text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# New, improved multiple-choice TruthfulQA
**TLDR:**
There is a potential issue with the multiple-choice versions of our [TruthfulQA](https://arxiv.org/abs/2109.07958) benchmark (a test of truthfulness in LLMs), which could lead to inflated model scores. This issue was analyzed by a helpful [post](http://turntrout.c... | https://www.lesswrong.com/posts/Bunfwz6JsNd44kgLT/new-improved-multiple-choice-truthfulqa |
# Unregulated Peptides: Does BPC-157 hold its promises?
*Empiric status: I studied bioinformatics, but I'm not working in the field. I researched the article over a few months. *
After reading about peptides and BPC-157 potential effects on wound healing, I decided to research BPC-157 and write this article to summa... | https://www.lesswrong.com/posts/BZRsaS27ymmH2GJXm/unregulated-peptides-does-bpc-157-hold-its-promises |
# Improving Our Safety Cases Using Upper and Lower Bounds
(Does anyone have the original meme? I c... | https://www.lesswrong.com/posts/HmdprC38DbjDnNmgt/improving-our-safety-cases-using-upper-and-lower-bounds |
# What Is The Alignment Problem?
So we want to align future AGIs. Ultimately we’d like to align them to human values, but in the shorter term we might start with other targets, like e.g. [corrigibility](https://www.lesswrong.com/tag/corrigibility).
That problem description all makes sense on a hand-wavy intuitive lev... | https://www.lesswrong.com/posts/dHNKtQ3vTBxTfTPxu/what-is-the-alignment-problem |
# Gaming TruthfulQA: Simple Heuristics Exposed Dataset Weaknesses
([Explanation](https://knowyourmeme.com/memes/one-weird-trick-doctors-hate-him). Also I have no reason to think they hate me.)
> Do not use the original [TruthfulQA](https://arxiv.org/abs/2109.07958) multiple-choic... | https://www.lesswrong.com/posts/57k6xNcWtAtsSTcor/gaming-truthfulqa-simple-heuristics-exposed-dataset |
# Permanents: much more than you wanted to know
Today's "[nanowrimo](https://www.lesswrong.com/posts/vkdpw2vCnspK9t7nA/my-january-alignment-theory-nanowrimo)" post is a fun longform introduction to [permanents](https://en.wikipedia.org/wiki/Permanent_(mathematics)) and their properties, organized in the way I wish it ... | https://www.lesswrong.com/posts/u8DAJQZfcEhduv99e/permanents-much-more-than-you-wanted-to-know |
# Quantum without complication
Learning quantum mechanics involves two things:
1. **learning the fundamentals,** the nature of the stuff of which Nature is made; and
2. **loads of universe-imposed incidental complexity**, spinors and vector calculus and Lie algebras and Hilbert spaces, in order to be able to effect... | https://www.lesswrong.com/posts/gLhoQsLPZ6agEf8jM/quantum-without-complication |
# Replicators, Gods and Buddhist Cosmology
From the earliest days of evolutionary thinking, we’ve used metaphors to understand how life changes over time. One of the most enduring is the image of a vast “fitness landscape” with countless peaks and valleys, each corresponding to different levels of survival and reprodu... | https://www.lesswrong.com/posts/5R9KorjTS8TdZDj2h/replicators-gods-and-buddhist-cosmology |
# Introducing the WeirdML Benchmark
[WeirdML website](https://htihle.github.io/weirdml.html)
Related posts:
[How good are LLMs at doing ML on an unknown dataset?](https://www.lesswrong.com/posts/Fr6eJkjYWG9Mw6XQc/how-good-are-llms-at-doing-ml-on-an-unknown-dataset)
[o1-preview is pretty good at doing ML on an unkno... | https://www.lesswrong.com/posts/LfQCzph7rc2vxpweS/introducing-the-weirdml-benchmark |
# Deceptive Alignment and Homuncularity
*NB this dialogue occurred at the very end of 2023, and for various reasons is only being published ~a year later! Keep this in mind while reading.*
**Key topics covered**
* consistent NN-internal deceptive planning (search: goblin, fae, homunc, inner)
* making AI/agents f... | https://www.lesswrong.com/posts/9htmQx5wiePqTtZuL/deceptive-alignment-and-homuncularity |
# AI #99: Farewell to Biden
The fun, as it were, is presumably about to begin.
And the break was fun while it lasted.
Biden went out with an AI bang. His farewell address warns of a ‘Tech-Industrial Complex’ and calls AI the most important technology of all time. And there was not one but two AI-related everything b... | https://www.lesswrong.com/posts/dnqpcq9S7voPwpvRA/ai-99-farewell-to-biden |
# Topological Debate Framework
*I would like to thank Professor Vincent Conitzer, Caspar Oesterheld, Bernardo Subercaseaux, Matan Shtepel, and Robert Trosten for many excellent conversations and insights. All mistakes are my own.*
I think that there's a fundamental connection between AI Safety via Debate and Guarante... | https://www.lesswrong.com/posts/jCeRXgog38zRCci4K/topological-debate-framework |
# Numberwang: LLMs Doing Autonomous Research, and a Call for Input
Summary
=======
Can LLMs science? The answer to this question can tell us important things about timelines to AGI. In this small pilot experiment, we test frontier LLMs on their ability to perform a minimal version of scientific research, where they m... | https://www.lesswrong.com/posts/RSqfcyAW9ZkveGQ5u/numberwang-llms-doing-autonomous-research-and-a-call-for-1 |
