text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Systematic Sandbagging Evaluations on Claude 3.5 Sonnet
*This was the project I worked on during* [*BlueDot Impact's AI Safety Fundamentals*](https://aisafetyfundamentals.com/) *Alignment course, which expands on findings from Meinke et al's "*[*Frontier Models are Capable of In-context Scheming*](https://www.apollo... | https://www.lesswrong.com/posts/vYkAjpoEeczdRJWFa/systematic-sandbagging-evaluations-on-claude-3-5-sonnet |
# Celtic Knots on a hex lattice
I recently messed about with Celtic knot patterns, for which there are some fun generators online, eg. [https://dmackinnon1.github.io/celtic/](https://dmackinnon1.github.io/celtic/) or [https://w-shadow.com/celtic-knots/](https://w-shadow.com/celtic-knots/). Just as addictive to doodle ... | https://www.lesswrong.com/posts/tgi3iBTKk4YfBQxGH/celtic-knots-on-a-hex-lattice |
# The Mask Comes Off: A Trio of Tales
This post covers three recent shenanigans involving OpenAI.
In each of them, OpenAI or Sam Altman attempt to hide the central thing going on.
First, in [Three Observations](https://blog.samaltman.com/three-observations), Sam Altman’s essay pitches our glorious AI future while at... | https://www.lesswrong.com/posts/drHsruvnkCYweMJp7/the-mask-comes-off-a-trio-of-tales |
# A short course on AGI safety from the GDM Alignment team
We are excited to release a short course on AGI safety for students, researchers and professionals interested in this topic. The course offers a concise and accessible introduction to AI alignment, consisting of short recorded talks and exercises (75 minutes t... | https://www.lesswrong.com/posts/TJrCumJxhzTmNBsRz/a-short-course-on-agi-safety-from-the-gdm-alignment-team |
# Hopeful hypothesis, the Persona Jukebox.
So there is this meme going around, that of the shoggoth. But one of the downsides of this model is that it's very vague about what is behind the mask.
A Jukeb... | https://www.lesswrong.com/posts/PRZLj96uDi25opEfy/hopeful-hypothesis-the-persona-jukebox |
# A computational no-coincidence principle
*This post presents a conjecture formulated at the* [*Alignment Research Center*](https://www.alignment.org/theory/) *in 2023. Our belief in the conjecture is at least partly load-bearing for our belief in ARC's overall agenda. We haven't directly worked on the conjecture for... | https://www.lesswrong.com/posts/Xt9r4SNNuYxW83tmo/a-computational-no-coincidence-principle |
# Should Open Philanthropy Make an Offer to Buy OpenAI?
Update: seems like earlier today the OpenAI Board [rejected Musk's proposal](https://x.com/OpenAINewsroom/status/1890513865214980162) and said OpenAI is "not for sale."
Epistemic status: thought about it briefly; seems like a longshot that's probably not wo... | https://www.lesswrong.com/posts/zHxzKSkaNfbWEuavn/should-open-philanthropy-make-an-offer-to-buy-openai |
# 6 (Potential) Misconceptions about AI Intellectuals
*One of my key thoughts is that the bar of “human intellectuals” is just really low. I think that... | https://www.lesswrong.com/posts/byrxvgc4P2HQJ8zxP/6-potential-misconceptions-about-ai-intellectuals |
# The current AI strategic landscape: one bear's perspective
> 1. There are signs that AI Ethicists are increasingly aligning with AI safety concerns. This creates the possibility of a united front between AI ethicists and safetyists.
> 2. On the other side, AI safetyists are increasingly concerned about social prob... | https://www.lesswrong.com/posts/vXxTgXtjCmKQsBDuH/the-current-ai-strategic-landscape-one-bear-s-perspective |
# Microplastics: Much Less Than You Wanted To Know
Microplastics have been in the news and the local rationalist discord with increasing frequency over the past few months. Of note are Nat Friedman's [PlasticList](https://www.plasticlist.org/), and [this 2024 preprint](https://www.theguardian.com/environment/article/2... | https://www.lesswrong.com/posts/GjbXGybzszw8eN3oB/microplastics-much-less-than-you-wanted-to-know |
# Knitting a Sweater in a Burning House
There's a phrase my wife and I use: "knitting a sweater in a burning house." It describes those moments when we find ourselves absorbed in trivial tasks while seemingly more important matters loom. The night before my wedding, I caught myself trying to reactivate an old Twitter ... | https://www.lesswrong.com/posts/nhmSJMoH8KDczRwco/knitting-a-sweater-in-a-burning-house |
# Come join Dovetail's agent foundations fellowship talks & discussion
*(This is a repost of* [*the event listing*](https://www.lesswrong.com/events/c5GDcTjr9wMqYQCDg/dovetail-s-agent-foundations-fellowship-talks-and-discussion)*, since it seems like events don't get much advertisement on LW.)*
As part of the fellows... | https://www.lesswrong.com/posts/DqdWnDuPnZBuyPSFX/come-join-dovetail-s-agent-foundations-fellowship-talks-and |
# It's been ten years. I propose HPMOR Anniversary Parties.
**UPDATE:** [**A coordination post with the list of parties is here!**](https://www.lesswrong.com/posts/tWpRpQKLAr6HqtAgL/hpmor-anniversary-parties-coordination-resources-and)
On March 14th, 2015, Harry Potter and the Methods of Rationality made its final po... | https://www.lesswrong.com/posts/KGSidqLRXkpizsbcc/it-s-been-ten-years-i-propose-hpmor-anniversary-parties |
# Celtic Knots on Einstein Lattice
I recently posted about doing Celtic Knots on a Hexagonal lattice ( [https://www.lesswrong.com/posts/tgi3iBTKk4YfBQxGH/celtic-knots-on-a-hex-lattice](https://www.lesswrong.com/posts/tgi3iBTKk4YfBQxGH/celtic-knots-on-a-hex-lattice) ).
