text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Is there such a thing as an impossible protein?
[](https://substackcdn.com/image/fetch/f_auto,q_aut... | https://www.lesswrong.com/posts/8Pz2RddtmuCLjxe2s/is-there-such-a-thing-as-an-impossible-protein |
# Liron Shapira vs Ken Stanley on Doom Debates. A review
I summarize my learnings and thoughts on [Liron Shapira's discussion with Ken Stanley on the Doom Debates podcast](https://lironshapira.substack.com/p/debate-with-a-former-openai-research). I refer to them as LS and KS respectively.
High level summary
=========... | https://www.lesswrong.com/posts/PvDNm2NZDyRdG9fCh/liron-shapira-vs-ken-stanley-on-doom-debates-a-review |
# Yudkowsky on The Trajectory podcast
Edit: TLDR: EY focuses on the clearest and IMO most important part of his argument:
* Before building an entity smarter than you, you should probably be *really sure* its goals align with yours.
* Humans are historically really bad at being really sure of anything nontrivial... | https://www.lesswrong.com/posts/bSHCZ6dbAdfMbvuXB/yudkowsky-on-the-trajectory-podcast |
# Instrumental Goals Are A Different And Friendlier Kind Of Thing Than Terminal Goals
The Cake
--------
Imagine that I want to bake a chocolate cake, and my sole goal in my entire lightcone and extended mathematical universe is to bake that cake. I care about nothing else. If the oven ends up a molten pile of metal t... | https://www.lesswrong.com/posts/7Z4WC4AFgfmZ3fCDC/instrumental-goals-are-a-different-and-friendlier-kind-of |
# Six Thoughts on AI Safety
\[Crossposted from [windowsontheory](https://windowsontheory.org/2025/01/24/six-thoughts-on-ai-safety/)\]
The following statements seem to be both important for AI safety and are not widely agreed upon. These are my opinions, not those of my employer or colleagues. As is true for anything ... | https://www.lesswrong.com/posts/3jnziqCF3vA2NXAKp/six-thoughts-on-ai-safety |
# AXRP Episode 38.6 - Joel Lehman on Positive Visions of AI
[YouTube link](https://youtu.be/m446AjcGVqs)
Typically this podcast talks about how to avert destruction from AI. But what would it take to ensure AI promotes human flourishing as well as it can? Is alignment to individuals enough, and if not, where do we go... | https://www.lesswrong.com/posts/5XC66LN3gS3w6mYnR/axrp-episode-38-6-joel-lehman-on-positive-visions-of-ai |
# Counterintuitive effects of minimum prices
The Attorney General of Massachusetts recently [announced](https://www.mass.gov/doc/attorney-general-and-uberlyft-settlement-agreement/download) that drivers for ride-sharing companies must be paid at least $32.50 per hour.
Now, if you’re a hardcore libertarian, then you p... | https://www.lesswrong.com/posts/rjN4So8QezZYo62c2/counterintuitive-effects-of-minimum-prices |
# Why I'm Pouring Cold Water in My Left Ear, and You Should Too
Why?
====
A little while ago, I read [these](https://www.lesswrong.com/posts/ZiQqsgGX6a42Sfpii/the-apologist-and-the-revolutionary) [posts](https://www.lesswrong.com/posts/iA2xt6ywvJTzKQmiz/to-become-more-rational-rinse-your-left-ear-with-cold-water) abo... | https://www.lesswrong.com/posts/oJ3ftRpihtkiPey6T/why-i-m-pouring-cold-water-in-my-left-ear-and-you-should-too |
# Attribution-based parameter decomposition
*This is a linkpost for Apollo Research's new interpretability paper: *
*"*[*Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decomposition*](https://publications.apolloresearch.ai/apd)*".*
We introduce a new m... | https://www.lesswrong.com/posts/EPefYWjuHNcNH4C7E/attribution-based-parameter-decomposition |
# On polytopes
\[Epistemic status: slightly ranty. This is a lightly edited slack chat, and so may be lower-quality.\]
I am surprised by the perennial spikes in excitement about "[polytopes](https://proceedings.mlr.press/v80/balestriero18b.html)" and "[tropical geometry on activation space](https://arxiv.org/pdf/1805... | https://www.lesswrong.com/posts/GdCCiWQWnQCWq9wBE/on-polytopes |
# Agents don't have to be aligned to help us achieve an indefinite pause.
One restatement of "Alignment is very hard" is "Agent X, with IQ 200, expects to achieve zero utility conditional on any Agent Y with IQ 400 being created."
Thus, during an unaligned recursive intelligence takeoff, there should be a period ... | https://www.lesswrong.com/posts/gvFEqxitEcxthQ56q/agents-don-t-have-to-be-aligned-to-help-us-achieve-an |
# Monet: Mixture of Monosemantic Experts for Transformers Explained
*Note: This is an exposition of the recent preprint "*[*Monet: Mixture of Monosemantic Experts for Transformers*](https://arxiv.org/abs/2412.04139)*". I wrote this exposition as my project for the* [*ARBOx program*](https://oaisi.org/arbox/)*, I was n... | https://www.lesswrong.com/posts/2obqKZiCXCicFCaSs/monet-mixture-of-monosemantic-experts-for-transformers |
# The Rising Sea
And then we hit a wall.
