url
stringlengths
52
124
post_id
stringlengths
17
17
title
stringlengths
2
248
author
stringlengths
2
49
content
stringlengths
22
295k
date
stringclasses
376 values
https://www.lesswrong.com/posts/SFsifzfZotd3NLJax/utility-engineering-analyzing-and-controlling-emergent-value
SFsifzfZotd3NLJax
Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs
Matrice Jacobine
null
2025-02-12
https://www.lesswrong.com/posts/Sz6xDcS9JHdwMgXPz/why-you-maybe-should-lift-weights-and-how-to
Sz6xDcS9JHdwMgXPz
Why you maybe should lift weights, and How to.
samusasuke
Who this post is for? Someone who either: Wonders if they should start lifting weights, and could be convinced of doing so.Wants to lift weights, and doesn't know where to begin. If this is you, you can skip this first section, though I'm guessing you don't know all the benefits yet. The WHY Benefits of ANY EXERCISE: G...
2025-02-12
https://www.lesswrong.com/posts/k9zLfq2nnqMGJpEAT/teaching-ai-to-reason-this-year-s-most-important-story
k9zLfq2nnqMGJpEAT
Teaching AI to reason: this year's most important story
Benjamin_Todd
This doesn't contain much new to LW readers. I wrote it to try to explain what's going on to a broader audience – I'm posting in case people find it helpful for that. Feedback welcome. Most people think of AI as a pattern-matching chatbot – good at writing emails, terrible at real thinking. They've missed something hug...
2025-02-13
https://www.lesswrong.com/posts/tybYgrTYp3xTaBjox/how-do-the-ceos-respond-to-our-concerns
tybYgrTYp3xTaBjox
how do the CEOs respond to our concerns?
avery-liu
When a decent rationalist walks up to Sam Altman, for example, and presents our arguments for AI doom, how does he respond? What stops us from simply walking up to the people in charge of these training runs, explaining to them the concept of AI doom very slowly and carefully while rebutting all their counterarguments,...
2025-02-11
https://www.lesswrong.com/posts/vvgND6aLjuDR6QzDF/my-model-of-what-is-going-on-with-llms
vvgND6aLjuDR6QzDF
My model of what is going on with LLMs
Amyr
Epistemic status: You probably already know if you want to read this kind of post, but in case you have not decided: my impression is that people are acting very confused about what we can conclude about scaling LLMs from the evidence, and I believe my mental model cuts through a lot of this confusion - I have tried to...
2025-02-13
https://www.lesswrong.com/posts/r86BBAqLHXrZ4mWWA/what-goals-will-ais-have-a-list-of-hypotheses
r86BBAqLHXrZ4mWWA
What goals will AIs have? A list of hypotheses
daniel-kokotajlo
My colleagues and I have written a scenario in which AGI-level AI systems are trained around 2027 using something like the current paradigm: LLM-based agents (but with recurrence/neuralese) trained with vast amounts of outcome-based reinforcement learning on diverse challenging short, medium, and long-horizon tasks, wi...
2025-03-03
https://www.lesswrong.com/posts/bsTzgG3cRrsgbGtCc/extended-analogy-between-humans-corporations-and-ais
bsTzgG3cRrsgbGtCc
Extended analogy between humans, corporations, and AIs.
daniel-kokotajlo
There are three main ways to try to understand and reason about powerful future AGI agents: Using formal models designed to predict the behavior of powerful general agents, such as expected utility maximization and variants thereof (explored in game theory and decision theory).Comparing & contrasting powerful future AG...
2025-02-13
https://www.lesswrong.com/posts/7KijyCL8WNP8JnWCR/gradient-anatomy-s-hallucination-robustness-in-medical-q-and
7KijyCL8WNP8JnWCR
Gradient Anatomy's - Hallucination Robustness in Medical Q&A
diego-sabajo
TL;DR We investigated reducing hallucinations in medical question-answering with Llama-3.1-8B-Instruct. Using Goodfire's Sparse Auto-Encoder (SAE) we identified neural features associated with accurate and hallucinated responses. Our study found that features related to the model’s awareness of its own knowledge limita...
2025-02-12
https://www.lesswrong.com/posts/3XaizFzbcWAEp8G6o/ai-safety-at-the-frontier-paper-highlights-january-25
3XaizFzbcWAEp8G6o
AI Safety at the Frontier: Paper Highlights, January '25
gasteigerjo
This is the selection of AI safety papers from my blog "AI Safety at the Frontier". The selection primarily covers ML-oriented research and frontier models. It's primarily concerned with papers (arXiv, conferences etc.). tl;dr Paper of the month: Constitutional Classifiers demonstrate a promising defense against univer...
2025-02-11
https://www.lesswrong.com/posts/boB3hJiZijxM3J6Ed/comparing-the-effectiveness-of-top-down-and-bottom-up
boB3hJiZijxM3J6Ed
Comparing the effectiveness of top-down and bottom-up activation steering for bypassing refusal on harmful prompts
ana-kapros
TL;DR This project compares the effectiveness of top-down and bottom-up activation steering methods in controlling refusal behaviour. In line with prior work,[1] we find that top-down methods outperform bottom-up ones in behaviour steering, as measured using HarmBench. Yet, a hybrid approach is even more effective (pro...
2025-02-12
https://www.lesswrong.com/posts/MmxtzkXLDWnyicQpF/the-news-is-never-neglected
MmxtzkXLDWnyicQpF
The News is Never Neglected
lsusr
Dear Lsusr, I am inspired by your stories about Effective Evil. My teachers at school tell me it is my civic responsibility to watch the news. Should I reverse this advice? Or should I watch the news like everyone else, except use what I learn for evil? Sincerely, [redacted] Dear [redacted], If you want to make an impa...
2025-02-11
https://www.lesswrong.com/posts/J6rgqYjj7Cm89Xu2w/where-would-good-forecasts-most-help-ai-governance-efforts-1
J6rgqYjj7Cm89Xu2w
Where Would Good Forecasts Most Help AI Governance Efforts?
Violet Hour
Thanks to Josh Rosenberg for comments and discussion. Introduction One of LessWrong’s historical troves is its pre-ChatGPT AGI forecasts. Not just for the specific predictions people offered, but for observing which sorts of generative processes produced which kinds of forecasts. For instance: [Nuno (Median AGI Timelin...