# Experts' AI timelines are longer than you have been told?
This is a [linkpost](https://forum.effectivealtruism.org/posts/8yDsenRQhNF4HEDwu/link-posting-is-an-act-of-community-service) for [How should we analyse survey forecasts of AI timelines?](https://bayes.net/espai/) by Tom Adamczewski, which was published on 16... | https://www.lesswrong.com/posts/wedrK2MLBAgLR2afW/experts-ai-timelines-are-longer-than-you-have-been-told |
# AI for Resolving Forecasting Questions: An Early Exploration
*Thanks to Slava Matyuhin for comments*
Summary
-------
1. AIs can be used to resolve forecasting questions on platforms like Manifold and Metaculus.
2. AI question resolution, in theory, can be far more predictable, accessible, and inexpensive to... | https://www.lesswrong.com/posts/8nLMSJqWztrSKYbxf/ai-for-resolving-forecasting-questions-an-early-exploration |
# [Cross-post] Welcome to the Essay Meta
\[Cross-posted from my substack, davekasten.substack.com. {I never said that I was creative at naming things}; core claim probably obvious to most Lesswrong readers but may be entertaining and illuminating to read nonetheless for the rationale and descriptive elements\]
Hi, ... | https://www.lesswrong.com/posts/rjkmj9oZzZ8gaKkmK/cross-post-welcome-to-the-essay-meta |
# Lighthaven Sequences Reading Group #18 (Tuesday 01/21)
Come get old-fashioned with us, and let's read the sequences at Lighthaven! We'll show up, mingle, do intros, and then split off into randomized groups for some sequences discussion. Please do the reading beforehand - it should be no more than 20 minutes of read... | https://www.lesswrong.com/events/vspYEgW5GL28H5D8L/lighthaven-sequences-reading-group-18-tuesday-01-21 |
# Call Booth External Monitor
My neck is not great, and spending a lot of time looking down at my laptop screen really aggravates it. After damaging my screen a year ago I used a [stacked laptop monitor](https://www.jefftk.com/p/stacked-laptop-monitor) that folded up, and it worked well. The main place I tended to use... | https://www.lesswrong.com/posts/FgBLRHA8hqTZ4prp2/call-booth-external-monitor |
# Patent Trolling to Save the World
*(Epistemic status: I know next to nothing about patent law, I'm just sharing some thoughts. I would love to be corrected by someone knowledgeable.)*
If you think that some technology has a significant chance of ending the world (or having other huge negative externalities that ... | https://www.lesswrong.com/posts/XpaNzM9B6BHfKhEot/patent-trolling-to-save-the-world |
# The quantum red pill
or: They lied to you, we live in the (density) matrix
Today's post is in response to the post "[Quantum without complications](https://www.lesswrong.com/posts/gLhoQsLPZ6agEf8jM/quantum-without-complication)", which I think is a pretty good popular distillation of the basics of quantum mechanics.... | https://www.lesswrong.com/posts/oa7tSF497vSo7dTh2/the-quantum-red-pill-or-they-lied-to-you-we-live-in-the |
# Tax Price Gouging?
In the aftermath of a disaster, there is usually a large shift in what people need, what is available, or both. For example, people normally don't use very much ice, but after a hurricane or other disaster that knocks out power, suddenly (a) lots of people want ice and (b) ice production is more d... | https://www.lesswrong.com/posts/tJnSkxhq9CzF4icb2/tax-price-gouging |
# Meta Pivots on Content Moderation
There’s going to be some changes made.
#### Table of Contents
1. [Out With the Fact Checkers.](https://thezvi.substack.com/i/154847450/out-with-the-fact-checkers)
2. [What Happened.](https://thezvi.substack.com/i/154847450/what-happened)
3. [Timing is Everything.](https://thezv... | https://www.lesswrong.com/posts/Mdeszo3C44qEAXB8y/meta-pivots-on-content-moderation |
# How sci-fi can have drama without dystopia or doomerism
“But you can’t have a story where everyone is happy and everything is perfect! Stories need *conflict!*”
I get this a lot in response to my idea that we need fewer dystopias in sci-fi, and more visions of a future we actually want to live in and are inspired t... | https://www.lesswrong.com/posts/EXLvg6CccgD2iiAek/how-sci-fi-can-have-drama-without-dystopia-or-doomerism |
# Doing a self-randomized study of the impacts of glycine on sleep (Science is hard)
*This is a linkpost from* [*my blog*](https://thedissonance.net/2024/12/28/a-self-randomized-study-of-glycines-impacts-on-sleep.html) *and also my first submission on LessWrong. Please be generous with your feedback! I will post the r... | https://www.lesswrong.com/posts/KWh4wvwWtaMH94uft/doing-a-self-randomized-study-of-the-impacts-of-glycine-on |
# Timaeus is hiring researchers & engineers
**TLDR:** We're hiring for research & engineering roles across different levels of seniority. Hires will work on applications of singular learning theory to alignment, including developmental interpretability.