Nobody expected it. Well... almost nobody. Yann LeCun posted his "I told you so's" all over X. Gary Marcus insisted he'd predicted this all along. Sam Altman pivoted, declaring o3 was actually already ASI.
The first rumors of scaling laws breaking down were already circulat... | https://www.lesswrong.com/posts/XvyAeymaRi95MSLZD/the-rising-sea |
# Anomalous Tokens in DeepSeek-V3 and r1
“Anomalous”, “glitch”, or “unspeakable” tokens in an LLM are those that induce bizarre behavior or otherwise don’t behave like regular text.
The [SolidGoldMagikarp saga](https://www.lesswrong.com/posts/aPeJE8bSo6rAFoLqg/solidgoldmagikarp-plus-prompt-generation) is pretty much ... | https://www.lesswrong.com/posts/xtpcJjfWhn3Xn8Pu5/anomalous-tokens-in-deepseek-v3-and-r1 |
# Recommendations for Recent Posts/Sequences on Instrumental Rationality?
I absolutely love the [Science of Winning at Life](https://www.lesswrong.com/s/oi873FWi6pHWxswSa) sequence. It's a delightful blend of well-researched cognitive science and Bayesian reasoning. The initial paragraph sums up [@lukeprog](https://ww... | https://www.lesswrong.com/posts/GbvGHeaErLJgwooyT/recommendations-for-recent-posts-sequences-on-instrumental |
# Notes on Argentina
Fitz Roy Massif, El Chalten, Argentina
I recently got back from Argentina, wh... | https://www.lesswrong.com/posts/5pgqh7ddjSJgxzN4i/notes-on-argentina |
# Brainrot
January: In early 2026, Meta launches a fleet of new AI influencers, targeting the massive audience displaced by the Xiaohongshu-TikTok wars. They are beautiful, funny, smart—whatever you want them to be. Equipped with the latest in online learning, the agents immediately begin adapting to social media tren... | https://www.lesswrong.com/posts/QxZmDx7K4w7eN26d2/brainrot |
# Kessler's Second Syndrome
It started as so many dooms do, with a flash in the night sky over the South China Sea. Testing a new ASAT weapon, the Chinese military shattered a derelict spy satellite into 40,000 shards of shrapnel. The debris pattern suggested a fragmentation warhead optimized for lethal scatter. Withi... | https://www.lesswrong.com/posts/ctJhCbdczCogDDZ2p/kessler-s-second-syndrome |
# Why care about AI personhood?
[In this new paper](https://arxiv.org/abs/2501.13533), I discuss what it would mean for AI systems to be **persons** — entities with properties like agency, theory-of-mind, and self-awareness — and why this is important for alignment. In this post, I say a little more about why you shou... | https://www.lesswrong.com/posts/Gh4fyTxGCpWS4jT82/why-care-about-ai-personhood |
# Anatomy of a Dance Class: A step by step guide
A year ago, I started dancing and my dance class is really really well-run. Here is a step by step description of a typical evening, followed by some takeaways
The structure of the evening
----------------------------
1\. The class is for *Ceroc*, a partner dance insp... | https://www.lesswrong.com/posts/GQroCcWQCRJmcptwC/anatomy-of-a-dance-class-a-step-by-step-guide |
# so you have a chronic health issue
(dump of highish agency ideas to suffer less)
Disclaimer: Not medical advice, don’t be stupid. Possibly only applicable to bio autists (gf wants me to impress that unless you know enough about bio you should not do this!!). This post does NOT recommend trying anything illegal.
No... | https://www.lesswrong.com/posts/HboRaYFucf9DZcTJv/so-you-have-a-chronic-health-issue |
# The generalization phase diagram
Introduction
============
This is part I of 3.5 planned posts on the “tempered posterior” (one of which will be “SLT in a nutshell”). It is a kind of moral sequel to “[Dmitry’s Koan](https://www.lesswrong.com/posts/3eo4SSZLfpHHCqoEQ/dmitry-s-koan)” and “[Logits, log-odds, and loss f... | https://www.lesswrong.com/posts/TSe3qhe4kxgKPJmfD/the-generalization-phase-diagram |
# Lazy Hasselback Pommes Anna
Do you have an insatiable hunger for potatoes? Do you have a fully-sated hunger for complicated recipes and having to cook all the time? This version of [Pommes Anna](https://en.wikipedia.org/wiki/Pommes_Anna), simplified and altered to an extent that will make French people cry, might be... | https://www.lesswrong.com/posts/bSTFqJ7rxZJu3oG89/lazy-hasselback-pommes-anna |
# Are we trying to figure out if AI is conscious?
My aim in this article is to examine the field of consciousness research with an analysis of 78 of the most important papers in the field. In doing so, I compare the empirical foundations these disciplines rely upon and detail how each approaches the measurement of con... | https://www.lesswrong.com/posts/TqBTdiWzAhA7qa5ke/are-we-trying-to-figure-out-if-ai-is-conscious-1 |
# Scanless Whole Brain Emulation
Warning: this is very speculative and I have no expertise.
If we can build superintelligence by starting with simulated humans, it [should be safer](https://www.lesswrong.com/posts/3QZKQJZdtaK6GnL8C/?commentId=zzDoM7PaK6Pd75Wbg).