2025-02-11
https://www.lesswrong.com/posts/ysghKGYev8DwPDY32/what-about-the-horses
ysghKGYev8DwPDY32
What About The Horses?
maxwell-tabarrok
In a previous post, I argued that AGI would not make human labor worthless. One of the most common responses was to ask about the horses. Technology resulted in mass unemployment and population collapse for horses even though they must have had some comparative advantage with more advanced engines. Why couldn’t the sam...
2025-02-11
https://www.lesswrong.com/posts/dLnwRFLFmHKuurTX2/rethinking-ai-safety-approach-in-the-era-of-open-source-ai
dLnwRFLFmHKuurTX2
Rethinking AI Safety Approach in the Era of Open-Source AI
weibing-wang
Open-Source AI Undermines Traditional AI Safety Approach In the past years, the mainstream approach to AI safety has been "AI alignment + access control." In simple terms, this means allowing a small number of regulated organizations to develop the most advanced AI systems, ensuring that these AIs' goals are aligned wi...
2025-02-11
https://www.lesswrong.com/posts/CJ4yywLBkdRALc4sT/on-deliberative-alignment
CJ4yywLBkdRALc4sT
On Deliberative Alignment
Zvi
Not too long ago, OpenAI presented a paper on their new strategy of Deliberative Alignment. The way this works is that they tell the model what its policies are and then have the model think about whether it should comply with a request. This is an important transition, so this post will go over my perspective on the n...
2025-02-11
https://www.lesswrong.com/posts/uA3JcjbmrZRohToRY/world-citizen-assembly-about-ai-announcement
uA3JcjbmrZRohToRY
World Citizen Assembly about AI - Announcement
Camille Berger
null
2025-02-11
https://www.lesswrong.com/posts/sekmz9EiBD6ByZpyp/detecting-ai-agent-failure-modes-in-simulations
sekmz9EiBD6ByZpyp
Detecting AI Agent Failure Modes in Simulations
michael-soareverix
AI agents have become significantly more common in the last few months. They’re used for web scraping,[1][2] robotics and automation,[3] and are even being deployed for military use.[4] As we integrate these agents into critical processes, it is important to simulate their behavior in low-risk environments. In this pos...
2025-02-11
https://www.lesswrong.com/posts/LaruPAWaZk9KpC25A/rational-utopia-and-narrow-way-there-place-asi-dealing-with
LaruPAWaZk9KpC25A
Rational Utopia & Narrow Way There: Place ASI, Dealing With Dystopias, New Ethics... (V. 4)
ank
(This is the result of three years of thinking and modeling hyper‑futuristic and current ethical systems. The first post in the series. Everything described here can be modeled mathematically—it’s essentially geometry. I take as an axiom that every agent in the multiverse experiences real pain and pleasure. Sorry for t...
2025-02-11
https://www.lesswrong.com/posts/tdb76S4viiTHfFr2u/why-did-elon-musk-just-offer-to-buy-control-of-openai-for
tdb76S4viiTHfFr2u
Why Did Elon Musk Just Offer to Buy Control of OpenAI for $100 Billion?
garrison
null
2025-02-11
https://www.lesswrong.com/posts/wpz5bXhGxPBY6LGDk/positive-directions
wpz5bXhGxPBY6LGDk
Positive Directions
geoffrey-wood
Let's set the scene - Nihilism We are chunks of self important jelly staggering about on the surface of a tiny nugget of rock in a second rate solar system. Sol's system is ~ four billion years old We have been “Ourselves” (Anatomically modern) for ~200,000 years Countless people have lived and died, and you are but on...
2025-02-11
https://www.lesswrong.com/posts/dJ7XFvqh5oWQbB4CJ/arguing-for-the-truth-an-inference-only-study-into-ai-debate
dJ7XFvqh5oWQbB4CJ
Arguing for the Truth? An Inference-Only Study into AI Debate
denisemester
💡 TL;DR: Can AI debate be a reliable tool for truth-seeking? In this inference-only experiment (no fine-tuning), I tested whether Claude 3.5 Sonnet and Gemini 1.5 Pro could engage in structured debates over factual questions from BoolQ and MMLU datasets, with GPT-3.5 Turbo acting as an impartial judge. The findings we...
2025-02-11
https://www.lesswrong.com/posts/yBtXoDqfjFXkEWLMc/logical-correlation
yBtXoDqfjFXkEWLMc
Logical Correlation
niplav
In which to compare how similarly programs compute their outputs, naïvely and less naïvely. Logical Correlation Attention conservation notice: Premature formalization, ab-hoc mathematical definition. Motivation, Briefly In the twin prisoners dilemma, I cooperate with my twin because we're implementing the same algorith...
2025-02-10
https://www.lesswrong.com/posts/zEKzCLzGJTXgoqSXf/lw-acx-social-meetup
zEKzCLzGJTXgoqSXf
LW/ACX social meetup
stefan-1
Come by! Meet interesting people, chat interesting chat! Normally we just chat about whatever comes up. Past topics of conversation have included AI alignment, decision theory (Newcomb's paradox etc), progress in AI and much much more. (We will be on the second floor of the Condeco café, look for a book on the table)
2025-02-10
https://www.lesswrong.com/posts/J59hfYefh6yA4wzLD/a-bearish-take-on-ai-as-a-treat
J59hfYefh6yA4wzLD
A Bearish Take on AI, as a Treat
cartier-gucciscarf
The implicit model that I have regarding the world around me on most topics is there is a truth on a matter, a select group of people and organizations who are closest to that truth, and an assortment of groups who espouse bad takes either out of malice or stupidity. This was, to a close approximation, my opinion about...
2025-02-10
https://www.lesswrong.com/posts/S8mEHmTnCPgYqfazv/notes-on-occam-via-solomonoff-vs-hierarchical-bayes
S8mEHmTnCPgYqfazv
Notes on Occam via Solomonoff vs. hierarchical Bayes
JesseClifton
Crossposted from my Substack. Intuitively, simpler theories are better all else equal. It also seems like finding a way to justify assigning higher prior probability to simpler theories is one of the more promising ways of approaching the problem of induction. In some places, Solomonoff induction (SI) seems to be consi...