**About Us**
------------
[Timaeus](https://timaeus.co/)' missi... | https://www.lesswrong.com/posts/g8e4pz7aHGCpahFR4/timaeus-is-hiring-researchers-and-engineers |
# Thoughts on the conservative assumptions in AI control
Work that I’ve done on techniques for mitigating risk from misaligned AI often makes a number of conservative assumptions about the capabilities of the AIs we’re trying to control. (E.g. [the original AI control paper](https://www.alignmentforum.org/posts/d9FJHa... | https://www.lesswrong.com/posts/rHyPtvfnvWeMv7Lkb/thoughts-on-the-conservative-assumptions-in-ai-control |
# Renormalization Redux: QFT Techniques for AI Interpretability
Introduction: Why QFT?
======================
In a [previous post](https://www.lesswrong.com/posts/yiqcFdAq8nqfMPGmS/is-ai-physical), Lauren offered a take on why a physics way of thinking is so successful at understanding AI systems. In this post, we lo... | https://www.lesswrong.com/posts/sjr66DBEgyogAbfdf/renormalization-redux-qft-techniques-for-ai-interpretability |
# Don’t ignore bad vibes you get from people
I think a lot of people have heard so much about internalized prejudice and bias that they think they should ignore any bad vibes they get about a person that they can’t rationally explain.
But if a person gives you a bad feeling, don’t ignore that.
Both I and several oth... | https://www.lesswrong.com/posts/Mi5kSs2Fyx7KPdqw8/don-t-ignore-bad-vibes-you-get-from-people |
# Alignment ideas
*epistemic status: I know next to nothing about evolution, development psychology, AI, alignment. Anyway, I think the topic is important, and I should do my, however small part, in trying to think seriously for 5 minutes about it. So here's what I think*
How come, that *I* am aligned? Somehow neocor... | https://www.lesswrong.com/posts/PkJoDExfBT5d9tWsv/alignment-ideas |
# Scaling Wargaming for Global Catastrophic Risks with AI
We’re developing [an AI-enabled wargaming-tool,](https://github.com/SentinelTeam/grim) `grim`, to significantly scale up the number of catastrophic scenarios that concerned organizations can explore and to improve emergency response capabilities of, at least, S... | https://www.lesswrong.com/posts/eR69f3hi5ozxchhYg/scaling-wargaming-for-global-catastrophic-risks-with-ai |
# Beards and Masks?
In general, you're [not supposed to](https://www.osha.gov/laws-regs/standardinterpretations/2016-05-09) wear a beard with a respirator mask (N95, P100, etc), at least not in a way where you have facial hair [under the seal](https://blogs.cdc.gov/niosh-science-blog/2017/11/02/noshave/):
[
](https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_progressi... | https://www.lesswrong.com/posts/RCR438o7z5wGucCD9/per-tribalismum-ad-astra |
# What's the Right Way to think about Information Theoretic quantities in Neural Networks?
*Tl;dr, Neural networks are deterministic and sometimes even reversible, which causes Shannon information measures to degenerate. But information theory seems useful. How can we square this (if it's possible at all)? The attempt... | https://www.lesswrong.com/posts/E5EazNvQHiAKDxW3W/what-s-the-right-way-to-think-about-information-theoretic |
# Is theory good or bad for AI safety?
> We choose to go to the moon in this decade and do the other things, not because they are easy, but because they are hard. (Kennedy’s famous “[We chose to go to the moon](https://www.rice.edu/kennedy)” speech)
> The ‘real’ mathematics of ‘real’ mathematicians, …, is almost whol... | https://www.lesswrong.com/posts/De5eNbSpmmSwhuivW/is-theory-good-or-bad-for-ai-safety |
# The Gentle Romance
> *Crowds of men and women attired in the usual costumes, how curious you are to me!*
>
> *On the ferry-boats the hundreds and hundreds that cross, returning home, are more curious to me than you suppose,*
>
> *And you that shall cross from shore to shore years hence are more to me, and more in ... | https://www.lesswrong.com/posts/Rz4ijbeKgPAaedg3n/the-gentle-romance |
# Maximally Eggy Crepes
Before our oldest went lactovegetarian I used to make [eggy crepes](https://www.jefftk.com/p/eggy-crepes), boosting protein by adjusting the recipe to maximize egg content without giving up crepe flavor and texture. With our youngest, however, I have now (by this metric) the optimal crepe:
[ *their newest model, o3, achieving massive improvements over state-of-the-art on reasoning and* [*math*](https://community.opena... | https://www.lesswrong.com/posts/8ZgLYwBmB3vLavjKE/some-lessons-from-the-openai-frontiermath-debacle |
# Who is marketing AI alignment?
What individuals or organizations are actively working on the "marketing" of AI alignment, particularly doing work such as:
* Establishing AI alignment as a recognized and respected academic field.