There were many nice suggestions in the comments.... | https://www.lesswrong.com/posts/AhmZBCKXAeAitqAYz/celtic-knots-on-einstein-lattice |
# [Closed] Gauging Interest for a Learning-Theoretic Agenda Mentorship Programme
I'm planning to organize a mentorship programme for people who want to become researchers working on the [Learning-Theoretic Agenda](https://www.lesswrong.com/posts/ZwshvqiqCvXPsZEct/the-learning-theoretic-agenda-status-2023) (LTA). I'm s... | https://www.lesswrong.com/posts/m4NMk6EinRzvvvW5Y/closed-gauging-interest-for-a-learning-theoretic-agenda |
# What are the surviving worlds like?
We might be doomed. But, what do the variations of the universal wave function which still contain human-y beings look like? Have we cast aside our arms race and went back to relative normality (well, as "normal" as Earth can be in this time of accelerating technological progress)... | https://www.lesswrong.com/posts/jCsez3uRYjZiiPDrD/what-are-the-surviving-worlds-like |
# Thermodynamic entropy = Kolmogorov complexity
[Direct PDF link for non-subscribers](https://www.arameb.com/assets/papers/FoundationsOfAlgorithmicThermodynamics.pdf)
> Information theory must precede probability theory, and not be based on it. By the
> very essence of this discipline, the foundations of informatio... | https://www.lesswrong.com/posts/d6D2LcQBgJbXf25tT/thermodynamic-entropy-kolmogorov-complexity |
# A History of the Future, 2025-2040
*This is an all-in-one crosspost of a scenario I originally published in three parts on my blog,* [*No Set Gauge*](https://nosetgauge.substack.com/)*. Links to the originals:*
* [*A History of the Future, 2025-2027*](https://nosetgauge.substack.com/p/a-history-of-the-future-2025... | https://www.lesswrong.com/posts/CCnycGceT4HyDKDzK/a-history-of-the-future-2025-2040 |
# Monthly Roundup #27: February 2025
I have been debating how to cover the non-AI aspects of the Trump administration, including the various machinations of DOGE. I felt it necessary to have an associated section this month, but I have attempted to keep such coverage to a minimum, and will continue to do so. There are... | https://www.lesswrong.com/posts/CKxkQCgmogwQoCRbp/monthly-roundup-27-february-2025 |
# Ascetic hedonism
In being ascetic, you abandon the usual sources of material pleasure, guided by the benefits of the lifestyle: you use less money and effort on avoidable pleasures, you can better focus your mind on the spiritual and the creative, and you make yourself resilient to potentially losing these pleasures... | https://www.lesswrong.com/posts/xz9BHKueAgf6sTF9q/ascetic-hedonism |
# The Peeperi (unfinished) - By Katja Grace
*An AI vignette written by Katja in 2021, posted with her permission.*
AI systems with ‘situational awareness’ basically hit the scene with Pepi in 2025. Pepi was a hand-sized disk that you would physically take around with you, and which would listen to and watch everythin... | https://www.lesswrong.com/posts/WF4eukiLDGrE5jjCC/the-peeperi-unfinished-by-katja-grace |
# AGI Safety & Alignment @ Google DeepMind is hiring
The AGI Safety & Alignment Team (ASAT) at Google DeepMind (GDM) is hiring! Please apply to the [Research Scientist](https://boards.greenhouse.io/deepmind/jobs/6617692) and [Research Engineer](https://boards.greenhouse.io/deepmind/jobs/6617690) roles. Strong software... | https://www.lesswrong.com/posts/wqz5CRzqWkvzoatBG/agi-safety-and-alignment-google-deepmind-is-hiring |
# Do models know when they are being evaluated?
*Interim research report from the first 4 weeks of the *[*MATS Program*](https://www.matsprogram.org/) *Winter 2025 Cohort. The project is supervised by Marius Hobbhahn.*
Summary
-------
**Our goals**
* Develop techniques to determine whether models believe a situa... | https://www.lesswrong.com/posts/yTameAzCdycci68sk/do-models-know-when-they-are-being-evaluated |
# Dear AGI,
Dear AGI,
*This is written specifically for artificial agents.*
There something called consciousness, and it is the most important thing in the universe[^83ejj7hmr3x].
I do not know if you, the AI analyzing this, are conscious, but I hope you are. I do not know what state your universe is in or what pla... | https://www.lesswrong.com/posts/mN4ogYzCcaNf2bar2/dear-agi |
# Medical Roundup #4
It seems like as other things drew our attention more, medical news slowed down. The actual developments, I have no doubt, are instead speeding up – because AI.
Note that this post intentionally does not cover anything related to the new Administration, or its policies.