2025-02-10
https://www.lesswrong.com/posts/3ZBmKDpAJJahRM248/proof-idea-slt-to-ait
3ZBmKDpAJJahRM248
Proof idea: SLT to AIT
Lblack
I think we may be able to prove that Bayesian learning on transformers[1] or recurrent neural networks with a uniform[2] prior over parameters is equivalent to a form of Solomonoff induction over a set of computationally-bounded programs. This bounded Solomonoff induction would still be 'approximately optimal' in a sen...
2025-02-10
https://www.lesswrong.com/posts/ECLEsydXxvtK3XxMs/sleeping-beauty-an-accuracy-based-approach
ECLEsydXxvtK3XxMs
Sleeping Beauty: an Accuracy-based Approach
glauberdebona
This post does not propose a solution to the Sleeping Beauty problem, but presents arguments based on accuracy for thirders, halfers and double-halfers. A more detailed draft paper can be found here. Summary Accuracy-based arguments claim that one should plan to adopt posterior credences that maximize the expected (acc...
2025-02-10
https://www.lesswrong.com/posts/Ddc5dArm8DXQK9ChC/political-idolatry
Ddc5dArm8DXQK9ChC
Political Idolatry
arturo-macias
Idolatry is the worship of non-conscious objects, sometimes falsely attributing consciousness to them, sometimes putting the value of some admittedly nonconscious being over that of conscious beings. Idolatry leads to human sacrifice because to prove your idol more important than the human soul the natural test is to s...
2025-02-10
https://www.lesswrong.com/posts/QpaWHYEQomyQTBKw5/nonpartisan-ai-safety
QpaWHYEQomyQTBKw5
Nonpartisan AI safety
yair-halberstadt
AI alignment is probably the most pressing issue of our time. Unfortunately it's also become one of the most controversial, with AI accelerationists accusing AI doomers/ai-not-kill-everyoneism-ers of being luddites who would rather keep humanity shackled to the horse and plow than risk any progress, whilst the doomers ...
2025-02-10
https://www.lesswrong.com/posts/ARZ5c99k9M2RJJtdT/opinion-article-scoring-system
ARZ5c99k9M2RJJtdT
Opinion Article Scoring System
ciaran
Here I propose a system for scoring media opinion articles. It is part prediction markets - as there is a small amount of money involved - and part forecasting science mechanism design. Journalists that publish an article on the platform must do so with an accompanying stake. Readers (whether human or AI) that wish to ...
2025-02-10
https://www.lesswrong.com/posts/L7xmssgoKXPJAbz4D/beyond-elo-rethinking-chess-skill-as-a-multidimensional
L7xmssgoKXPJAbz4D
Beyond ELO: Rethinking Chess Skill as a Multidimensional Random Variable
oliver-oswald
Introduction The traditional ELO rating system reduces a player's ability to a single scalar value E, from which win probabilities are computed via a logistic function of the rating difference. While pragmatic, this one-dimensional approach may obscure the rich, multifaceted nature of chess skill. For instance, factors...
2025-02-10
https://www.lesswrong.com/posts/xcMngBervaSCgL9cu/levels-of-friction
xcMngBervaSCgL9cu
Levels of Friction
Zvi
Scott Alexander famously warned us to Beware Trivial Inconveniences. When you make a thing easy to do, people often do vastly more of it. When you put up barriers, even highly solvable ones, people often do vastly less. Let us take this seriously, and carefully choose what inconveniences to put where. Let us also take ...
2025-02-10
https://www.lesswrong.com/posts/ds98kG3FKpKv7665W/a-simulation-of-automation-economics
ds98kG3FKpKv7665W
A Simulation of Automation economics?
qbolec
Looks like even respected people disagree about the effects of automation of most jobs on prices, ability to earn and trade, ect. Is there a game which I could play to gain intuitions about it? A game where there are some things and services people need to get to survive, some resources like time, abilities or real est...
2025-02-10
https://www.lesswrong.com/posts/geRo75Xi9baHcwzht/claude-is-more-anxious-than-gpt-personality-is-an-axis-of-2
geRo75Xi9baHcwzht
Claude is More Anxious than GPT; Personality is an axis of interpretability in language models
future_detective
Aggregate Personality Differences Users of Claude and GPT will be the first to tell you that the models have their own personality. Some users make decisions based on “who” they prefer to talk to. In my own experience, I’ve found Claude to be more deferential, GPT more clinical. In "We Can Solve Psychology ith Text Emb...
2025-02-10
https://www.lesswrong.com/posts/e8giAk8bmwDFrJxx2/should-i-divest-from-ai
e8giAk8bmwDFrJxx2
Should I Divest from AI?
OKlogic
In recent years, AI has been all the rage in the stock market, and there is no reason to see that slowing down. With the picture of AGI on the horizon becoming clearer and clearer, faster and smarter models being released, and more and more investment being poured into AI stocks, it seems inevitable that prices will co...
2025-02-10
https://www.lesswrong.com/posts/emdeWndtjD8QxzgS5/openai-lied-about-sft-vs-rlhf
emdeWndtjD8QxzgS5
OpenAI lied about SFT vs. RLHF
sanxiyn
I used to think while OpenAI is pretty deceitful (eg for-profit conversion) it generally won't lie about its research. This is a pretty definitive case of lying, so I updated accordingly. I am posting here because it doesn't seem to be widely known.
2025-02-10
https://www.lesswrong.com/posts/dhLmbpk346e7ARdnP/self-blackmail-and-alternatives
dhLmbpk346e7ARdnP
"Self-Blackmail" and Alternatives
jessica.liu.taylor
Ziz has been in the news lately. Instead of discussing that, I'll discuss an early blog post, "Self-Blackmail". This is a topic I also talked with Ziz about in person, although not a lot. Let's start with a very normal thing people do: make New Year's resolutions. They might resolve that, for example, they will do stre...
2025-02-09
https://www.lesswrong.com/posts/X9dy7LLaBbLcq8jky/altman-blog-on-post-agi-world
X9dy7LLaBbLcq8jky
Altman blog on post-AGI world
Julian Bradshaw
First part just talks about scaling laws, nothing really new. Second part is apparently his latest thoughts on a post-AGI world. Key part: While we never want to be reckless and there will likely be some major decisions and limitations related to AGI safety that will be unpopular, directionally, as we get closer to ach...