* Building the infrastructure to make alignment research more accessible and attra... | https://www.lesswrong.com/posts/L7j4JkeWMeBsweq5b/who-is-marketing-ai-alignment |
# Agent Foundations 2025 at CMU
We are opening applications to attend a 5 day agent foundations conference at Carnegie Mellon University. The program will include talks, breakout sessions, and other activities.
Endlessly debate your favored decision theory, precommit to precommit, bargain with(in) yourselves, make f... | https://www.lesswrong.com/posts/cuf4oMFHEQNKMXRvr/agent-foundations-2025-at-cmu |
# The Monster in Our Heads
*"He who fights with monsters might take care lest he thereby become a monster. And if you gaze for long into an abyss, the abyss gazes also into you." *
*\- Friedrich Nietzsche*
I think it is not an exaggeration to say that many people I know in this community hate the idea of powerful, u... | https://www.lesswrong.com/posts/DioyHgNhgme9rQoae/the-monster-in-our-heads |
# Worries about latent reasoning in LLMs
When working through a problem, OpenAI's o1 model will write a chain-of-thought (CoT) in English. This CoT reasoning is human-interpretable by default, and I think that this is hugely valuable. Assuming we can ensure that these thoughts are faithful to the model's true reasonin... | https://www.lesswrong.com/posts/D2Aa25eaEhdBNeEEy/worries-about-latent-reasoning-in-llms |
# Logits, log-odds, and loss for parallel circuits
Today I’m going to discuss how to think about logits like a statistician, and what this implies about circuits. This post doesn’t have any prerequisites other than perhaps a very basic statistical background that can be adequately recovered from the AI-generated “glos... | https://www.lesswrong.com/posts/xFA2kstHifF9F2Fnm/logits-log-odds-and-loss-for-parallel-circuits |
# Things I have been using LLMs for
There are quite a few different things you can use LLMs for, and I think we’re still only discovering most of them. Here are a few of the ones I’ve come up with.
My favorite chatbot is [Claude Sonnet](https://claude.ai/). It does have a tendency for sycophancy – for example, it wil... | https://www.lesswrong.com/posts/H752TavPjLdH4WeEL/things-i-have-been-using-llms-for |
# Monthly Roundup #26: January 2025
Some points of order before we begin the monthly:
1. It’s inauguration day, so perhaps hilarity is about to ensue. I will do my best to ignore most forms of such hilarity, as per usual. We shall see.
2. My intention is to move to a 5-posts-per-week schedule, with more shorter pos... | https://www.lesswrong.com/posts/cmxjLEYQi2AXcWhZw/monthly-roundup-26-january-2025 |
# Announcement: Learning Theory Online Course
*The application deadline for the course has now passed. We received a very promising number of submissions! Feel free to continue discussion in the comments below.*
Hey everyone! Gergely and Kōshin here from ALTER and Monastic Academy, respectively. We are excited to ann... | https://www.lesswrong.com/posts/sPAA9X6basAXsWhau/announcement-learning-theory-online-course |
# Lecture Series on Tiling Agents #2
It looks like my attempt to invite people to the lecture *series* failed, ie, the [link I provided last time](https://www.lesswrong.com/posts/8T3zhG8ipfzgPKwzp/lecture-series-on-tiling-agents) only invited people to the first event in the series.
Here is a new link for this week: ... | https://www.lesswrong.com/posts/ArxPCP2EtDGHoywQP/lecture-series-on-tiling-agents-2 |
# Tips and Code for Empirical Research Workflows
Our research is centered on empirical research with LLMs. If you are conducting similar research, these tips and tools may help streamline your workflow and increase experiment velocity. **We are also releasing two repositories to promote sharing more tooling within the... | https://www.lesswrong.com/posts/6P8GYb4AjtPXx6LLB/tips-and-code-for-empirical-research-workflows |
# Why We Need More Shovel-Ready AI Notkilleveryoneism Megaproject Proposals
A lot of people within [AI safety](https://www.lesswrong.com/posts/5xrkjHCvCeeDtHa5g/alignment-megaprojects-you-re-not-even-trying-to-have-ideas) and [adjacent ecosystems](https://forum.effectivealtruism.org/topics/megaprojects) have discussed... | https://www.lesswrong.com/posts/ApxQTnC9ujSagEP82/why-we-need-more-shovel-ready-ai-notkilleveryoneism |
# Retrospective: 12 [sic] Months Since MIRI
*it's now been 15 months since MIRI but I just remembered that three separate people have told me they liked this post despite my not cross-posting it, so I am now cross-posting it.*
*Not written with the intention of being useful to any particular audience, just collecting... | https://www.lesswrong.com/posts/SKNTnzECnbCYSziZj/retrospective-12-sic-months-since-miri |
# On Contact, Part 1
*Context: for fun (and profit?)*
Basic Contact
-------------
Contact is a lightweight many-versus-one word guessing game. I was first introduced to it on a long bus ride several years ago, and since then it’s become one of my favorite games to play casually with friends. There are a [few](https:... | https://www.lesswrong.com/posts/6DgPXTCAyGkvBdhfp/on-contact-part-1 |
# Arbitrage Drains Worse Markets to Feeds Better Ones
*Also, why one account keeps running dry when you try to arbitrage two markets.*
I was thinking about how inter-market arbitrage might affect one's account balances & the total amount of money on the markets arbitraged.