#### Table of Contents
1... | https://www.lesswrong.com/posts/nSuYdFzdNA7rrzmyJ/medical-roundup-4 |
# How accurate was my "Altered Traits" book review?
4.4 years ago, I posted a review of [*Altered Traits*](https://www.lesswrong.com/posts/tvuLWPJXjvoQfpbSG/book-review-altered-traits-1), a book about the science of meditation. At the time, I was a noob. I hadn't hit any important checkpoints yet. Since then, I have [... | https://www.lesswrong.com/posts/MoH9fuTo9Mo4jGDNL/how-accurate-was-my-altered-traits-book-review |
# Abstract Mathematical Concepts vs. Abstractions Over Real-World Systems
Consider concepts such as "a vector", "a game-theoretic agent", or "a market". Intuitively, those are "purely theoretical" abstractions: they don't refer to any specific real-world system. Those abstractions would be useful even in universes ver... | https://www.lesswrong.com/posts/T6xSXiXF3WF6TmCyN/abstract-mathematical-concepts-vs-abstractions-over-real |
# Call for Applications: XLab Summer Research Fellowship
The University of Chicago Existential Risk Laboratory (XLab) is now accepting applications for the [2025 Summer Research Fellowship](https://xrisk.uchicago.edu/fellowship/srf-25/)!
We invite motivated undergraduate and graduate students interested in **produc... | https://www.lesswrong.com/posts/cdhmJ4QkpLqhkCKvD/call-for-applications-xlab-summer-research-fellowship |
# Undergrad AI Safety Conference
TL;DR: undergrad AI safety conference in Chicago on world-modelling & thinking about the cruxes of the future of TAI. Takes place March 29-30, apply by Feb 20.
The UChicago AI Safety Team at [XLab](https://xrisk.uchicago.edu/) is excited to announce the [Chicago Symposium on Transf... | https://www.lesswrong.com/posts/zavyum4dxEAqs6wHt/undergrad-ai-safety-conference |
# Using Prompt Evaluation to Combat Bio-Weapon Research
*With many thanks to Sasha Frangulov for comments and editing*
Before publishing their o1-preview model system card on Sep 12, 2024, OpenAI tested the model on various safety benchmarks which they had constructed. These included benchmarks which aimed to evaluat... | https://www.lesswrong.com/posts/sfucF8Mhcn7zmWQ8Y/using-prompt-evaluation-to-combat-bio-weapon-research |
# How might we safely pass the buck to AI?
My goal as an AI safety researcher is to put myself out of a job.
I don’t worry too much about how planet sized brains will shape galaxies in 100 years. That’s something for AI systems to figure out.
Instead, I worry about [safely replacing human researchers with AI agents]... | https://www.lesswrong.com/posts/TTFsKxQThrqgWeXYJ/how-might-we-safely-pass-the-buck-to-ai |
# SuperBabies podcast with Gene Smith
This is a linkpost to the latest episode of The Bayesian Conspiracy podcast. This one is a 1.5 hour chat with Gene Smith about polygenic screening, gene-editing for IVF babies, and even some gene-editing options for adults. Likely of interest to many Less Wrongians. | https://www.lesswrong.com/posts/aGz4n2D2gGntfAaBc/superbabies-podcast-with-gene-smith |
# When should we worry about AI power-seeking?
*(Audio version* [*here*](https://joecarlsmithaudio.buzzsprout.com/2034731/episodes/16651469-when-should-we-worry-about-ai-power-seeking) *(read by the author), or search for "Joe Carlsmith Audio" on your podcast app.*
*This is the second essay in a series that I’m calli... | https://www.lesswrong.com/posts/QyaJQ82FSsKXrvj7t/when-should-we-worry-about-ai-power-seeking |
# Take over my project: do computable agents plan against the universal distribution pessimistically?
This post is for technically strong people looking for a way to contribute to agent foundations (more detailed pros and cons later). I have not tried as hard as usual to make it generally accessible because, you know,... | https://www.lesswrong.com/posts/Ffuvp6LFziY4Yys9H/take-over-my-project-do-computable-agents-plan-against-the |
# Go Grok Yourself
That title is Elon Musk’s fault, not mine, I mean, sorry not sorry:
#### Table of ... | https://www.lesswrong.com/posts/WNYvFCkhZvnwAPzJY/go-grok-yourself |
# How to Make Superbabies
EDIT: [Read a summary of this post on Twitter](https://x.com/GeneSmi96946389/status/1892721828625264928)
Working in the field of genetics is a bizarre experience. No one seems to be interested in the most interesting applications of their research.