2025-02-09
https://www.lesswrong.com/posts/3L3ZGSucpx4ypEou7/ml4good-colombia-applications-open-to-latam-participants
3L3ZGSucpx4ypEou7
ML4Good Colombia - Applications Open to LatAm Participants
alejandro-acelas
Applications are open for ML4Good Colombia April 2025 In partnership with AI Safety Colombia, ML4Good is running an intensive 10-day bootcamp focusing on upskilling in deep learning, exploring governance, and delving into conceptual topics for individuals who are motivated to work on addressing the risks posed by advan...
2025-02-10
https://www.lesswrong.com/posts/g3fH7YzthnXwtCt6g/forecasting-newsletter-2-2025-forecasting-meetup-network
g3fH7YzthnXwtCt6g
Forecasting newsletter #2/2025: Forecasting meetup network
Radamantis
Highlights Forecasting meetup network (a) looking for volunteers. If you want to host a meetup in your city, send an email at forecastingmeetupnetwork@gmail.com.Caroline Pham moves up to Chairman of the CFTC. She is much friendlier to prediction markets and has spent years writting dissents againsts regulatory overreac...
2025-02-09
https://www.lesswrong.com/posts/HZ4sM28jc8JBcznDG/how-identical-twin-sisters-feel-about-nieces-vs-their-own
HZ4sM28jc8JBcznDG
How identical twin sisters feel about nieces vs their own daughters
dave-lindbergh
(cross posted from https://mugwumpery.com/how-identical-twin-sisters-feel-about-nieces-vs-their-own-daughters/) It seems to be generally assumed that twin sisters feel the same way as other sisters – closer to their own children. But per Hamilton/Trivers, they shouldn’t. They should feel equally related and care equall...
2025-02-09
https://www.lesswrong.com/posts/XnAHe6iFfkEwTYgsA/the-structure-of-professional-revolutions
XnAHe6iFfkEwTYgsA
The Structure of Professional Revolutions
JohnBuridan
An expert is not merely someone who has memorized data but someone who has internalized the structure of knowledge itself. This is why we call them PhDs—Doctors of Philosophy. Their expertise extends beyond isolated facts to the organizing principles that connect those facts, allowing them to wield knowledge in novel w...
2025-02-09
https://www.lesswrong.com/posts/2q4qbwEFEjhRcHSk2/how-do-you-make-a-250x-better-vaccine-at-1-10-the-cost
2q4qbwEFEjhRcHSk2
How do you make a 250x better vaccine at 1/10 the cost? Develop it in India.
abhishaike-mahajan
(I made a vaccinology/policy-based podcast! A very long one! If you'd like to avoid the summary below, here is the Youtube link and Substack link.). Summary: There's a lot of discussion these days on how China's biotech market is on track to bypass the US's. I wondered: shouldn't we have observed the exact same phenome...
2025-02-09
https://www.lesswrong.com/posts/dZfFLKwHWXbvamdja/less-laptop-velcro
dZfFLKwHWXbvamdja
Less Laptop Velcro
jkaufman
A year ago I broke my laptop screen, and took the opportunity to build something I've always wanted: a monitor that folds vertically so I don't have to bend my neck: A few months ago my cracked-screen laptop finished dying, and I got a new one. I use the stacked monitor a lot less now, since for quick things the built...
2025-02-09
https://www.lesswrong.com/posts/zavyum4dxEAqs6wHt/undergrad-ai-safety-conference
zavyum4dxEAqs6wHt
Undergrad AI Safety Conference
joanna-j-1
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 is excited to announce the Chicago Symposium on Transformative AI, an undergraduate AI Safety conference taking place on the...
2025-02-19
https://www.lesswrong.com/posts/q56ZEDpQtqTbJ8af2/axrp-episode-38-7-anthony-aguirre-on-the-future-of-life
q56ZEDpQtqTbJ8af2
AXRP Episode 38.7 - Anthony Aguirre on the Future of Life Institute
DanielFilan
YouTube link The Future of Life Institute is one of the oldest and most prominant organizations in the AI existential safety space, working on such topics as the AI pause open letter and how the EU AI Act can be improved. Metaculus is one of the premier forecasting sites on the internet. Behind both of them lie one man...
2025-02-09
https://www.lesswrong.com/posts/dovhoCuaEzSGBLzsB/job-ad-lisa-ceo
dovhoCuaEzSGBLzsB
[Job ad] LISA CEO
ryankidd44
Overview Job Title: Chief Executive OfficerCompany Name: London Initiative for Safe AI (LISA)Location: Old Street, LondonDuration: Full-timeSalary Range: £95-125k (more may be available for an exceptional candidate)Application Deadline: Monday 24th February We are looking for a Chief Executive Officer with experience a...
2025-02-09
https://www.lesswrong.com/posts/98K94XXGxfxdc9Pyd/p-s-risks-to-contemporary-humans
98K94XXGxfxdc9Pyd
p(s-risks to contemporary humans)?
mhampton
Epistemic status These are my cursory thoughts on this topic after having read about it over a few days and conversed with some other people. I still have high uncertainty and am raising questions that may address some uncertainties. Content warning Discussion of risks of astronomical suffering[1] Why focus on s-risks ...
2025-02-08
https://www.lesswrong.com/posts/skEhdgiK6T5HzMhis/think-it-faster-worksheet
skEhdgiK6T5HzMhis
"Think it Faster" worksheet
Raemon
This is a succinct worksheet version of the "Think It Faster" Exercise. [1] You can use this worksheet either for purposeful practice, after completing some kind of challenging/confusing intellectual exercise (such as Thinking Physics or Baba is You). Or, if in your real life work you find something took a noticeably l...
2025-02-08
https://www.lesswrong.com/posts/ZB3DqPp3CwcPNNk7y/visual-reference-for-frontier-large-language-models
ZB3DqPp3CwcPNNk7y
Visual Reference for Frontier Large Language Models
kenakofer
Hopefully this can be a helpful visual reference for the development and features of frontier large language models in the last year-ish. We are always open to feedback on how the reference could be improved. FAQ: Q: Which models/companies are included? A: We include LLMs that are noteworthy in capabilities, price, or ...