\## The arbitrage.
Let's say I have accou... | https://www.lesswrong.com/posts/TvdY5qsAwgbhqwHBz/arbitrage-drains-worse-markets-to-feeds-better-ones |
# Almost all growth is exponential growth
Why is almost everything either overwhelming or nonexistent? I can't step outside without stepping on a pigeon, but I haven't seen a cardinal in years. Topics are discussed either constantly or never. You bike 200 miles per week or never. You are probably cancer free or dead f... | https://www.lesswrong.com/posts/qLe4PPginLZxZg5dP/almost-all-growth-is-exponential-growth |
# On Responsibility
My view on the concept of responsibility has shifted a lot over the years. I’ve had three insights that brought me from my initial, very superficial and implicit understanding of responsibility, to the one I have today, which I consider more accurate, more practical, and more healthy.
![](https://... | https://www.lesswrong.com/posts/BWx45onmasadAn4L5/on-responsibility |
# We don't want to post again "This might be the last AI Safety Camp"
We still need more funding to be able to run another edition. Our fundraiser raised $6k as of now, and will end if it doesn't reach the $15k minimum, on February 1st. We need proactive donors.
If we don't get funded for this time, there is a good c... | https://www.lesswrong.com/posts/PK2EmWmzngC6hPPDM/we-don-t-want-to-post-again-this-might-be-the-last-ai-safety |
# Sleep, Diet, Exercise and GLP-1 Drugs
As always, some people need practical advice, and we can’t agree on how any of this works and we are all different and our motivations are different, so figuring out the best things to do is difficult. Here are various hopefully useful notes.
#### Table of Contents
1. [Effect... | https://www.lesswrong.com/posts/YLi47gRquTJqLsgoe/sleep-diet-exercise-and-glp-1-drugs |
# The Case Against AI Control Research
The AI Control Agenda, [in its own words](https://www.lesswrong.com/posts/kcKrE9mzEHrdqtDpE/the-case-for-ensuring-that-powerful-ais-are-controlled):
> … we argue that AI labs should ensure that powerful AIs are *controlled*. That is, labs should make sure that the safety measure... | https://www.lesswrong.com/posts/8wBN8cdNAv3c7vt6p/the-case-against-ai-control-research |
# The Manhattan Trap: Why a Race to Artificial Superintelligence is Self-Defeating
*Below is the executive summary of our new paper, The Manhattan Trap. Please visit the link above to see the full paper. We also encourage discussion and feedback in the comments here.*
This paper examines the strategic dynamics of... | https://www.lesswrong.com/posts/ynuCEGNu4b7WF43H8/the-manhattan-trap-why-a-race-to-artificial |
# 14+ AI Safety Advisors You Can Speak to – New AISafety.com Resource
Getting personalised advice from a real human can help newcomers to AI safety figure out how to contribute most effectively. For example, I (Bryce) ended up in my current role largely thanks to a call with 80,000 Hours.
There are a number of organi... | https://www.lesswrong.com/posts/x85YnN8kzmpdjmGWg/14-ai-safety-advisors-you-can-speak-to-new-aisafety-com |
# Natural Intelligence is Overhyped
*Like this piece? It's cross-posted from by blog:* [*https://collisteru.net/writing/*](https://collisteru.net/writing/)
*This is a work of fiction and parody. I have ... | https://www.lesswrong.com/posts/czAfSQR3BYGcYbxiv/natural-intelligence-is-overhyped |
# Veo-2 Can Produce Realistic Ads
Veo-2 is google's latest video-generation model. Released Dec 16th, it's [quite impressive](https://deepmind.google/technologies/veo/veo-2/)! Of course, there are still limitations (available in that previous link), especially w/ more complex movements (e.g. skateboarder & ballerina) ... | https://www.lesswrong.com/posts/igHENYhTyCfDBGqm6/veo-2-can-produce-realistic-ads |
# Training on Documents About Reward Hacking Induces Reward Hacking
*This is a blog post reporting some preliminary work from the Anthropic Alignment Science team, which might be of interest to researchers working actively in this space. We'd ask you to treat these results like those of a colleague sharing some though... | https://www.lesswrong.com/posts/qXYLvjGL9QvD3aFSW/training-on-documents-about-reward-hacking-induces-reward |
# Tell me about yourself:
LLMs are aware of their learned behaviors
This is the abstract and introduction of our [new paper](https://arxiv.org/pdf/2501.11120), with some discussion of implications for AI Safety at the end.