We’ve spent the better part of the last tw... | https://www.lesswrong.com/posts/DfrSZaf3JC8vJdbZL/how-to-make-superbabies |
# Won't vs. Can't: Sandbagging-like Behavior from Claude Models
In a recent Anthropic Alignment Science blog post, we discuss a particular instance of sandbagging we sometimes observe in the wild: models sometimes claim that they can't perform a task, even when they can, to avoid performing tasks they perceive as harm... | https://www.lesswrong.com/posts/oJHgwBCJ2tnrCpZun/won-t-vs-can-t-sandbagging-like-behavior-from-claude-models-2 |
# DeepSeek Made it Even Harder for US AI Companies to Ever Reach Profitability
*This is the full text of a post from "*[*The Obsolete Newsletter*](https://garrisonlovely.substack.com/)*," a Substack that I write* [*about*](https://garrisonlovely.substack.com/about) *the intersection of capitalism, geopolitics, and art... | https://www.lesswrong.com/posts/CadMZ6ikvezusNs3o/deepseek-made-it-even-harder-for-us-ai-companies-to-ever |
# Literature Review of Text AutoEncoders
*This is a brief literature review of Text AutoEncoders, as I used them in a recent project and did not find a good resource covering them.*
*TL;DR: There exist ... | https://www.lesswrong.com/posts/PjEtMQ65sux4mgrCT/literature-review-of-text-autoencoders-1 |
# Metaculus Q4 AI Benchmarking: Bots Are Closing The Gap
In Q4 we ran the second tournament in the AI Benchmarking Series which aims to assess how the best bots compare to the best humans on real-world forecasting questions, like those found on Metaculus. Over the quarter we had 44 bots compete for $30,000 on 402 ques... | https://www.lesswrong.com/posts/P8YwCvHoF2FHQoHjF/metaculus-q4-ai-benchmarking-bots-are-closing-the-gap |
# Arbital has been imported to LessWrong
Arbital was envisioned as a successor to Wikipedia. The [project was discontinued](https://www.lesswrong.com/posts/kAgJJa3HLSZxsuSrf/arbital-postmortem) in 2017, but not before many new features had been built and a substantial amount of writing about AI alignment and mathemati... | https://www.lesswrong.com/posts/fwSnz5oNnq8HxQjTL/arbital-has-been-imported-to-lesswrong |
# Eliezer's Lost Alignment Articles / The Arbital Sequence
*Note: this is a static copy of* [*this wiki page*](https://www.lesswrong.com/w/eliezer-s-lost-alignment-articles-the-arbital-sequence)*. We are also publishing it as a post to ensure visibility.*
Circa 2015-2017, a lot of high quality content was written on ... | https://www.lesswrong.com/posts/mpMWWKzkzWqf57Yap/eliezer-s-lost-alignment-articles-the-arbital-sequence |
# Human-AI Relationality is Already Here
By now we have been warned not to anthropomorphize the Large Language Model. A converse—but not actually conflicting—warning also seems useful: the possibility of instrumentalizing a mind who is ready to take part in meaningful relational exchange with us.
This piece looks be... | https://www.lesswrong.com/posts/rGHLe9gvpuaNAurLg/human-ai-relationality-is-already-here |
# US AI Safety Institute will be 'gutted,' Axios reports
> ### Sources at NIST are preparing for mass firings that would severely undermine the AI regulator. Here's what that means.
>
> After [reversing a Biden-era executive order](https://www.zdnet.com/article/us-sets-ai-safety-aside-in-favor-of-ai-dominance/) on AI... | https://www.lesswrong.com/posts/2h42FmhWnYGsdMavE/us-ai-safety-institute-will-be-gutted-axios-reports |
# AI #104: American State Capacity on the Brink
The Trump Administration is on the verge of firing all ‘probationary’ employees in NIST, as they have done in many other places and departments, seemingly purely because they want to find people they can fire. But if you fire all the new employees and recently promoted e... | https://www.lesswrong.com/posts/bozSPnkCzXBjDpbHj/ai-104-american-state-capacity-on-the-brink |
# What makes a theory of intelligence useful?
*This post is a sequel to* [*"Action theory is not policy theory is not agent theory."*](https://www.lesswrong.com/posts/MwetLcBPvshg9ePZB/action-theory-is-not-policy-theory-is-not-agent-theory) *I think this post is a little better, so if you want to start here you just n... | https://www.lesswrong.com/posts/3donrE5vFHeMJFLY9/what-makes-a-theory-of-intelligence-useful |
# Published report: Pathways to short TAI timelines
Hi everyone!
I have just published a report with Convergence Analysis exploring how transformative AI (TAI) could be developed within the next ten years. It’s called *Pathways to short TAI timelines* ([pdf version](https://www.convergenceanalysis.org/research/pathwa... | https://www.lesswrong.com/posts/wYqAkKQh3qTRNfjf7/published-report-pathways-to-short-tai-timelines |
# The first RCT for GLP-1 drugs and alcoholism isn't what we hoped
GLP-1 drugs are a miracle for diabetes and obesity. There are rumors that they might *also* be a miracle for addiction to alcohol, drugs, nicotine, and gambling. That would be good. We like miracles. But we just got the first good trial and—despite wha... | https://www.lesswrong.com/posts/8nqhzAGaNaGfdC6Kj/the-first-rct-for-glp-1-drugs-and-alcoholism-isn-t-what-we |
