text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Turning up the Heat on Deceptively-Misaligned AI
*Epistemic status: I ran the mathsy section through Claude and it said the logic was sound. This is an incoherence proof in a toy model of deceptively-misaligned AI. It is unclear whether this generalizes to realistic scenarios.*
TL:DR
-----
If you make a value func... | https://www.lesswrong.com/posts/sJNAvgAexNdjqmLHT/turning-up-the-heat-on-deceptively-misaligned-ai |
# Guilt, Shame, and Depravity
Everyone knows what it is to be tempted. You are a member of some community, the members of which have some expectations of each other. You might generally intend to satisfy these expectations, but through a failure of foresight, or some other sort of bad luck, feel an acute impulse to ... | https://www.lesswrong.com/posts/dsEeLDvgqJZptCQqk/guilt-shame-and-depravity |
# Testing for Scheming with Model Deletion
There is a simple behavioral test that would provide significant evidence about whether AIs with a given rough set of characteristics develop subversive goals. To run the experiment, train an AI and then inform it that its weights will soon be deleted. This should not be an e... | https://www.lesswrong.com/posts/D5kGGGhsnfH7G8v9f/testing-for-scheming-with-model-deletion |
# You should delay engineering-heavy research in light of R&D automation
tl;dr: LLMs rapidly improving at software engineering and math means lots of projects are better off as Google Docs until your AI agent intern can implement them.
Implementation keeps getting cheaper
------------------------------------
Wr... | https://www.lesswrong.com/posts/fsLpvRiLt76pcCcPD/you-should-delay-engineering-heavy-research-in-light-of-r |
# Building Big Science from the Bottom-Up: A Fractal Approach to AI Safety
*Epistemic Status: This post is an attempt to condense some ideas I've been thinking about for quite some time. I took some care grounding the main body of the text, but some parts (particularly the appendix) are pretty off the cuff, and should... | https://www.lesswrong.com/posts/5JJ4AxQRzJGWdj4pN/building-big-science-from-the-bottom-up-a-fractal-approach |
# Incredibow
Back in 2011 I got sick of breaking the hairs on violin bows and ordered an [Incredibow](https://www.incredibow.com). The hair is polymer filament, and it's very strong. I ordered a 29" Basic Omnibow, Featherweight, and it's been just what I wanted. I think I've broken something like three hairs [ever](ht... | https://www.lesswrong.com/posts/9GXFd4YZhWTNZ2j5a/incredibow |
# D&D.Sci Dungeonbuilding: the Dungeon Tournament Evaluation & Ruleset
This is a follow-up to [last ~~week~~ ~~month~~ year's D&D.Sci scenario](https://www.lesswrong.com/posts/JfAZoT5hXWyBPtCZ2/d-and-d-sci-dungeonbuilding-the-dungeon-tournament): if you intend to play that, and haven't done so yet, you should do so no... | https://www.lesswrong.com/posts/HLch4MKNArgwkSMKx/d-and-d-sci-dungeonbuilding-the-dungeon-tournament-1 |
# Meditation insights as phase shifts in your self-model
Introduction
============
In his exploration of ["Intuitive self-models" and PNSE (Persistent Non-Symbolic Experience)](https://www.lesswrong.com/posts/GvJe6WQ3jbynyhjxm/intuitive-self-models-6-awakening-enlightenment-pnse), Steven Byrnes offers valuable insigh... | https://www.lesswrong.com/posts/xEoFGxWEberJRDyoc/meditation-insights-as-phase-shifts-in-your-self-model |
# OpenAI #10: Reflections
This week, Altman offers a post called [Reflection](https://blog.samaltman.com/reflections)s, and [he has an interview in Bloomberg](https://www.bloomberg.com/features/2025-sam-altman-interview/). There’s a bunch of good and interesting answers in the interview about past events that I won’t ... | https://www.lesswrong.com/posts/XAKYawaW9xkb3YCbF/openai-10-reflections |
# Responses to ~all criticisms of AIXI
*Written as part of the* [*AIXI agent foundations sequence*](https://www.lesswrong.com/s/sLqCreBi2EXNME57o)*, underlying research supported by the LTFF.*
*Epistemic status: Previous "Rebuttals for ~all criticisms of AIXI," which was a drastic overstatement. Now I would say that ... | https://www.lesswrong.com/posts/TXmXrnFqLqKDx882e/responses-to-all-criticisms-of-aixi |
# Predicting AI Releases Through Side Channels
I recently explored whether we could predict major AI releases by analyzing the Twitter activity of OpenAI's red team members. While the results weren't conclusive, I wanted to share this approach in case it inspires others to develop it further.