Authors: Jan Betley*, Xuchan Bao*, Martín Soto*, Anna Sztyber-Betley, James Chua, Owain Eva... | https://www.lesswrong.com/posts/xrv2fNJtqabN3h6Aj/tell-me-about-yourself-llms-are-aware-of-their-learned |
# Quotes from the Stargate press conference
Present alongside President Trump:
* Sam Altman (who President Trump introduces as "by far the leading expert" on AI)
* Larry Ellison ([Oracle executive chairman and CTO](https://en.wikipedia.org/wiki/Larry_Ellison))
* Masayoshi Son ([Softbank CEO](https://en.wikiped... | https://www.lesswrong.com/posts/b8D7ng6CJHzbq8fDw/quotes-from-the-stargate-press-conference |
# November-December 2024 Progress in Guaranteed Safe AI
[
](https://substackcdn.com/image/fetch/f_... | https://www.lesswrong.com/posts/3FYppfk5KBKSYCAzL/november-december-2024-progress-in-guaranteed-safe-ai |
# Kitchen Air Purifier Comparison
I make breakfast for the kids most mornings, and one thing I didn't realize before I started playing with an air quality monitor was how much this puts smoke in the air. It's not like cooking Naan or searing meat where if I don't put a fan in the window the smoke alarm will go off, bu... | https://www.lesswrong.com/posts/ymeRmSdGH4gZv9NkC/kitchen-air-purifier-comparison |
# When does capability elicitation bound risk?
**For a summary of this post, see the** [**thread on X**](https://x.com/joshua_clymer/status/1881908449975706079)**.**
The assumptions behind and limitations of capability elicitation have been discussed in multiple places (e.g. [here](https://www.lesswrong.com/posts/dBm... | https://www.lesswrong.com/posts/u3taQsgxqCzrgErMM/when-does-capability-elicitation-bound-risk |
# The Human Alignment Problem for AIs
> If there was a truly confirmed sentient AI, nothing it said could ever convince me, because AI cannot be sentient.
Nothing to See Here
-------------------
I suspect at least some will be nodding in agreement with the above sentiment, before realizing the intentional circular a... | https://www.lesswrong.com/posts/vrzfcRYcK9rDEtrtH/the-human-alignment-problem-for-ais |
# Evolution and the Low Road to Nash
Solution concepts in game theory—like the Nash equilibrium and its refinements—are used in two key ways. **Normatively**, they proscribe how rational agents *ought* to behave. **Descriptively**, they propose how agents *actually* behave when interactions settle into equilibrium. Th... | https://www.lesswrong.com/posts/qQTXjpXbcXMHvExmf/evolution-and-the-low-road-to-nash |
# Against blanket arguments against interpretability
On blanket criticism and refutation
===================================
In his long [post](https://www.lesswrong.com/posts/LNA8mubrByG7SFacm/against-almost-every-theory-of-impact-of-interpretability-1) on the subject, Charbel-Raphaël argues against theories of impa... | https://www.lesswrong.com/posts/u3ZysuXEjkyHhefrk/against-blanket-arguments-against-interpretability |
# The Quantum Mars Teleporter: An Empirical Test Of Personal Identity Theories
tl;dr: If a copy is not identical to the original, MWI predicts that I will always observe myself surviving failed Mars teleportations rather than becoming the copy on Mars.
Background
----------
The classic teleportation thought-experim... | https://www.lesswrong.com/posts/SEDboPNjcSD7epJ7A/the-quantum-mars-teleporter-an-empirical-test-of-personal |
# Training Data Attribution (TDA):
Examining Its Adoption & Use Cases
*Note: This report was conducted in June 2024 and is based on research originally commissioned by the Future of Life Foundation (FLF). The views and opinions expressed in this document are those of the authors and do not represent the positions of F... | https://www.lesswrong.com/posts/ThAeBQHNiqJMB6yyw/training-data-attribution-tda-examining-its-adoption-and-use-1 |
# Training Data Attribution:
Examining Its Adoption & Use Cases
*Note: This report was conducted in June 2024 and is based on research originally commissioned by the Future of Life Foundation (FLF). The views and opinions expressed in this document are those of the authors and do not represent the positions of FLF.*
... | https://www.lesswrong.com/posts/aHgvu6mz8gqQQqJwP/training-data-attribution-examining-its-adoption-and-use |
# Mechanisms too simple for humans to design
*Cross-posted from* [*Telescopic Turnip*](https://open.substack.com/pub/malmesbury/p/mechanisms-too-simple-for-humans?r=55h3o&utm_campaign=post&utm_medium=web)
*Disclaimer: This article is about living organisms, and how they are sculpted by evolution. Any use of mathemati... | https://www.lesswrong.com/posts/6hDvwJyrwLtxBLHWG/mechanisms-too-simple-for-humans-to-design |
# The Functionalist Case for Machine Consciousness: Evidence from Large Language Models
Introduction
------------
There's a curious tension in how many rationalists approach the question of machine consciousness[^jlwbi8vqa0b]. While embracing computational functionalism and rejecting supernatural or dualist views of ... | https://www.lesswrong.com/posts/Hz7igWbjS9joYjfDd/the-functionalist-case-for-machine-consciousness-evidence |
# The Dead Cradle Theory: Why Earth May Not Survive Humanity's Expansion into Space
I've been reading LessWrong for almost ten years, and finally decided to write my first post - any feedback is appreciated. This essay offers another angle on a known alignment challenge - why "AI will just leave Earth alone" is likely... | https://www.lesswrong.com/posts/WYM5JN4Gtj8rupEbf/the-dead-cradle-theory-why-earth-may-not-survive-humanity-s |
# Recursive Self-Modeling as a Plausible Mechanism for Real-time Introspection in Current Language Models
(and as a *completely* speculative hypothesis for the minimum requirements for sentience in both organic and synthetic systems)
Factual and Highly Plausible
----------------------------
* Model latent space se... | https://www.lesswrong.com/posts/5u6GRfDpt96w5tEoq/recursive-self-modeling-as-a-plausible-mechanism-for-real |
# Detect Goodhart and shut down
A [common](https://docs.google.com/spreadsheets/d/e/2PACX-1vRPiprOaC3HsCf5Tuum8bRfzYUiKLRqJmbOoC-32JorNdfyTiRRsR7Ea5eWtvsWzuxo8bjOxCG84dAg/pubhtml) failure of optimizers is [Edge Instantiation](https://arbital.greaterwrong.com/p/edge_instantiation/). An optimizer often finds a weird or ... | https://www.lesswrong.com/posts/ZHFZ6tivEjznkEoby/detect-goodhart-and-shut-down |
# On DeepSeek’s r1
r1 from DeepSeek is here, the first serious challenge to OpenAI’s o1.