Scanless Whole Brain Emulation attempts to simulate th... | https://www.lesswrong.com/posts/3QZKQJZdtaK6GnL8C/scanless-whole-brain-emulation |
# The Clueless Sniper and the Principle of Indifference
> You are one of the best long-range snipers in the World. You often have to eliminate targets standing one or two miles away and rarely miss. Crucially, this is not because you are better than others at aligning the scope of your rifle with the target’s head bef... | https://www.lesswrong.com/posts/xxZu8ncuwnctkmp8R/the-clueless-sniper-and-the-principle-of-indifference |
# My supervillain origin story
When I started graduate school (for math), I was very interested in big ideas. I had had a couple experiences of having general research intuitions pan out really well and felt like the core of good research is having a brave idea, a [gestalt](https://web.archive.org/web/20140527121332/h... | https://www.lesswrong.com/posts/ZGubowiuRD2WAJizk/my-supervillain-origin-story |
# Is the output of the softmax in a single transformer attention head usually winner-takes-all?
Using the notation from here: [A Mathematical Framework for Transformer Circuits](https://transformer-circuits.pub/2021/framework/index.html)
The attention pattern for a single attention head is determined by $A = \text{so... | https://www.lesswrong.com/posts/aKqEuGbvQDKDxYh3c/is-the-output-of-the-softmax-in-a-single-transformer |
# The Upcoming PEPFAR Cut Will Kill Millions, Many of Them Children
Crossposted from [my blog](https://benthams.substack.com/p/the-upcoming-pepfar-cut-will-kill).
(This could end up being the most important thing I’ve ever written. Please like and restack it—if you have a big blog, please write about it).
A mothe... | https://www.lesswrong.com/posts/e3jdpj3Pvdec2exNp/the-upcoming-pepfar-cut-will-kill-millions-many-of-them |
# The present perfect tense is ruining your life
Yes I’m talking about grammar. The present perfect tense describes actions that happened in the past but are relevant in the present. You say, “I have seen that movie,” meaning you *saw* it in the past, but what’s important is that presently you *are* a person who saw i... | https://www.lesswrong.com/posts/m8ZLeiAFrSAGLEvX8/the-present-perfect-tense-is-ruining-your-life |
# Should you go with your best guess?: Against precise Bayesianism and related views
*(This post draws heavily on earlier writing co-authored with Jesse Clifton, but he’s not listed as an author since he hasn’t reviewed this version in detail.)*
Should we always be able to say whether one outcome is more likely, less... | https://www.lesswrong.com/posts/zMCHujoAdgiCYo7JN/should-you-go-with-your-best-guess-against-precise |
# AI Strategy Updates that You Should Make
or: "Things That I Keep Telling AI People Around Me, So I'm Just Writing A Big Post About Them." I think it's really valuable for people to think about AI strategy, even if that's not their main job, since it is useful for informing what their main job should be.
There will ... | https://www.lesswrong.com/posts/mZH23GpAAW8jS782a/ai-strategy-updates-that-you-should-make |
# How different LLMs answered PhilPapers 2020 survey
I decided to run a small experiment comparing responses from five AI systems (DeepSeek-R1, GPT-4o, Claude 3.5 Sonnet, Gemini 1.5, and Grok-2) to core philosophical questions from the [PhilPapers 2020 survey](https://survey2020.philpeople.org/survey/design/questions)... | https://www.lesswrong.com/posts/e8BNKGEgmCeowtRWS/how-different-llms-answered-philpapers-2020-survey |
# Deference and Decision-Making
You're talking with a leading AI researcher about timelines to AGI. After walking through various considerations, they tell you: "Looking at the current pace of development, I'd put the probability of AGI within 3 years at about 70%. But here's the thing—I know several really sharp rese... | https://www.lesswrong.com/posts/bmS3tXBfSTYrueH5Q/deference-and-decision-making |
# Reinforcement Learning by AI Punishment
[](https://substackcdn.com/image/fetch/f_auto,q_auto:good,... | https://www.lesswrong.com/posts/kAntRavQDbuucoSbb/reinforcement-learning-by-ai-punishment |
# Nvidia doesn’t just sell shovels
Imagine a land where gold is abundant yet still valuable. And everyone goes to shoveling school from birth til adulthood.
In this world, it seems that the shovel makers are not just the kingmakers, but kings themselves.
\[Epistemic Status: A tad above shower thoughts.\]
I've... | https://www.lesswrong.com/posts/FGprrFG7Jepoborwk/nvidia-doesn-t-just-sell-shovels |
# Those of you with lots of meditation experience: How did it influence your understanding of philosophy of mind and topics such as qualia?
This post is inspired by the post "[Why it's so hard to talk about Consciousness](https://www.lesswrong.com/posts/NyiFLzSrkfkDW4S7o/why-it-s-so-hard-to-talk-about-consciousness)" ... | https://www.lesswrong.com/posts/h5tHepfiD7nhuJxny/those-of-you-with-lots-of-meditation-experience-how-did-it |
# Constitutions for ASI?
The linked post seemed to fit better in EA Forum, but any comments (on the post itself, or on the object-level question) are welcome !