2025-02-11
https://www.lesswrong.com/posts/vQYPBatJKaGkfDfCt/closed-ended-questions-aren-t-as-hard-as-you-think
vQYPBatJKaGkfDfCt
Closed-ended questions aren't as hard as you think
electroswing
Summary In this short post, I argue that closed-ended questions, even those of arbitrary difficulty, are not as difficult as they may appear. In particular, I argue that the benchmark HLE is probably easier than it may first seem.[1] Specifically, I argue: Crowd workers find it easier to write easy questions than hard ...
2025-02-19
https://www.lesswrong.com/posts/3MCwMkP6cJmrxMmat/can-knowledge-hurt-you-the-dangers-of-infohazards-and
3MCwMkP6cJmrxMmat
Can Knowledge Hurt You? The Dangers of Infohazards (and Exfohazards)
aggliu
In this Rational Animations video, we look at dangerous knowledge: information hazards (infohazards) and external information hazards (exfohazards).  We talk about one way they can be classified, what kinds of dangers they pose, and the dangers that come from too much secrecy. The primary scriptwriter was Allen Liu (th...
2025-02-08
https://www.lesswrong.com/posts/oBo7tGTvP9f26M98C/gary-marcus-now-saying-ai-can-t-do-things-it-can-already-do
oBo7tGTvP9f26M98C
Gary Marcus now saying AI can't do things it can already do
Benjamin_Todd
January 2020, Gary Marcus wrote GPT-2 And The Nature Of Intelligence, demonstrating a bunch of easy problems that GPT-2 couldn’t get right. He concluded these were “a clear sign that it is time to consider investing in different approaches.” Two years later, GPT-3 could get most of these right. Marcus wrote a new list ...
2025-02-09
https://www.lesswrong.com/posts/HqHcvKb6kw5aRdoMA/preserving-epistemic-novelty-in-ai-experiments-insights-and
HqHcvKb6kw5aRdoMA
Preserving Epistemic Novelty in AI: Experiments, Insights, and the Case for Decentralized Collective Intelligence
andy-e-williams
Introduction In my recent experiments with AI models, I have encountered a fundamental problem: even when novel epistemic insights are introduced into AI interactions, the models tend to “flatten” or reframe these ideas into existing, consensus‐based frameworks. This compression of novelty limits an AI’s ability to evo...
2025-02-08
https://www.lesswrong.com/posts/KEdr7E5SfaqjczFgD/technical-comparison-of-deepseek-novasky-s1-helix-p0
KEdr7E5SfaqjczFgD
Technical comparison of Deepseek, Novasky, S1, Helix, P0
Juliezhanggg
Comparing Novasky with S1: NovaSky by Berkeley club, S1 by Feifei Li (arXiv:2501.19393), are the players who don't have capital or compute, mainly focuses on developing method that finetune a large language model with curated minimal reasoning datasets. Sky-T1 trained an entire model, with datasets from diverse domains...
2025-02-25
https://www.lesswrong.com/posts/feknAa3hQgLG2ZAna/cross-layer-feature-alignment-and-steering-in-large-language-2
feknAa3hQgLG2ZAna
Cross-Layer Feature Alignment and Steering in Large Language Model
dlaptev
The text below is a brief summary of our research in mechanistic interpretability. First, this article discusses the motivation behind our work. Second, it provides an overview of our previous work. Finally, we outline the future directions we consider important. Introduction and Motivation Large language models (LLMs)...
2025-02-08
https://www.lesswrong.com/posts/cSPcey7FsKpNzLXuF/ai-safety-oversights
cSPcey7FsKpNzLXuF
AI Safety Oversights
davey-morse
The field of AI Safety at large is making four key oversights: LLMs vs. Agents. AI safety researchers have been thorough in examining safety concerns from LLMs (bias, deception, accuracy, child safety, etc). Agents powered by LLMs, however,  are more dangerous and dangerous in different ways than LLMs are alone. The fi...
2025-02-08
https://www.lesswrong.com/posts/ZpnPxn433EMWw2t6m/wiki-on-suspects-in-lind-zajko-and-maland-killings
ZpnPxn433EMWw2t6m
Wiki on Suspects in Lind, Zajko, and Maland Killings
Rebecca_Records
Hey everyone, I've been following the news about the killings linked to LaSota, Zajko and associates, and right now, finding all the relevant information is a challenge. Either you have to dig through scattered sources, or you’re stuck reading a single, extremely long Google Doc to get the full picture. To make things ...
2025-02-08
https://www.lesswrong.com/posts/Cs8n4zeaGtxnvCjhe/two-hemispheres-i-do-not-think-it-means-what-you-think-it
Cs8n4zeaGtxnvCjhe
Two hemispheres - I do not think it means what you think it means
Viliam
I am going to address some misconceptions about brain hemispheres -- in popular culture, and in Zizian theory. The latter, because the madness must stop. The former, because it provided a foundation for the latter. * Two hemispheres in popular culture About 99% of animals are bilaterally symmetric -- the left side and ...
2025-02-09
https://www.lesswrong.com/posts/Hgj84BSitfSQnfwW6/so-you-want-to-make-marginal-progress
Hgj84BSitfSQnfwW6
So You Want To Make Marginal Progress...
johnswentworth
Once upon a time, in ye olden days of strange names and before google maps, seven friends needed to figure out a driving route from their parking lot in San Francisco (SF) down south to their hotel in Los Angeles (LA). The first friend, Alice, tackled the “central bottleneck” of the problem: she figured out that they p...
2025-02-07
https://www.lesswrong.com/posts/s5SjpHGaKFAizbieu/reasons-based-choice-and-cluelessness
s5SjpHGaKFAizbieu
Reasons-based choice and cluelessness
JesseClifton
Crossposted from my Substack. Rational choice theory is commonly thought of as being about what to do in light of our beliefs and preferences. But our beliefs and preferences come from somewhere. I would say that we believe and prefer things for reasons. My evidence gives me reason to believe I am presently in an airpo...
2025-02-07
https://www.lesswrong.com/posts/26SHhxK2yYQbh7ors/research-directions-open-phil-wants-to-fund-in-technical-ai
26SHhxK2yYQbh7ors
Research directions Open Phil wants to fund in technical AI safety
jake_mendel
The Open Philanthropy has just launched a large new Request for Proposals for technical AI safety research. Here we're sharing a reference guide, created as part of that RFP, which describes what projects we'd like to see across 21 research directions in technical AI safety. This guide provides an opinionated overview ...