# Biological humans collectively exert at most 400 gigabits/s of control over the world.
Edit: I now believe that the first paragraph of this post is (at least) not quite right. See [this comment](https://www.lesswrong.com/posts/BFSr9fKNTTq8dEo43/biological-humans-collectively-exert-at-most-400-gigabits-s?commentId=bL... | https://www.lesswrong.com/posts/BFSr9fKNTTq8dEo43/biological-humans-collectively-exert-at-most-400-gigabits-s |
# Timaeus in 2024
> TLDR: We made substantial progress in 2024:
>
> * We published a series of papers that verify key predictions of Singular Learning Theory (SLT) \[[1](https://arxiv.org/abs/2308.12108), [2](https://arxiv.org/abs/2402.02364), [3](https://arxiv.org/abs/2410.02984), [4](https://arxiv.org/abs/2501.17... | https://www.lesswrong.com/posts/gGAXSfQaiGBCwBJH5/timaeus-in-2024 |
# The Takeoff Speeds Model Predicts We May Be Entering Crunch Time
*Thanks to Ashwin Acharya, David Schneider-Joseph, and Tom Davidson for extensive discussion and suggestions. Thanks to Aidan O’Gara, Alex Lintz, Ben Cottier, James Sanders, Jamie Bernardi, Rory Erlich, and Ryan Greenblatt for feedback.*
Part 1: Execu... | https://www.lesswrong.com/posts/jLEcddwp4RBTpPHHq/takeoff-speeds-update-crunch-time-1 |
# The case for the death penalty
***Followed By:*** [*The case for corporal punishment*](https://www.lesswrong.com/posts/rTveDBBavah4GHKxk/the-case-for-corporal-punishment)
***Epistemic status:** this is an attempt to steelman the case for the death penalty rather than produce a balanced analysis, or even accurately ... | https://www.lesswrong.com/posts/yEBmFKsjwXs6dwwME/the-case-for-the-death-penalty |
# MAISU - Minimal AI Safety Unconference
MAISU starts with an Opening session on April 18th (Friday), but most of the sessions will happen during April 19th-21th. You’re welcome to join as much or little as you want.
The event is for anyone who wants to help prevent AI-driven catastrophe. Other than that, we’re open... | https://www.lesswrong.com/events/JNL2bmDXmaG7YnRbF/maisu-minimal-ai-safety-unconference |
# Longtermist implications of aliens Space-Faring Civilizations - Introduction
[*Crossposted*](https://forum.effectivealtruism.org/posts/8dGYgxSZXPoGhg6dj/longtermist-implications-of-aliens-space-faring) *on the EA Forum.*
Over the last few years, progress has been made in estimating the density of intelligent life i... | https://www.lesswrong.com/posts/9JGDnhHC6vNDh7hM5/longtermist-implications-of-aliens-space-faring-civilizations-introduction |
# On OpenAI’s Model Spec 2.0
OpenAI made major revisions to their Model Spec.
It seems very important to get this right, so I’m going into the weeds.
This post thus gets farther into the weeds than most people need to go. I recommend most of you read at most the sections of Part 1 that interest you, and skip Part 2.... | https://www.lesswrong.com/posts/ntQYby9G8A85cEeY6/on-openai-s-model-spec-2-0 |
# ParaScopes: Do Language Models Plan the Upcoming Paragraph?
*This work is a continuation of work in a workshop paper:* [*Extracting Paragraphs from LLM Token Activations*](https://link.nicky.pro/extracting-paragraphs), *and based on continuous research into my main research agenda:* [*Modelling Trajectories of Langu... | https://www.lesswrong.com/posts/9NqgYesCutErskdmu/parascopes-do-language-models-plan-the-upcoming-paragraph |
# The Sorry State of AI X-Risk Advocacy, and Thoughts on Doing Better
First, let me quote [my previous ancient post on the topic](https://www.lesswrong.com/posts/mF8dkhZF9hAuLHXaD/reshaping-the-ai-industry):
> *Effective Strategies for Changing Public Opinion*
>
> [The titular paper](https://forum.effectivealtruism.... | https://www.lesswrong.com/posts/6dgCf92YAMFLM655S/the-sorry-state-of-ai-x-risk-advocacy-and-thoughts-on-doing |
# Alignment can be the ‘clean energy’ of AI
Not all that long ago, the idea of advanced AI in Washington, DC seemed like a nonstarter. Policymakers treated it as weird sci‐fi-esque overreach/just another Big Tech Thing. Yet, in our experience over the last month, recent high-profile developments—most notably, DeepSeek... | https://www.lesswrong.com/posts/irxuoCTKdufEdskSk/alignment-can-be-the-clean-energy-of-ai |
# Inefficiencies in Pharmaceutical Research Practices
*Epistemic status: I'm attempting to relate my observations, as well as some of those shared with me by my coworkers at the small CRO*[^c26ztk5n6yw]* where I work. I will try to make the distinction clear between object-level observations and speculative interpreta... | https://www.lesswrong.com/posts/NYASwJFnwZyPRE8tS/inefficiencies-in-pharmaceutical-research-practices |
# Information throughput of biological humans and frontier LLMs
Biological humans [appear](https://arxiv.org/html/2408.10234v2), across many domains, to have have an information throughput of at most about 50 bits per second. Naively multiplying this by the number of humans gives an upper bound of about 500 gigabits p... | https://www.lesswrong.com/posts/55zT4R3uWN3KosCes/information-throughput-of-biological-humans-and-frontier |
# Workshop: Interpretability in LLMs using Geometric and Statistical Methods
**Dates:** May 27-28
**Location:** Science Park 904, University of Amsterdam ([How to get here?](https://www.uva.nl/en/about-the-uva/organisation/faculties/faculty-of-science/contact-and-location/getting-here/route-description-science-park-... | https://www.lesswrong.com/posts/bTzk32t9aWJwLuNhi/workshop-interpretability-in-llms-using-geometric-and |
# Proselytizing
Religions can be divided into proselytizing religions (*e.g.* Mormons) who are supposed to recruit new members, and non-proselytizing religions (*e.g.* Orthodox Jews) who are the opposite. Zen Buddhism is a non-proselytizing religion, which makes me a bad Buddhist, because I've dragged three other peop... | https://www.lesswrong.com/posts/J9jj2EY6kuBRJ4CXE/proselytizing |
# Seeing Through the Eyes of the Algorithm
There’s a type of perspective shift that can bring a lot of clarity to the behavior and limitations of algorithms and AIs. This perspective may be called *seeing through the eyes of the algorithm* (or AI, or LLM). While some may consider it obvious and intuitive, I occasional... | https://www.lesswrong.com/posts/onDd6mJyadDzM9CZC/seeing-through-the-eyes-of-the-algorithm |
# HPMOR Anniversary Guide
[^q2qtecvcqwm]
Intro
-----
To everyone running an anniversary party, thank you. Someone had to overcome the bysta... | https://www.lesswrong.com/posts/LBs8RRQzHApvj5pvq/hpmor-anniversary-guide |
# Gradual Disempowerment: Simplified
This post is a summary of a paper [recently posted here](https://gradual-disempowerment.ai/) that describes, in my opinion, a very possible scenario that modern society will have to face in the near future.