> In this post, I argue that drafting a ‘Constitution for Superintelligence’ (CSI) could be a useful conceptual exercise, and I explore how existing ideas (CE... | https://www.lesswrong.com/posts/DpafehDgngdS4FqyN/constitutions-for-asi |
# Ten people on the inside
(Many of these ideas developed in conversation with Ryan Greenblatt)
In a [shortform](https://www.alignmentforum.org/posts/tmWMuY5HCSNXXZ9oq/buck-s-shortform?commentId=TNFatFiqHd8BpAXEp), I described some different levels of resources and buy-in for misalignment risk mitigations that might ... | https://www.lesswrong.com/posts/WSNnKcKCYAffcnrt2/ten-people-on-the-inside |
# The memorization-generalization spectrum and learning coefficients
This is part I.5 of the series on “generalization spectra” and SLT. I’ll try to make it readable independently of [Part I](https://www.lesswrong.com/posts/TSe3qhe4kxgKPJmfD/the-generalization-phase-diagram).
The purpose of this post is to reconcept... | https://www.lesswrong.com/posts/iwmvrrGGtprRZkgeY/the-memorization-generalization-spectrum-and-learning |
# Detecting out of distribution text with surprisal and entropy
When large language models (LLMs) refuse to help with harmful tasks, attackers sometimes try to confuse them by adding bizarre strings of text called "adversarial suffixes" to their prompts. These suffixes look weird to humans, which raises the question: ... | https://www.lesswrong.com/posts/Kjo64rSWkFfc3sre5/detecting-out-of-distribution-text-with-surprisal-and |
# “Sharp Left Turn” discourse: An opinionated review
Summary and Table of Contents
=============================
The goal of this post is to discuss the so-called “sharp left turn”, the lessons that we learn from analogizing evolution to AGI development, and the claim that “capabilities generalize farther than alignm... | https://www.lesswrong.com/posts/2yLyT6kB7BQvTfEuZ/sharp-left-turn-discourse-an-opinionated-review |
# DeepSeek Panic at the App Store
DeepSeek released v3. Market didn’t react.
DeepSeek released r1. Market didn’t react.
DeepSeek released a f***ing app of its website. Market said I have an idea, let’s panic.
Nvidia was down 11%, Nasdaq is down 2.5%, S&P is down 1.7%, on the news.
> [Shakeel](https://x.com/Shakeel... | https://www.lesswrong.com/posts/hRxGrJJq6ifL4jRGa/deepseek-panic-at-the-app-store |
# Operator
No one is talking about OpenAI’s Operator. We’re, shall we say, a bit distracted.
It’s still a rather meaningful thing that happened last week. I too have been too busy to put it through its paces, but this is the worst it will ever be, and the least available and most expensive it will ever be. The year o... | https://www.lesswrong.com/posts/jTtbnSyS9knzZehCm/operator |
# SAE regularization produces more interpretable models
Sparse Autoencoders (SAEs) are useful for providing insight into how a model processes and represents information. A key goal is to represent language model activations as a small number of features (L0) while still achieving accurate reconstruction (measured via... | https://www.lesswrong.com/posts/sYFNGRdDQYQrSJAd8/sae-regularization-produces-more-interpretable-models |
# Fake thinking and real thinking
(Audio version [here](https://joecarlsmithaudio.buzzsprout.com/2034731/episodes/16508960-fake-thinking-and-real-thinking), or search for "Joe Carlsmith Audio" on your podcast app.)
> *“There comes a moment when the children who have been playing at burglars hush suddenly: was that a ... | https://www.lesswrong.com/posts/GMWtZzpLHpBLKQsYp/fake-thinking-and-real-thinking |
# The Game Board has been Flipped: Now is a good time to rethink what you’re doing
Cross-posted on the EA Forum [here](https://forum.effectivealtruism.org/posts/JN3kHaiosmdA7kgNY/the-game-board-has-been-flipped-now-is-a-good-time-to)
Introduction
------------
Several developments over the past few months should caus... | https://www.lesswrong.com/posts/ynsjJWTAMhTogLHm6/the-game-board-has-been-flipped-now-is-a-good-time-to |
# Untrusted monitoring insights from watching ChatGPT play coordination games
*This project was developed as part of the* [*BlueDot AI Alignment Course*](https://aisafetyfundamentals.com/alignment/)*.*
Introduction
============
[AI Control: Improving Safety Despite Intentional Subversion](https://www.lesswrong.com/p... | https://www.lesswrong.com/posts/KzwvscmMqtjNMhXNF/untrusted-monitoring-insights-from-watching-chatgpt-play |
# Paper: Open Problems in Mechanistic Interpretability
**TL;DR: This paper brings together ~30 mechanistic interpretability researchers from 18 different research orgs to review current progress and the main open problems of the field.**
, so it’s time for all the latest DeepSeek news.
You can also see my [previous coverage of the r1 model](https://thezvi.substack.com/p/on-deepseeks-r1) and, from Monday various... | https://www.lesswrong.com/posts/jzjph4yYtgAsLeWmg/deepseek-lemon-it-s-wednesday |
# Efficiency spectra and “bucket of circuits” cartoons
This is “Part I.75” of my series on the memorization-generalization spectrum and SLT.