2025-02-08
https://www.lesswrong.com/posts/pzhykwLuaNDFPvqsh/request-for-information-for-a-new-us-ai-action-plan-ostp-rfi
pzhykwLuaNDFPvqsh
Request for Information for a new US AI Action Plan (OSTP RFI)
agucova
null
2025-02-07
https://www.lesswrong.com/posts/4x4QFzmdWadgr7mdj/translation-in-the-age-of-ai-don-t-look-for-unicorns
4x4QFzmdWadgr7mdj
[Translation] In the Age of AI don't Look for Unicorns
mushroomsoup
Translator's note: This is another article from Jeffery Ding's Around the Horn. In short, the article suggests that a good metric of AI adoption is the number of daily average tokens used by a company. Companies which achieved successful adoption are those who use more than one billion tokens everyday. In China there a...
2025-02-07
https://www.lesswrong.com/posts/vqsXncwxsxti5L9Ne/racing-towards-fusion-and-ai
vqsXncwxsxti5L9Ne
Racing Towards Fusion and AI
jeffrey-heninger
Racing Towards a New Technology Is a Collective Choice, Not an Inevitable Consequence of Incentives Before I started thinking about AI policy, I was working in the trying-to-get-fusion industry. There are some significant similarities between AI and fusion. Both are emerging technologies.Both have the potential to have...
2025-02-07
https://www.lesswrong.com/posts/KFJ2LFogYqzfGB3uX/how-ai-takeover-might-happen-in-2-years
KFJ2LFogYqzfGB3uX
How AI Takeover Might Happen in 2 Years
joshua-clymer
I’m not a natural “doomsayer.” But unfortunately, part of my job as an AI safety researcher is to think about the more troubling scenarios. I’m like a mechanic scrambling last-minute checks before Apollo 13 takes off. If you ask for my take on the situation, I won’t comment on the quality of the in-flight entertainment...
2025-02-07
https://www.lesswrong.com/posts/kpGEp4Jem7PrQhPRW/high-level-machine-intelligence-and-full-automation-of-labor
kpGEp4Jem7PrQhPRW
'High-Level Machine Intelligence' and 'Full Automation of Labor' in the AI Impacts Surveys
jeffrey-heninger
The 60+ Year Gap AI Impacts has run three surveys (2016, 2022, & 2023) asking AI researchers about how they expect AI to develop in the future.[1] One of the key questions addressed was when AI capabilities will exceed human capabilities. The surveys did not ask directly about 'Artificial General Intelligence' (AGI). I...
2025-02-07
https://www.lesswrong.com/posts/rGvkirbemm9deAM3g/the-devil-s-ontology
rGvkirbemm9deAM3g
the devil's ontology
lostinwilliamsburg
imagine you’re playing a game where some rules encoded in words are so special that no one is allowed to touch or change them. these special rules have special properties, so no one questions them. the devil, in this case, is like a sneaky player who hides behind those rules, using them to confuse everyone else. they c...
2025-02-07
https://www.lesswrong.com/posts/gsj3TWdcBxwkm9eNt/10-year-timelines-remain-unlikely-despite-deepseek-and-o3
gsj3TWdcBxwkm9eNt
≤10-year Timelines Remain Unlikely Despite DeepSeek and o3
sil-ver
[Thanks to Steven Byrnes for feedback and the idea for section §3.1. Also thanks to Justis from the LW feedback team.] Remember this? Or this? The images are from WaitButWhy, but the idea was voiced by many prominent alignment people, including Eliezer Yudkowsky and Nick Bostrom. The argument is that the difference in ...
2025-02-13
https://www.lesswrong.com/posts/etqbEF4yWoGBEaPro/on-the-meta-and-deepmind-safety-frameworks
etqbEF4yWoGBEaPro
On the Meta and DeepMind Safety Frameworks
Zvi
This week we got a revision of DeepMind’s safety framework, and the first version of Meta’s framework. This post covers both of them. Table of Contents Meta’s RSP (Frontier AI Framework). DeepMind Updates its Frontier Safety Framework. What About Risk Governance. Where Do We Go From Here? Here are links for previous co...
2025-02-07
https://www.lesswrong.com/posts/mNKRibWTsx32J8GzW/request-for-proposals-improving-capability-evaluations
mNKRibWTsx32J8GzW
Request for proposals: improving capability evaluations
cb
Open Philanthropy is launching an RFP for work on AI capability evaluations. We're looking to fund three types of work: Global Catastrophic Risk (GCR)-relevant capability benchmarks for AI agentsResearch to improve our understanding of how capabilities develop and scaleSolutions for enabling meaningful third-party eval...
2025-02-07
https://www.lesswrong.com/posts/GxpSFtnHccNNqBz4N/baumol-effect-vs-jevons-paradox
GxpSFtnHccNNqBz4N
Baumol effect vs Jevons paradox
Hzn
Key points. The Baumol effect & Jevons paradox are 2 claims regarding the effect of increasing efficiency of a good or sector. Although not incompatible, they are at odds; one suggesting relative decline, the other suggesting absolute growth. I examine these & find that they are often defined & discussed in a confused ...
2025-02-10
https://www.lesswrong.com/posts/Xt9r4SNNuYxW83tmo/a-computational-no-coincidence-principle
Xt9r4SNNuYxW83tmo
A computational no-coincidence principle
UnexpectedValues
This post presents a conjecture formulated at the Alignment Research Center 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 a while now, but we believe the conjecture is interesting in its own right. In a rece...
2025-02-14
https://www.lesswrong.com/posts/YXNeA3RyRrrRWS37A/a-problem-to-solve-before-building-a-deception-detector
YXNeA3RyRrrRWS37A
A Problem to Solve Before Building a Deception Detector
ea-1
TL;DR: If you are thinking of using interpretability to help with strategic deception, then there's likely a problem you need to solve first: how are intentional descriptions (like deception) related to algorithmic ones (like understanding the mechanisms models use)? We discuss this problem and try to outline some cons...
