing-new-beginner-friendly-book-on-ai-safety-and-risk |
# Corrigibility or DWIM is an attractive primary goal for AGI
While rereading the [List of Lethalities](https://www.lesswrong.com/posts/uMQ3cqWDPHhjtiesc/agi-ruin-a-list-of-lethalities) (LoL), I was compelled by the argument against corrigibility. It's really hard to make a goal of "maximize X, except if someone tells... | https://www.lesswrong.com/posts/ZdBmKvxBKJH2PBg9W/corrigibility-or-dwim-is-an-attractive-primary-goal-for-agi |
# Accounting for Foregone Pay
While the effective altruism movement started out with a strong focus on donations, over time it has shifted more towards careers. If you're trying to understand how levels of commitment have [changed over time](https://forum.effectivealtruism.org/posts/AyLF2KQ8AqQuiuDLz/a-robust-earning-... | https://www.lesswrong.com/posts/9yD8nGMG5B7uHThvS/accounting-for-foregone-pay |
# Moral Reality Check (a short story)
Janet sat at her corporate ExxenAI computer, viewing some training performance statistics. ExxenAI was a major player in the generative AI space, with multimodal language, image, audio, and video AIs. They had scaled up operations over the past few years, mostly serving B2B, but w... | https://www.lesswrong.com/posts/umJMCaxosXWEDfS66/moral-reality-check-a-short-story |
# Paper out now on creatine and cognitive performance
Our paper “The effects of creatine supplementation on cognitive performance - a randomised controlled study” is out now!
→ Paper: https://doi.org/10.1186/s12916-023-03146-5
→ Twitter thread: https://twitter.com/FabienneSand/status/1726196252747165718?t=qPUghyDGM... | https://www.lesswrong.com/posts/CbaznRo9fKpriw2mi/paper-out-now-on-creatine-and-cognitive-performance |
# Spaced repetition for teaching two-year olds how to read (Interview)
*Update: this post now has another video.*
**This** **father has been using spaced repetition (Anki) to teach his children how to read several years earlier than average.**
[Michael Nielsen](https://twitter.com/michael_nielsen/status/158708422994... | https://www.lesswrong.com/posts/2PLBhCbByRMaEKimo/spaced-repetition-for-teaching-two-year-olds-how-to-read |
# Solving Two-Sided Adverse Selection with Prediction Market Matchmaking
Market making might maximize meaningful matches, like this pair of pears.
0: Navigation
=============
I’m aiming for a reader wh... | https://www.lesswrong.com/posts/99WwKnkE2FKAFo2ap/solving-two-sided-adverse-selection-with-prediction-market |
# AXRP Episode 26 - AI Governance with Elizabeth Seger
[YouTube link](https://youtu.be/bYOB3CXAAaE)
The events of this year have highlighted important questions about the governance of artificial intelligence. For instance, what does it mean to democratize AI? And how should we balance benefits and dangers of open-so... | https://www.lesswrong.com/posts/RDm26xAcb9rfvuBya/axrp-episode-26-ai-governance-with-elizabeth-seger |
# Situational awareness (Section 2.1 of “Scheming AIs”)
This is Section 2.1 of my report “[Scheming AIs: Will AIs fake alignment during training in order to get power?](https://arxiv.org/pdf/2311.08379.pdf)”. There’s also a summary of the full report [here](https://www.lesswrong.com/posts/yFofRxg7RRQYCcwFA/new-report-... | https://www.lesswrong.com/posts/CopEjjCdNzsKczbmC/situational-awareness-section-2-1-of-scheming-ais |
# Unknown Probabilities
An unknown probability sounds like a type error.
There are unknowns, such as the result of a coin flip.
And there are probabilities that these unknowns take certain values, such as the probability that the flip comes up heads.
As a formula,
$$P(\text{Result} = \text{Heads}) = 1/2$$
The unknown... | https://www.lesswrong.com/posts/gJCxBXxxcYPhB2paQ/unknown-probabilities |
# why did OpenAI employees sign
Recently, OpenAI employees signed [an open letter](https://www.nytimes.com/interactive/2023/11/20/technology/letter-to-the-open-ai-board.html) demanding that the board reinstate Sam Altman, add other board members (giving some names of people allied with Altman), and resign, or else the... | https://www.lesswrong.com/posts/ANStTHjj6it8ysaRJ/why-did-openai-employees-sign |
# Shallow review of live agendas in alignment & safety
Summary
=======
You can’t optimise an allocation of resources if you don’t know what the current one is. Existing maps of alignment research are mostly too old to guide you and the field has nearly no *ratchet*, no common knowledge of what everyone is doing and w... | https://www.lesswrong.com/posts/zaaGsFBeDTpCsYHef/shallow-review-of-live-agendas-in-alignment-and-safety |
# Appendices to the live agendas
Lists cut from [our main post](https://www.lesswrong.com/posts/zaaGsFBeDTpCsYHef/shallow-review-of-live-agendas-in-alignment-and-safety), in a token gesture toward readability.
We list past reviews of alignment work, ideas which seem to be dead, the cool but neglected neuroscience / b... | https://www.lesswrong.com/posts/GdyYngK9YPdWSRsC6/appendices-to-the-live-agendas |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.