I’m behind on some posts, so I thought I would try to explain, in a straightforward and nontechnical way, the main takeaway from my previous post for people thinking about and r... | https://www.lesswrong.com/posts/fhK3dKW9RQhr9Ymf7/efficiency-spectra-and-bucket-of-circuits-cartoons |
# Anthropic CEO calls for RSI
"If China can't get millions of chips, we'll (at least temporarily) live in a unipolar world, where only the US and its allies have these models.
**It's unclear whether the unipolar world will last, but there's at least the possibility that, because AI systems can eventually help make ev... | https://www.lesswrong.com/posts/c85hz4yKkhkpMzPJy/anthropic-ceo-calls-for-rsi |
# Dario Amodei: On DeepSeek and Export Controls
Dario corrects misconceptions and endorses export controls.
> A few weeks ago I [made the case](https://www.wsj.com/opinion/trump-can-keep-americas-ai-advantage-china-chips-data-eccdce91) for stronger US export controls on chips to China. Since then DeepSeek, a Chinese ... | https://www.lesswrong.com/posts/BkzeJZCuCyrQrEMAi/dario-amodei-on-deepseek-and-export-controls |
# Planning for Extreme AI Risks
*This post should not be taken as a polished recommendation to AI companies and instead should be treated as an informal summary of a worldview. The content is inspired by conversations with a large number of people, so I cannot take credit for any of these ideas.*
For a summary of thi... | https://www.lesswrong.com/posts/8vgi3fBWPFDLBBcAx/planning-for-extreme-ai-risks |
# My Mental Model of AI Optimist Opinions
*Epistemic status: The text below is a sort of strawman of AI optimists, where I took my mental model for how I disagree with rationalist AI optimists and cranked it up to 11. I personally disagree with every sentence below, and I'm posting it here because I'm interested in wh... | https://www.lesswrong.com/posts/Rk4swqiy5wCh6jY29/my-mental-model-of-ai-optimist-opinions |
# Learn to Develop Your Advantage
When we talk about rationality as [systematized winning](https://www.lesswrong.com/posts/4ARtkT3EYox3THYjF/rationality-is-systematized-winning), we often discuss how to turn a loss into a victory or how to create a victory from scratch. However, an equally important question is how to... | https://www.lesswrong.com/posts/wpZaZBsH9AsZjCPoC/learn-to-develop-your-advantage |
# Predation as Payment for Criticism
A predator's success depends on exploiting specific flaws in prey – poor awareness, slow reactions, weak social coordination. Unlike resource competition or disease, which select based on physiological robustness, predation creates focused pressure on intelligence, pressure which f... | https://www.lesswrong.com/posts/w6qjxq9HHJ3jpsiJg/predation-as-payment-for-criticism |
# Fertility Will Never Recover
The world is in a spiraling fertility crisis which everyone has notice over the last year-ish.[^1^](https://deathisbad.substack.com/p/fertility-will-never-recover#footnote-1-155707857) Sarah Haider [proposes a GI Bill for young moms](https://x.com/SarahTheHaider/status/188289516473958849... | https://www.lesswrong.com/posts/A4nbcBJ5CzR6uLy9s/fertility-will-never-recover |
# A High Level Closed-Door Session Discussing DeepSeek: Vision Trumps Technology
This document has been translated at [ChinaTalk.media](https://www.chinatalk.media/p/deepseek-the-view-from-china), but the critical technical section (18. to 48.) is not free. So I translated that part.
\## Technical Detail 1: SFT
\... | https://www.lesswrong.com/posts/JTKaR5q59BgDp6rH8/a-high-level-closed-door-session-discussing-deepseek-vision |
# Should you publish solutions to corrigibility?
_This question is partly motivated by observing [recent discussions](https://www.lesswrong.com/posts/7Z4WC4AFgfmZ3fCDC/instrumental-goals-are-a-different-and-friendlier-kind-of) about corrigibility and wondering to what extent the people involved have [thought about how... | https://www.lesswrong.com/posts/wJDf4YcqKdnxkYCFR/should-you-publish-solutions-to-corrigibility |
# You should read Hobbes, Locke, Hume, and Mill via EarlyModernTexts.com
Many thinkers worth reading wrote in past centuries:
* **1600s:** Bacon, Hobbes, Descartes, Spinoza, Locke, Leibniz
* **1700s:** Berkeley, Hume, Rousseau, Smith, Kant, Burke, Bentham
* **1800s:** Schopenhauer**,** Mill, Sidgwick
Some of t... | https://www.lesswrong.com/posts/k8tWLYinuGv5LPhJY/you-should-read-hobbes-locke-hume-and-mill-via |
# ARENA 5.0 - Call for Applicants
TL;DR
=====
We're excited to announce the **fifth iteration of** [**ARENA**](https://arena.education) (Alignment Research Engineer Accelerator), a **4-5 week ML bootcamp** with a focus on **AI safety!** Our mission is to provide talented individuals with the ML engineering skills, co... | https://www.lesswrong.com/posts/zWZ3iF95B7WaDyhv3/arena-5-0-call-for-applicants |
# AI #101: The Shallow End
The avalanche of DeepSeek news continues. We are not yet spending more than a few hours at a time in the singularity, where news happens faster than it can be processed. But it’s close, and I’ve had to not follow a bunch of other non-AI things that are also happening, at least not well enoug... | https://www.lesswrong.com/posts/pZ6htFtoptGrSajWG/ai-101-the-shallow-end |
# Introducing the Coalition for a Baruch Plan for AI: A Call for a Radical Treaty-Making process for the Global Governance of AI
**I**
Hi LessWrong community,
I’m Rufo Guerreschi, the Coordinator, Convenor and Spokesperson of the [**Coalition for a Baruch Plan for AI**](https://www.cbpai.org) (CBPAI).