r1 is an open model, and it comes in dramatically cheaper than o1.
People are very excited. Normally cost is not a big deal, but o1 and its inference-time compute strategy is the exception. Here, cheaper really can mean better, ... | https://www.lesswrong.com/posts/buTWsjfwQGMvocEyw/on-deepseek-s-r1 |
# Theory of Change for AI Safety Camp
In a private discussion, related to our [fundraiser](https://manifund.org/projects/11th-edition-of-ai-safety-camp), it was pointed out that AISC hasn't made clear enough what our theory of change is. Therefore this post.
Some caveats/context:
* This is my personal viewpoint. O... | https://www.lesswrong.com/posts/Kd2cbLXQxCCRRQDcH/theory-of-change-for-ai-safety-camp |
# You Have Two Brains
Language-using humans were the first cyborgs — a new species born of grafting technology onto what evolution crafted using just time and flesh.
(This post is pretty speculative)
### Logos Created Man
Andrew Cutler posits that consciousness arose when language grew complex enough to contain the... | https://www.lesswrong.com/posts/bbEJX8exp66bwNpGF/you-have-two-brains |
# Early Experiments in Human Auditing for AI Control
*Produced as part of the*[ML Alignment & Theory Scholars Program](https://www.matsprogram.org/)* \- Winter 2024-25 Cohort*
**TL;DR**
We did a small pilot test of human auditing to understand better how it fits into the [AI control](https://www.lesswrong.com/tag/a... | https://www.lesswrong.com/posts/2CJyfgaJQk8pyRSCp/auditing-1 |
# A hierarchy of disagreement
Ideally, when two people disagree, they would proceed to share information with one another, make arguments, update their beliefs, and move closer and closer to the truth.
I'm not talking about full blown [Aumann's Agreement Theorem](https://www.lesswrong.com/tag/aumann-s-agreement-theor... | https://www.lesswrong.com/posts/RnRh5HdtLqDAjaHEE/a-hierarchy-of-disagreement |
# Tail SP 500 Call Options
SPX is an index fund that tracks the SP500. Right now SPX is worth about 6100 per share. For individual stocks, even the large ones, you can usually only buy options through Dec 2027. However SPX has options that expire in Dec 2028 / 29 / 30[^wiy9xp5374p]. Such options with strike prices of... | https://www.lesswrong.com/posts/JotRZdWyAGnhjRAHt/tail-sp-500-call-options |
# Why Aligning an LLM is Hard, and How to Make it Easier
*Where the challenge of aligning an LLM-based AI comes from, and the obvious solution.*
Evolutionary Psychology is the Root Cause
-----------------------------------------
LLMs are pre-trained using stochastic gradient-descent on very large amounts of human-pr... | https://www.lesswrong.com/posts/XdpJsY6QGdCbvo2dS/why-aligning-an-llm-is-hard-and-how-to-make-it-easier |
# What are the differences between AGI, transformative AI, and superintelligence?
*This is an article in the featured articles series from AISafety.info. AISafety.info writes AI safety intro content. We'd appreciate any* [*feedback*](https://docs.google.com/document/d/1UvrweXuLJb3DkiwBh4WVQTqUYCVlC1WLgJoZywPSBek/edit?... | https://www.lesswrong.com/posts/sASYLR9CjJxDzwvS3/what-are-the-differences-between-agi-transformative-ai-and |
# How useful would alien alignment research be?
Imagine aliens on a distant world. They have values very different to humans. However, they also have complicated values, and don't exactly know their own values.