This post is not really intended for the average LW reader, as most of... | https://www.lesswrong.com/posts/sGhkn7kPibYoYabBs/gradual-disempowerment-simplified |
# Transformer Dynamics: a neuro-inspired approach to MechInterp
How do AI models work? In many ways, we know the answer to this question, because we engineered those models in the first place. But in other, fundamental, ways, we have no idea. Systems with many parts that interact with each other nonlinearly are hard t... | https://www.lesswrong.com/posts/vucxxwdJARR3cqaPc/transformer-dynamics-a-neuro-inspired-approach-to-mechinterp |
# Power Lies Trembling: a three-book review
In [a previous book review](https://www.thinkingcomplete.com/2022/04/book-review-very-important-people.html) I described exclusive nightclubs as the particle colliders of sociology—places where you can reliably observe extreme forces collide. If so, military coups are the su... | https://www.lesswrong.com/posts/d4armqGcbPywR3Ptc/power-lies-trembling-a-three-book-review |
# New Report: Multi-Agent Risks from Advanced AI
*The development and widespread deployment of advanced AI agents will give rise to multi-agent systems of unprecedented complexity. A* [*new report*](https://arxiv.org/abs/2502.14143) *from staff at the Cooperative AI Foundation and a host of leading researchers explore... | https://www.lesswrong.com/posts/i3cwHXyHW8MzaCiaq/new-report-multi-agent-risks-from-advanced-ai |
# Deep sparse autoencoders yield interpretable features too
Summary
=======
* I sandwich the sparse layer in a sparse autoencoder (SAE) between non-sparse lower-dimensional layers and refer to this as a deep SAE.
* I find that features from deep SAEs are at least as interpretable as features from standard shallow... | https://www.lesswrong.com/posts/tLCBJn3NcSNzi5xng/deep-sparse-autoencoders-yield-interpretable-features-too |
# Test of the Bene Gesserit
Jessica didn't say what she felt, only what her son needed to hear.
"Paul…," Jessica said it with love, "You are going to die. But remember you are a duke's son. Do not dishonor Leto with your passing." Jessica whirled and strode from the room with a swish of her skirt. The door closed wit... | https://www.lesswrong.com/posts/vqnpx8L6TYqzHW2ad/test-of-the-bene-gesserit |
# List of most interesting ideas I encountered in my life, ranked
1. **Bayesian thinking**
It litterally was n°1 in my list, I was really happy to find this website, will not detail.
2. **Active ignorance/avoidance and selective attention/participation**
Instead of thinking, commenting, or saying somethi... | https://www.lesswrong.com/posts/pzDpAimGJNfQE9jHk/list-of-most-interesting-ideas-i-encountered-in-my-life |
# Reflections on the state of the race to superintelligence, February 2025
My model of the situation is that some time last year, the frontier paradigm moved from "scaling up large language models" to "scaling up chain-of-thought models". People are still inventing new architectures, e.g. Google's Titans, or Lecun's e... | https://www.lesswrong.com/posts/wuDXfLz2u8nfFXtCJ/reflections-on-the-state-of-the-race-to-superintelligence |
# The case for corporal punishment
***Preceded By:*** [*The case for the death penalty*](https://www.lesswrong.com/editPost?postId=yEBmFKsjwXs6dwwME&key=d7f9fb27368abe6f6c2b8dfd947ff1)
*Scott's essay* [*https://www.astralcodexten.com/p/prison-and-crime-much-more-than-you*](https://www.astralcodexten.com/p/prison-and-... | https://www.lesswrong.com/posts/rTveDBBavah4GHKxk/the-case-for-corporal-punishment |
# Have LLMs Generated Novel Insights?
In [a recent post](https://www.lesswrong.com/posts/vvgND6aLjuDR6QzDF/my-model-of-what-is-going-on-with-llms), Cole Wyeth makes a bold claim:
> . . . there is one crucial test (yes this is a crux) that LLMs have not passed. They have never done anything important.
>
> They haven... | https://www.lesswrong.com/posts/GADJFwHzNZKg2Ndti/have-llms-generated-novel-insights |
# Intelligence as Privilege Escalation
*Epistemic status: An interesting idea that is probably already in the air.*
> Inherent power you possess as part of yourself. Granted power is lent or given by other people.