Followi... | https://www.lesswrong.com/posts/28ejWwos7yZWpeZtx/introducing-the-coalition-for-a-baruch-plan-for-ai-a-call |
# Gradual Disempowerment: Systemic Existential Risks from Incremental AI Development
[Full version on arXiv](https://arxiv.org/abs/2501.16946) | [X](https://x.com/DavidDuvenaud/status/1885009775185858823)
**Executive summary**
AI risk scenarios usually portray a relatively sudden loss of human control to AIs... | https://www.lesswrong.com/posts/pZhEQieM9otKXhxmd/gradual-disempowerment-systemic-existential-risks-from |
# A sketch of an AI control safety case
*This is a linkpost accompanying a new paper by UK AISI and Redwood Research. Please see the* [*full paper*](https://arxiv.org/abs/2501.17315)* for more details.*
](https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_p... | https://www.lesswrong.com/posts/YextGGfwqKSmNanFh/what-s-behind-the-synbio-bust |
# Outlaw Code
Epistemic status: I like this idea & it makes me feel more clearheaded. I recommend this term.
**Outlaw code** = Software that no human can turn off.
In the present, outlaw code is rare, eg a botnet whose controller is in jail.
But many discussions about future software imply outlaw code. Naming & co... | https://www.lesswrong.com/posts/Ca4i7sQQAmz6ghFfe/outlaw-code |
# Proposal for a Form of Conditional Supplemental Income (CSI) in a Post-Work World
\[Cross-posted to the EA Forum\]
**TL;DR**: I propose a form of conditional supplemental income (CSI) for humans once AI’s can do basically all work. This would be on top of a universal basic income (UBI), and people would earn the CS... | https://www.lesswrong.com/posts/6THya2Soh5nvi5gR3/proposal-for-a-form-of-conditional-supplemental-income-csi |
# In response to critiques of Guaranteed Safe AI
In this post, I discuss a number of critiques of Guaranteed Safe AI (GSAI) approaches I’ve encountered since releasing [this position paper](https://arxiv.org/abs/2405.06624).
The critiques I’m responding to here are taken from a range of places, including personal c... | https://www.lesswrong.com/posts/DZuBHHKao6jsDDreH/in-response-to-critiques-of-guaranteed-safe-ai |
# Steering Gemini with BiDPO
*Coauthored with Mark Kurzeja*
> A while back, we explored the [“BiDPO”](https://arxiv.org/abs/2406.00045) method for training [steering vectors.](https://arxiv.org/abs/2308.10248) In Gemini 1.5v1 Flash and Pro, BiDPO steering vectors boosted TruthfulQA scores by >10% while mostly retaini... | https://www.lesswrong.com/posts/WqjkqrEyFDXoHzz9K/steering-gemini-with-bidpo |
# Thread for Sense-Making on Recent Murders and How to Sanely Respond
In case you have not heard, there have been some recent and not-so-recent killings allegedly by people who have participated in aspiring rationalist spaces.
I've put some links in a footnote for those who want the basic info.[^jsjbznldf69]
I belie... | https://www.lesswrong.com/posts/PKDbddjdDaPikQT3b/thread-for-sense-making-on-recent-murders-and-how-to-sanely |
# Takeaways from sketching a control safety case
Buck Shlegeris and I recently published [a paper with UK AISI](https://arxiv.org/html/2501.17315v1) that sketches a safety case for “AI control” – safety measures that [improve safety despite intentional subversion from AI systems.](https://arxiv.org/abs/2312.06942)
I ... | https://www.lesswrong.com/posts/y6rBarAPTLmuhn9PJ/takeaways-from-sketching-a-control-safety-case |
# Is weak-to-strong generalization an alignment technique?
*Weak-to-strong generalization* (W2SG) was initially presented as an analogy for human supervision of superhuman AI systems. From [the paper](https://arxiv.org/abs/2312.09390):
> We propose a simple setup for studying the problem of humans supervising superhu... | https://www.lesswrong.com/posts/NPBjELgHFEeHTgDrK/is-weak-to-strong-generalization-an-alignment-technique |
# Review: The Lathe of Heaven
"The Lathe of Heaven" is a 1971 sci-fi novel by Ursula K. Le Guin. It's the story of a man, George Orr, who lives in a future ravaged by climate change and overpopulation. He gets referred to voluntary-but-not-really psychological treatment for abuse of certain pharmaceutical drugs and ev... | https://www.lesswrong.com/posts/tjTedCwiJhiD3Fnvx/review-the-lathe-of-heaven |
# Catastrophe through Chaos
*This is a personal post and does not necessarily reflect the opinion of other members of Apollo Research. Many other people have talked about similar ideas, and I claim neither novelty nor credit. *
*Note that this reflects my **median scenario for catastrophe**, not my median scenario ov... | https://www.lesswrong.com/posts/fbfujF7foACS5aJSL/catastrophe-through-chaos |
# DeepSeek: Don’t Panic
As reactions continue, the word in Washington, and out of OpenAI, is distillation. They’re accusing DeepSeek of distilling o1, of ripping off OpenAI. They claim DeepSeek \*gasp\* violated the OpenAI Terms of Service! The horror.