2025-02-07
https://www.lesswrong.com/posts/7HpFuGBLcdyjHj8tc/when-you-downvote-explain-why
7HpFuGBLcdyjHj8tc
When you downvote, explain why
avery-liu
NOTE: this is not site policy, just my personal suggestion Being a newcomer and having your post downvoted can be very discouraging. This isn't necessarily a bad thing—obviously we want to discourage people from posting things that are not worth our time to read—but it doesn't provide much feedback other than "somethin...
2025-02-07
https://www.lesswrong.com/posts/kJzTmuhSZ7ufcgGTv/medical-windfall-prizes-1
kJzTmuhSZ7ufcgGTv
Medical Windfall Prizes
PeterMcCluskey
Summary AI may produce a windfall surge in government revenues in 5 to 10 years. I want governments to spending a small fraction of that windfall on retroactively rewarding entities in proportion to how they have contributed to medical advances, measured by lives saved and suffering avoided. Motivations This post was i...
2025-02-06
https://www.lesswrong.com/posts/fwSnz5oNnq8HxQjTL/arbital-has-been-imported-to-lesswrong
fwSnz5oNnq8HxQjTL
Arbital has been imported to LessWrong
T3t
Arbital was envisioned as a successor to Wikipedia. The project was discontinued in 2017, but not before many new features had been built and a substantial amount of writing about AI alignment and mathematics had been published on the website. If you've tried using Arbital.com the last few years, you might have noticed...
2025-02-20
https://www.lesswrong.com/posts/nHDhst47yzDCpGstx/seven-sources-of-goals-in-llm-agents
nHDhst47yzDCpGstx
Seven sources of goals in LLM agents
Seth Herd
LLM agents[1] seem reasonably likely to become our first takeover-capable AGIs.[2]  LLMs already have complex "psychologies," and using them to power more sophisticated agents will create even more complex "minds." A profusion of competing goals is one barrier to aligning this type of AGI. Goal sources for LLM agents c...
2025-02-08
https://www.lesswrong.com/posts/rJnygfgGyZ3ovo9Yu/aisn-47-reasoning-models
rJnygfgGyZ3ovo9Yu
AISN #47: Reasoning Models
corin-katzke
Welcome to the AI Safety Newsletter by the Center for AI Safety. We discuss developments in AI and AI safety. No technical background required. Listen to the AI Safety Newsletter for free on Spotify or Apple Podcasts. Reasoning Models DeepSeek-R1 has been one of the most significant model releases since ChatGPT. After ...
2025-02-06
https://www.lesswrong.com/posts/wbJxRNxuezvsGFEWv/open-philanthropy-technical-ai-safety-rfp-usd40m-available
wbJxRNxuezvsGFEWv
Open Philanthropy Technical AI Safety RFP - $40M Available Across 21 Research Areas
jake_mendel
Open Philanthropy is launching a big new Request for Proposals for technical AI safety research, with plans to fund roughly $40M in grants over the next 5 months, and available funding for substantially more depending on application quality. Applications (here) start with a simple 300 word expression of interest and ar...
2025-02-06
https://www.lesswrong.com/posts/viZhwKytJ6Rqahymf/wild-animal-suffering-is-the-worst-thing-in-the-world
viZhwKytJ6Rqahymf
Wild Animal Suffering Is The Worst Thing In The World
omnizoid
Crossposted from my blog which many people are saying you should check out! Imagine that you came across an injured deer on the road. She was in immense pain, perhaps having been mauled by a bear or seriously injured in some other way. Two things are obvious: If you could greatly help her at small cost, you should do s...
2025-02-06
https://www.lesswrong.com/posts/9pGbTz6c78PGwJein/detecting-strategic-deception-using-linear-probes
9pGbTz6c78PGwJein
Detecting Strategic Deception Using Linear Probes
nicholas-goldowsky-dill
Can you tell when an LLM is lying from the activations? Are simple methods good enough? We recently published a paper investigating if linear probes detect when Llama is deceptive. Abstract: AI models might use deceptive strategies as part of scheming or misaligned behaviour. Monitoring outputs alone is insufficient, s...
2025-02-06
https://www.lesswrong.com/posts/2w6hjptanQ3cDyDw7/methods-for-strong-human-germline-engineering
2w6hjptanQ3cDyDw7
Methods for strong human germline engineering
TsviBT
PDF version. Image sizes best at berkeleygenomics.org with a wide monitor. Twitter thread Introduction This article summarizes the technical pathways to make healthy humans with significantly modified genomes. These are the pathways that I'm aware of and that seem plausibly feasible in the next two decades. A short sum...
2025-03-03
https://www.lesswrong.com/posts/rAaGbh7w52soCckNC/ai-102-made-in-america
rAaGbh7w52soCckNC
AI #102: Made in America
Zvi
I remember that week I used r1 a lot, and everyone was obsessed with DeepSeek. They earned it. DeepSeek cooked, r1 is an excellent model. Seeing the Chain of Thought was revolutionary. We all learned a lot. It’s still #1 in the app store, there are still hysterical misinformed NYT op-eds and and calls for insane reacti...
2025-02-06
https://www.lesswrong.com/posts/PRZLj96uDi25opEfy/hopeful-hypothesis-the-persona-jukebox
PRZLj96uDi25opEfy
Hopeful hypothesis, the Persona Jukebox.
donald-hobson
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 Jukebox was an old machine that would pick up vinyl records and place them on a turntable to play them. So. What does the persona jukebox hypothesis say. It says that...
2025-02-14
https://www.lesswrong.com/posts/R3RWnjqQWJ9w88rGZ/don-t-go-bankrupt-don-t-go-rogue
R3RWnjqQWJ9w88rGZ
Don't go bankrupt, don't go rogue
Nathan Young
Would you bet your life’s savings on a 50% chance to triple it? Now imagine betting your sanity, values and goals. How much should you be willing to put on the table? You wanna bet? Let’s say you go to a casino, which offers a coin flip. Heads you triple it, tails you lose it. How much of your money should you bet on i...
2025-02-06
https://www.lesswrong.com/posts/oAaxcQJPAmWtkMC8p/biology-ideology-and-violence
oAaxcQJPAmWtkMC8p
Biology, Ideology and Violence
Zero Contradictions
The audio version can be listened to here: I often use the term "ideology", so I thought I should explain what I mean by it. The Wikipedia definition is: A comprehensive set of normative beliefs, conscious and unconscious ideas, that an individual, group or society has. I use the term in a more specific way. My definit...