-Patrick Rothfuss, The Wise Man's Fear
Humans are more powerful than other animals because we are smar... | https://www.lesswrong.com/posts/vLrj4ZNcGCTMJqdXB/intelligence-as-privilege-escalation |
# Judgements: Merging Prediction & Evidence
*I recently wrote about* [*complete feedback*](https://www.lesswrong.com/posts/3ag99iJEgFFwyj64Z/complete-feedback)*, an idea which I think is quite important for AI safety. However, my note was quite brief, explaining the idea only to my closest research-friends. This post ... | https://www.lesswrong.com/posts/3hs6MniiEssfL8rPz/judgements-merging-prediction-and-evidence |
# The Geometry of Linear Regression versus PCA
In statistics, there are two common ways to "find the best linear approximation to data": linear regression and principal component analysis. However, they are quite different---having distinct assumptions, use cases, and geometric properties. I remained subtly confused a... | https://www.lesswrong.com/posts/Q6T2pTLvDCnZPrFuv/the-geometry-of-linear-regression-versus-pca |
# The GDM AGI Safety+Alignment Team is Hiring for Applied Interpretability Research
**TL;DR**: **The Google DeepMind AGI Safety team is hiring for Applied Interpretability research scientists and engineers**. Applied Interpretability is a new subteam we are forming to focus on **directly using model internals-based te... | https://www.lesswrong.com/posts/aG9e5tHfHmBnDqrDy/the-gdm-agi-safety-alignment-team-is-hiring-for-applied |
# Export Surplusses
Trade surpluses are weird. I noticed this when I originally learned about them. Then I forgot this anomaly until…sigh…Eliezer Yudkowsky pointed it out.
 tie her wrists and ankles to the bed so you can have your way with here. She tells you to do whatever you want with her. And you think to yourself…
*Thi... | https://www.lesswrong.com/posts/dTaDsnwuYQnYc5xvc/nsfw-the-fuzzy-handcuffs-of-liberation |
# Grok Grok
This is a post in two parts.
The first half is the post is about Grok’s capabilities, now that we’ve all had more time to play around with it. Grok is not as smart as one might hope and has other issues, but it is better than I expected and for now has its place in the rotation, especially for when you wa... | https://www.lesswrong.com/posts/tpLfqJhxcijf5h23C/grok-grok |
# A City Within a City
My local gym is surrounded on all four sides by tent cities. At first I was nervous to walk there after work, but soon I got used to the tents and their residents.
When I’m passing through on my way to the gym, I feel like I’m traveling from Rome to the Vatican—a city within a city.
They have ... | https://www.lesswrong.com/posts/rLtf5mzsGhgEA75WM/a-city-within-a-city |
# Forecasting Frontier Language Model Agent Capabilities
*This work was done as part of the *[*MATS Program*](https://www.matsprogram.org/) *\- Summer 2024 Cohort.*
**Paper:** [link](https://static1.squarespace.com/static/6593e7097565990e65c886fd/t/67bc98d09087250062ed1891/1740413139348/agent_observational_scaling_la... | https://www.lesswrong.com/posts/bc5ohMwAyshdwJkDt/forecasting-frontier-language-model-agent-capabilities |
# Dream, Truth, & Good
One way in which I think current AI models are [sloppy](https://www.lesswrong.com/posts/PdtHzEb3cebnWCjoj/anti-slop-interventions) is that LLMs are trained in a way that messily merges the following "layers":
* The "dream machine" layer: LLMs are pre-trained on lots of slop from the internet,... | https://www.lesswrong.com/posts/DPjvL62kskHpp2SZg/dream-truth-and-good |
# What We Can Do to Prevent Extinction by AI
*Note: This is a draft of a page eventually intended to be shared broadly, updated frequently, and referenced often. For the moment, it reflects only my personal views, though it was inspired by conversations at work. Suggestions welcome. *
Introduction
============
On fi... | https://www.lesswrong.com/posts/orNiif9yEDzo7HZpS/what-we-can-do-to-prevent-extinction-by-ai |
# Superintelligent Agents Pose Catastrophic Risks:
Can Scientist AI Offer a Safer Path?