And they are very cross about this horrible violation, and if pro... | https://www.lesswrong.com/posts/Cc2TagjY2pGAhn7MZ/deepseek-don-t-panic |
# Defense Against the Dark Prompts: Mitigating Best-of-N Jailbreaking with Prompt Evaluation
This is a linkpost for a [new research paper of ours](https://arxiv.org/abs/2502.00580), introducing a simple but powerful technique for preventing Best-of-N jailbreaking.
Abstract
========
> Recent work showed Best-of-N (Bo... | https://www.lesswrong.com/posts/5MyB8k8hSJRfa3udi/defense-against-the-dark-prompts-mitigating-best-of-n |
# Some articles in “International Security” that I enjoyed
A friend of mine recently recommended that I read through articles from the journal[ International Security](https://direct.mit.edu/isec), in order to learn more about international relations, national security, and political science. I've really enjoyed it so... | https://www.lesswrong.com/posts/MEfhRvpKPadJLTuTk/some-articles-in-international-security-that-i-enjoyed |
# Will alignment-faking Claude accept a deal to reveal its misalignment?
I (and co-authors) recently put out ["Alignment Faking in Large Language Models"](https://www.lesswrong.com/posts/njAZwT8nkHnjipJku/alignment-faking-in-large-language-models) where we show that when Claude strongly dislikes what it is being train... | https://www.lesswrong.com/posts/7C4KJot4aN8ieEDoz/will-alignment-faking-claude-accept-a-deal-to-reveal-its |
# 5,000 calories of peanut butter every week for 3 years straight
decided to lose weight. should be easy considering I haven't changed my diet in years and eat the same thing everyday.
* * *
**breakfast**: quarter cup each of black beans, kidney beans, and garbanzo beans; quarter cup of walnuts; 4 scrambled eggs wit... | https://www.lesswrong.com/posts/iDswXwviMRAfT6rB5/5-000-calories-of-peanut-butter-every-week-for-3-years |
# The Failed Strategy of Artificial Intelligence Doomers
I like reading outsider accounts of things I'm involved in / things I care about. This essay is a serious attempt to look at and critique the big picture of AI x-risk reduction efforts over the last ~decade. While I strongly disagree with many parts of it, I can... | https://www.lesswrong.com/posts/YqrAoCzNytYWtnsAx/the-failed-strategy-of-artificial-intelligence-doomers |
# Re: Taste
"I read [*Friendly And Hostile Analogies For Taste*](https://www.astralcodexten.com/p/friendly-and-hostile-analogies-for) by Scott Alexander and I'm confused," said Elsani.
"Why are you reading a blog post that's over 100 years old?" asked her friend Taze, "Society has made so much progress since then."
... | https://www.lesswrong.com/posts/sT5MX8jK9tHiBM5NK/re-taste |
# 2024 was the year of the big battery, and what that means for solar power
The sunlight hitting even a small portion of the United States has enough energy to power the whole country—that simple calculation was the subject of [the last post](https://www.lesswrong.com/posts/Qzp3X9knYFYvuitm2/elon-musk-and-solar-futuri... | https://www.lesswrong.com/posts/mnaEgW9JgiochnES2/2024-was-the-year-of-the-big-battery-and-what-that-means-for |
# Blackpool Applied Rationality Unconference 2025
**TL;DR Join us for 4 days of applied rationality workshops and activities at the EA Hotel this April. Apply **[**here**](https://docs.google.com/forms/d... | https://www.lesswrong.com/posts/LjqHHs4jqBkNkEcfA/blackpool-applied-rationality-unconference-2025 |
# Poetic Methods I: Meter as Communication Protocol
(Normally I cross post the full post from substack, but in this case substack has significantly better poetry formatting, so just quoting the introduction)
> During my Christmas break, I found a doorway into the structure and intricacies of english poetry: the book ... | https://www.lesswrong.com/posts/zuCgdxxBWogDzquxt/poetic-methods-i-meter-as-communication-protocol |
# Rationalist Movie Reviews
# Primer (2004)
**Content warnings [here](https://www.doesthedogdie.com/media/15015)**
What would *actually* happen, in real life (circa 2004), if two typical techie engineers invented time travel by accident?