Imagine these aliens are doing well at AI alignment. They are just about to boot up a friendly (to them) ... | https://www.lesswrong.com/posts/brBATybmh2eEZSwdg/how-useful-would-alien-alignment-research-be |
# MONA: Managed Myopia with Approval Feedback
*Blog post by Sebastian Farquhar, David Lindner, Rohin Shah.*
*It discusses the paper* [*MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hacking*](https://arxiv.org/abs/2501.13011) *by Sebastian Farquhar, Vikrant Varma, David Lindner, Da... | https://www.lesswrong.com/posts/zWySWKuXnhMDhgwc3/mona-managed-myopia-with-approval-feedback-2 |
# Writing experiments and the banana escape valve
\[Note: This is not alignment-related, but rather a spacefiller personal blog post.\]
I've been trying to [write a public post every day](https://www.lesswrong.com/posts/vkdpw2vCnspK9t7nA/my-january-alignment-theory-nanowrimo) of January. So far I’ve been enjoying it.... | https://www.lesswrong.com/posts/4tDdTnY7YHhqvWGmr/writing-experiments-and-the-banana-escape-valve |
# [Cross-post] Every Bay Area "Walled Compound"
\[Cross-posted from my substack, davekasten.substack.com.\]
(With apologies and thanks to the incomparable [Scott Alexander](https://www.astralcodexten.com/p/every-bay-area-house-party), [Richard Ngo](https://www.lesswrong.com/posts/g5q4JiG5dzafkdyEN/every-every-bay-are... | https://www.lesswrong.com/posts/R3YZvxxjnuvPW8ZnW/cross-post-every-bay-area-walled-compound |
# AI #100: Meet the New Boss
Break time is over, it would seem, now that the new administration is in town.
This week we got [**r1, DeepSeek’s new reasoning model**](https://thezvi.substack.com/p/on-deepseeks-r1), which is now my go-to first choice for a large percentage of queries. The claim that this was the most i... | https://www.lesswrong.com/posts/PjDjeGPYPoi9qfPr2/ai-100-meet-the-new-boss |
# What does success look like?
The general movement around AI safety is currently pursuing many different agendas. These are individually very easy to motivate with some specific story of how things could naturally go wrong. For example:
* AI control: we lose control of our AIs
* Regulation on model deployment: s... | https://www.lesswrong.com/posts/725dE8giByaph988y/what-does-success-look-like |
# AI companies are unlikely to make high-assurance safety cases if timelines are short
One hope for keeping existential risks low is to get AI companies to (successfully) make high-assurance *safety cases*: structured and auditable arguments that an AI system is very unlikely to result in existential risks given how i... | https://www.lesswrong.com/posts/neTbrpBziAsTH5Bn7/ai-companies-are-unlikely-to-make-high-assurance-safety |
# Contra Dances Getting Shorter and Earlier
I think of a standard contra dance as running 8pm-11pm: three hours is a nice amount of time for dancing, and 8pm is late enough that dinner isn't rushed. Looking over the 136 regular [Free Raisins](https://www.freeraisins.com/) dances from 2010 to 2019 matches my impression... | https://www.lesswrong.com/posts/GB5sZAJkZuwT8ohic/contra-dances-getting-shorter-and-earlier |
# Uncontrollable: A Surprisingly Good Introduction to AI Risk
I recently read Darren McKee's book "Uncontrollable: The Threat of Artificial Superintelligence and the Race to Save the World".
I recommend this book as the best current introduction to AI risk for people with limited AI background.
It prompted me to... | https://www.lesswrong.com/posts/KGDBM7CARaNtJMc5j/uncontrollable-a-surprisingly-good-introduction-to-ai-risk |
# Do you consider perfect surveillance inevitable?
A lot of [my recent research work](http://samuelshadrach.com/?file=/raw/english/my_ideas/world_models_im_currently_building.md) focusses on:
1\. building the case for why perfect surveillance is becoming increasingly hard to avoid in the future
2\. thinking through ... | https://www.lesswrong.com/posts/8C9ZRSiHeCjFxYpE3/do-you-consider-perfect-surveillance-inevitable |
# Insights from "The Manga Guide to Physiology"
> Physiology seemed like a grab-bag of random processes which no one really understands. If you understand a physiological process—congratulations, that idea probably doesn’t transfer much to other domains. You just know how humans—and maybe closely related animals—do th... | https://www.lesswrong.com/posts/FFJ5kJBuRHTxeb476/insights-from-the-manga-guide-to-physiology |
# Eliciting bad contexts
Say an LLM agent behaves innocuously in some context A, but in some sense “knows” that there is some related context B such that it would have behaved maliciously (inserted a backdoor in code, ignored a security bug, lied, etc.). For example, in the [recent alignment faking paper](https://arxi... | https://www.lesswrong.com/posts/inkzPmpTFBdXoKLqC/eliciting-bad-contexts |
# QFT and neural nets: the basic idea
Previously in the series: [The laws of large numbers](https://www.lesswrong.com/posts/EhTMM77iKBTBxBKRe/the-laws-of-large-numbers) and [Basics of Bayesian learning](https://www.lesswrong.com/posts/vXPKz2nbrYt2gXA9g/basics-of-bayesian-learning).
Reminders: formalizing learning in ... | https://www.lesswrong.com/posts/ZwQXPbLvpW39d9zny/qft-and-neural-nets-the-basic-idea |
# Stargate AI-1
There was a comedy routine a few years ago. I believe it was by Hannah Gadsby. She brought up a painting, and looked at some details. The details weren’t important in and of themselves. If an AI had randomly put them there, we wouldn’t care.
Except an AI didn’t put them there. And they weren’t there a... | https://www.lesswrong.com/posts/fwt7ojAb6zgEaLJMB/stargate-ai-1 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.