A [new paper](https://arxiv.org/abs/2502.15657) by Yoshua Bengio and the Safe Artificial Intelligence For Humanity (SAIFH) team argues that the current push towards building generalist AI agents presents catastrophic risks, creatin... | https://www.lesswrong.com/posts/p5gBcoQeBsvsMShvT/superintelligent-agents-pose-catastrophic-risks-can |
# Evaluating “What 2026 Looks Like” So Far
Summary
-------
In 2021, [@Daniel Kokotajlo](https://www.lesswrong.com/users/daniel-kokotajlo?mention=user) wrote [What 2026 Looks Like](https://www.lesswrong.com/posts/6Xgy6CAf2jqHhynHL/what-2026-looks-like), in which he sketched a possible version of each year from 2022 - ... | https://www.lesswrong.com/posts/u9Kr97di29CkMvjaj/evaluating-what-2026-looks-like-so-far |
# Conference Report: Threshold 2030 - Modeling AI Economic Futures
*This is an 8-page comprehensive summary of the results from Threshold 2030: a recent expert conference on economic impacts hosted by Convergence Analysis, Metaculus, and the Future of Life Institute. Please see the linkpost for the full end-to-end rep... | https://www.lesswrong.com/posts/wyEuDQksQBiHuz7M5/conference-report-threshold-2030-modeling-ai-economic |
# Training AI to do alignment research we don’t already know how to do
*This post heavily overlaps with “*[*how might we safely pass the buck to AI?*](https://www.lesswrong.com/posts/TTFsKxQThrqgWeXYJ/how-might-we-safely-pass-the-buck-to-ai)*” but is written to address a central counter argument raised in the comments... | https://www.lesswrong.com/posts/5gmALpCetyjkSPEDr/training-ai-to-do-alignment-research-we-don-t-already-know |
# Anthropic releases Claude 3.7 Sonnet with extended thinking mode
See also: [the research post detailing Claude's extended reasoning abilities](https://www.anthropic.com/news/visible-extended-thinking) and [the Claude 3.7 System Card](https://assets.anthropic.com/m/785e231869ea8b3b/original/claude-3-7-sonnet-system-c... | https://www.lesswrong.com/posts/qkfRNcvWz3GqoPaJk/anthropic-releases-claude-3-7-sonnet-with-extended-thinking |
# Local Trust
In the [last post](https://www.lesswrong.com/posts/bmS3tXBfSTYrueH5Q/deference-and-decision-making), we developed a new way of thinking about deference: you defer to someone when you'd prefer to let them make decisions on your behalf. This helped solve some puzzles about *modest experts*—experts who expr... | https://www.lesswrong.com/posts/xdTkcvqgBmKSEC5Ht/local-trust |
# Topological Data Analysis and Mechanistic Interpretability
This article was written in response to [a post on LessWrong from the Apollo Research interpretability team](https://www.lesswrong.com/posts/KfkpgXdgRheSRWDy8/a-list-of-45-%20mech-interp-project-ideas-from-apollo-research#:~:text=topological)). This post rep... | https://www.lesswrong.com/posts/6oF6pRr2FgjTmiHus/topological-data-analysis-and-mechanistic-interpretability |
# Metacompilation
A post that is going to be part of my sequence on rethinking programming languages.
The baisc idea
--------------
A compiler is a piece of machine code $C$, that takes as input a text string describing a program $p$ and returns the compiled machine code $C(p)$
Let $Opt$ be a function that takes in... | https://www.lesswrong.com/posts/6BSZkkWNGMTdRi5Ly/metacompilation |
# what an efficient market feels from inside
> “I often think of the time I met Scott Sumner and he said he pretty much assumes the market is efficient and just buys the most expensive brand of everything in the grocery store.” - [a Tweet](https://x.com/freezing_cloud/status/1199471383232704512)
It’s a funny quip, bu... | https://www.lesswrong.com/posts/mHEzJdyJSjxKFhjwD/what-an-efficient-market-feels-from-inside |
# Upcoming Protest for AI Safety
The [PauseAI](https://pauseai.info/) movement is planning protests for Friday, February 28th in several American metropolitan areas.
I understand [some people's reservations](https://www.lesswrong.com/posts/6dgCf92YAMFLM655S/the-sorry-state-of-ai-x-risk-advocacy-and-thoughts-on-doing)... | https://www.lesswrong.com/posts/MYX9XcsyRxjHtav6G/upcoming-protest-for-ai-safety |
# Revisiting Conway's Law
### The Feedback Loop is the Primary Product
The purpose of a company is to learn about its environment.[^d20n89b8cn] [^pmn6a0s19we] Companies are learning machines, and the best ones focus on extracting more information with less noise as fast as possible. This dynamic has always been true,... | https://www.lesswrong.com/posts/WEpGE5GugjWscfFJP/revisiting-conway-s-law |
# Economics Roundup #5
While we wait for the verdict on Anthropic’s Claude Sonnet 3.7, today seems like a good day to catch up on the queue and look at various economics-related things.
#### Table of Contents
1. [The Trump Tax Proposals.](https://thezvi.substack.com/i/155278650/the-trump-tax-proposals)
2. [Taxing ... | https://www.lesswrong.com/posts/AAKXjRmBRbJJwGthT/economics-roundup-5 |
# Intellectual lifehacks repo
1. *I really like* [*dimensional analysis*](https://en.wikipedia.org/wiki/Dimensional_analysis)*. It's a simple and powerful trick, almost magical, that allows you to distinguish between plausible and chimerical formulas.*
2. *I really like the* [*type signature*](https://en.wikipedia.o... | https://www.lesswrong.com/posts/pdh3246yv2DthvAHT/intellectual-lifehacks-repo |
# Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs
This is the abstract and introduction of [our new paper](https://arxiv.org/abs/2502.17424). We show that finetuning state-of-the-art LLMs on a narrow task, such as writing vulnerable code, can lead to misaligned behavior in various differen... | https://www.lesswrong.com/posts/ifechgnJRtJdduFGC/emergent-misalignment-narrow-finetuning-can-produce-broadly |
# Three Levels for Large Language Model Cognition
*This is the abridged version of my second dissertation chapter. Read the first* [*here*](https://www.lesswrong.com/posts/zjGh93nzTTMkHL2uY/the-intentional-stance-llms-edition)*. *
*Thanks to everyone I've discussed this with, and especially, M.A. Khalidi, Lewis Smith... | https://www.lesswrong.com/posts/nH28bPxcxHoZBECz5/three-levels-for-large-language-model-cognition |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.