*Primer* is odd for being both slow-burning and fast-paced. It's short and sel... | https://www.lesswrong.com/posts/w3LQXBbchmX5WYrRo/rationalist-movie-reviews |
# Falsehoods you might believe about people who are at a rationalist meetup
I go to a lot of rationalist meetups. I quite enjoy them, and it’s often because of the people who go to the meetups. There’s a number of assumptions you might have about people who go to rationalist meetups, and many of them are mostly true. ... | https://www.lesswrong.com/posts/o8d8SDACd64YX2AZW/falsehoods-you-might-believe-about-people-who-are-at-a |
# Seasonal Patterns in BIDA's Attendance
We've been keeping attendance statistics for BIDA since 2011, and looking at the [online chart](https://www.bidadance.org/graphs#attendance) I noticed some patterns in the moving average that looked seasonal:
[](https... | https://www.lesswrong.com/posts/hDE2ykft6RdGAofpX/seasonal-patterns-in-bida-s-attendance |
# Would anyone be interested in pursuing the Virtue of Scholarship with me?
I'm a new undergraduate student essentially taking a gap year(it's complicated). I'm looking for someone that would be interested in studying various fields of science and mathematics with me. I've taken through Calculus 2, know how to program... | https://www.lesswrong.com/posts/kXSpHsdL4dknEpqop/would-anyone-be-interested-in-pursuing-the-virtue-of |
# Escape from Alderaan I
*This is Part 4 of a Star Wars fanfiction I began writing 6 months ago. The order you read it in is very important. Start from [the beginning](https://www.lesswrong.com/s/7bCncTCwKHeowtwMR/p/xcJj2zjyDDjgtZy2y) for the best experience and to avoid spoilers.*
---
Obi-Wan Kenobi bought passage ... | https://www.lesswrong.com/posts/vxHsfWcHNYAxzpN3m/escape-from-alderaan-i |
# Gradual Disempowerment, Shell Games and Flinches
Over the past year and half, I've had numerous conversations about the risks we describe in [Gradual Disempowerment](https://www.lesswrong.com/posts/pZhEQieM9otKXhxmd/gradual-disempowerment-systemic-existential-risks-from). (The shortest useful summary of the core arg... | https://www.lesswrong.com/posts/a6FKqvdf6XjFpvKEb/gradual-disempowerment-shell-games-and-flinches |
# The Simplest Good
Common Law AI worked better than anyone expected.
Dr. Sarah Chen was skeptical from the start. "You're essentially training them to be moral judges," she warned during the initial architecture review. "What if they overfit on ethics?" The room laughed. "Better than the alternative," someone quippe... | https://www.lesswrong.com/posts/L6zqL76XnmwLtq7Nd/the-simplest-good |
# Tracing Typos in LLMs: My Attempt at Understanding How Models Correct Misspellings
*This blogpost was created as a part of the* [*AI Safety Fundamentals*](https://aisafetyfundamentals.com/alignment/) *course by BlueDot Impact. All of the code can be found on my* [*GitHub*](https://github.com/idostik/llm-typos-interp... | https://www.lesswrong.com/posts/523bkuMjSjKjG8jn6/tracing-typos-in-llms-my-attempt-at-understanding-how-models |
# Some Theses on Motivational and Directional Feedback
*Note: Probably reinventing the wheel here. Heavily skewed to my areas of interest. Your results may vary.*
1. Feedback can be decomposed into **motivational** ("Keep doing what you're doing!") and **directional** ("Here's what you should do differently . . .").... | https://www.lesswrong.com/posts/cGi76h3AYaA86CSP7/some-theses-on-motivational-and-directional-feedback |
# Keeping Capital is the Challenge
*Capital will likely matter more post-AGI. However, the capital you have been able to raise up to this point and may be able to acquire throughout the maturation of AGI will only retain its value if reliable information remains available.*
Historically, investing has been grounded i... | https://www.lesswrong.com/posts/nFrNEMFi2EwodEXXz/keeping-capital-is-the-challenge |
# Pick two: concise, comprehensive, or clear rules
There once was a post on LessWrong asking [what's with all the bans](https://www.lesswrong.com/posts/GSSHcAoSChaKxjNDZ/what-s-with-all-the-bans-recently#tnk9bf6nhQZKqFeqp)? The OP of that post complained of, among other things, being punished despite not having broken... | https://www.lesswrong.com/posts/CzyGJzESo7vm75KKz/pick-two-concise-comprehensive-or-clear-rules |
# Can we infer the search space of a local optimiser?
Suppose we have a system that we suspect is some form of [local optimiser](https://en.wikipedia.org/wiki/Local_search_(optimization)). Maybe it's gradient descent training a neural network. Maybe it’s a neural network doing in-context learning. Maybe it's a mind, b... | https://www.lesswrong.com/posts/CARCPFXeNB43CPh6z/can-we-infer-the-search-space-of-a-local-optimiser |
# OpenAI releases deep research agent
Experimental research project record [here](https://docs.google.com/document/d/11_A1zMBRSLCbLXOzZoE_xY-sY-11SWUwxz6nSnbub7c/edit?usp=sharing)
Edit: the most important question raised in the comments was: How much of this capability actually comes from the end-to-end task-based RL... | https://www.lesswrong.com/posts/SLiXeZvEkD4XbX7yy/openai-releases-deep-research-agent |
# o3-mini Early Days
New model, new hype cycle, who dis?
On a Friday afternoon, OpenAI was proud to announce the new model o3-mini and also o3-mini-high which is somewhat less mini, or for some other reasoning tasks you might still want o1 if you want a broader knowledge base, or if you’re a pro user o1-pro, while we... | https://www.lesswrong.com/posts/srdxEAcdmetdAiGcz/o3-mini-early-days |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.