2025-02-06
https://www.lesswrong.com/posts/rhzXQyTQBdmjPQ76z/chicanery-no
rhzXQyTQBdmjPQ76z
Chicanery: No
Screwtape
There is a concept I picked up from the Onyx Path forums, from a link that is now dead and a post I can no longer find. That concept is the Chicanery tag, and while I’ve primarily used it for tabletop RPGs (think Dungeons & Dragons) I find it applicable elsewhere. If you happen to be better at searching the forums than...
2025-02-06
https://www.lesswrong.com/posts/zjqrSKZuRLnjAniyo/illusory-safety-redteaming-deepseek-r1-and-the-strongest
zjqrSKZuRLnjAniyo
Illusory Safety: Redteaming DeepSeek R1 and the Strongest Fine-Tunable Models of OpenAI, Anthropic, and Google
ccstan99
DeepSeek-R1 has recently made waves as a state-of-the-art open-weight model, with potentially substantial improvements in model efficiency and reasoning. But like other open-weight models and leading fine-tunable proprietary models such as OpenAI’s GPT-4o, Google’s Gemini 1.5 Pro, and Anthropic’s Claude 3 Haiku, R1’s g...
2025-02-07
https://www.lesswrong.com/posts/qGKq4G3HGRcSBc94C/mats-applications-research-directions-i-m-currently-excited
qGKq4G3HGRcSBc94C
MATS Applications + Research Directions I'm Currently Excited About
neel-nanda-1
I've just opened summer MATS applications (where I'll supervise people to write mech interp papers) I'd love to get applications from any readers who are interested! Apply here, due Feb 28 As part of this, I wrote up a list of research areas I'm currently excited about, and thoughts for promising directions within thos...
2025-02-06
https://www.lesswrong.com/posts/bo4pfdAyDQmkavqFY/hypnosis-question
bo4pfdAyDQmkavqFY
hypnosis question
avery-liu
hypnosis is really crazy and weird and hypnosis is a real thing, right?if so, can I use it to motivate myself to do stuff, nudge my aliefs and System 1 dispositions closer to actual reality, be happier, etc?
2025-02-06
https://www.lesswrong.com/posts/va7oH9XfjjmHaLnZF/bida-calendar-ical-feed
va7oH9XfjjmHaLnZF
BIDA Calendar iCal Feed
jkaufman
A while ago I played around with manually generating iCal feeds for events, encoding the BIDA's regular schedule (1st and 3rd Sundays back then, with a more complex pattern for family dances and open bands) as in a machine-readable format. This was a fun project, but it didn't work very well: it didn't have any way to...
2025-02-06
https://www.lesswrong.com/posts/yaL7ZdQqA2twbiEmZ/do-no-harm-navigating-and-nudging-ai-moral-choices
yaL7ZdQqA2twbiEmZ
Do No Harm? Navigating and Nudging AI Moral Choices
sinem-erisken
TL;DR: How do AI systems make moral decisions, and can we influence their ethical judgments? We probe these questions by examining Llama's 70B (3.1 and 3.3) responses to moral dilemmas, using Goodfire API to steer its decision-making process. Our experiments reveal that simply reframing ethical questions - from "harm o...
2025-02-06
https://www.lesswrong.com/posts/NYASwJFnwZyPRE8tS/inefficiencies-in-pharmaceutical-research-practices
NYASwJFnwZyPRE8tS
Inefficiencies in Pharmaceutical Research Practices
erioire
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[1] where I work. I will try to make the distinction clear between object-level observations and speculative interpretations of the former. The process of developing a new drug or medical ...
2025-02-22
https://www.lesswrong.com/posts/jEZpfsdaX2dBD9Y6g/the-risk-of-gradual-disempowerment-from-ai
jEZpfsdaX2dBD9Y6g
The Risk of Gradual Disempowerment from AI
Zvi
The baseline scenario as AI becomes AGI becomes ASI (artificial superintelligence), if nothing more dramatic goes wrong first and even we successfully ‘solve alignment’ of AI to a given user and developer, is the ‘gradual’ disempowerment of humanity by AIs, as we voluntarily grant them more and more power in a vicious ...
2025-02-05
https://www.lesswrong.com/posts/DnWkYz3w6xxsekok9/wired-on-doge-personnel-with-admin-access-to-federal-payment
DnWkYz3w6xxsekok9
Wired on: "DOGE personnel with admin access to Federal Payment System"
Raemon
I haven't looked into this in detail, and I'm not actually sure how unique a situation this is. But, it updated me on "institutional changes to the US that might be quite bad[1]", and it seemed good if LessWrong folk did some sort of Orient Step on it. (Please generally be cautious on LessWrong talking about politics. ...
2025-02-05
https://www.lesswrong.com/posts/WHrig3emEAzMNDkyy/on-ai-scaling
WHrig3emEAzMNDkyy
On AI Scaling
harsimony
I’ve avoided talking about AI, mostly because everyone is talking about it. I think the implications of AI scaling were clear a while ago, and I mostly said my piece back then. It’s time to take another crack at it. Scaling laws Scaling laws are a wonderful innovation[1]. With large enough training runs, performance is...
2025-02-05
https://www.lesswrong.com/posts/3donrE5vFHeMJFLY9/what-makes-a-theory-of-intelligence-useful
3donrE5vFHeMJFLY9
What makes a theory of intelligence useful?
Amyr
This post is a sequel to "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 need to know that I consider an action theory to discuss choosing the best action, a policy theory to discuss the best decision-making policy when the environmen...
2025-02-20
https://www.lesswrong.com/posts/q5ihyrBWiGcRJfd7a/the-state-of-metaculus
q5ihyrBWiGcRJfd7a
The State of Metaculus
ChristianWilliams
null
2025-02-05
https://www.lesswrong.com/posts/jfskr4ZZezpsq3pTd/alignment-paradox-and-a-request-for-harsh-criticism
jfskr4ZZezpsq3pTd
Alignment Paradox and a Request for Harsh Criticism
bridgett-kay
I’m not a scientist, engineer, or alignment researcher in any respect; I’m a failed science fiction writer. I have a tendency to write opinionated essays that I rarely finish. It’s good that I rarely finish them, however, because if I did, I would generate far too much irrelevant slop. My latest opinionated essay was t...
2025-02-05