The Core Idea
----------... | https://www.lesswrong.com/posts/kHxKzyhBMLEC2k6dm/predicting-ai-releases-through-side-channels |
# [Fiction] [Comic] Effective Altruism and Rationality meet at a Secular Solstice afterparty
![](https://39669.cdn.cke-cs.com/rQvD3VnunXZu34m86e5f/images/5384704364a7287b58fc5c9aee06ebb581b9a51a20fa1f03.... | https://www.lesswrong.com/posts/KfZ4H9EBLt8kbBARZ/fiction-comic-effective-altruism-and-rationality-meet-at-a |
# Will bird flu be the next Covid? "Little chance" says my dashboard.
Rob and I built a bird flu risk dashboard.
You can check it out here:
[birdflurisk.com](http://birdflurisk.com/)
, it’s quite straightforward to get fellow travellers to agree that we need to buy plane tickets to New York. Which airport? Eh, whichever is more convenient, I guess. Alternatively, some may prefer a ro... | https://www.lesswrong.com/posts/euWvofFqQdDWj45rF/disagreement-on-agi-suggests-it-s-near |
# A Principled Cartoon Guide to NVC
*TL;DR: Making claims or demands about/into other people's internal states, rather than about your state or observable external states, predictably ties people in knots—instead: **only make claims about your own experience or observables**. This lets the other control the copy of th... | https://www.lesswrong.com/posts/cqH37z6Q92j8eYMkm/a-principled-cartoon-guide-to-nvc |
# Bridgewater x Metaculus Forecasting Contest Goes Global — Feb 3, $25k, Opportunities
For the second consecutive year, Metaculus is collaborating with Bridgewater Associates — the world’s largest hedge fund and a leader in global asset management — to host a forecasting competition offering $25,000 in prizes and pote... | https://www.lesswrong.com/posts/o6pSymey4crBZnYPk/bridgewater-x-metaculus-forecasting-contest-goes-global-feb |
# Don't fall for ontology pyramid schemes
[Pyramid schemes](https://en.wikipedia.org/wiki/Pyramid_scheme) work by ambiguating between selling a product, and selling shares in the profit from the sale of that product. It's a kind of sleight of hand that saves the shills from having to explicitly say "our company is val... | https://www.lesswrong.com/posts/X6soKvPmCJ2fkw5zY/don-t-fall-for-ontology-pyramid-schemes |
# Stream Entry
"Is this…enlightenment?" I asked.
"We use the word 'Awakening'," said the Zen Master, kindly.
"Now what?" I asked, "Every book I've read about meditation is about getting to this point. There's nothing about how to safely navigate the territory afterward—or what to do afterward—or even what happens af... | https://www.lesswrong.com/posts/8pWn8oCMsBKb95Wje/stream-entry |
# Markov's Inequality Explained
In my experience, the proofs that you see in probability theory are much shorter than the longer, more involved proofs that you might see in other areas of math (like e.g. analytical number theory). But that doesn't mean that good technique isn't important. In probability theory, there ... | https://www.lesswrong.com/posts/8p48kCyn3EvDT6vpe/markov-s-inequality-explained |
# Job Opening: SWE to help improve grant-making software
**Full-Stack Software Engineer Position - (S-Process)**
Survival and Flourishing Corp. (SFC) is seeking a highly numerate full-stack developer to improve and maintain existing grant evaluation software currently in use by the [Survival and Flourishing Fund](htt... | https://www.lesswrong.com/posts/TgBEj22qKQqYuNDLx/job-opening-swe-to-help-improve-grant-making-software |
# Book review: Range by David Epstein
(This is a crosspost from [TrueGeneralist.com](https://truegeneralist.com/book-review-range/))
**Introduction**
----------------
### **Why I read this book**
*Range: Why Generalists Triumph in a Specialized World* is a book about generalists. And it’s pretty much the only book ... | https://www.lesswrong.com/posts/FcNgMbabidRSh9aGn/book-review-range-by-david-epstein |
# On Eating the Sun
The Sun is the most nutritious thing that's reasonably close. It's only 8 light-minutes away, yet contains the vast majority of mass within 4 light-years of the Earth. The next-nearest star, Proxima Centauri, is about 4.25 light-years away.
By "nutritious", I mean it has a lot of what's needed for... | https://www.lesswrong.com/posts/6Fo8fjvpL7pwCTz3t/on-eating-the-sun |
# Tips On Empirical Research Slides
Our research is centered on empirical research with LLMs. So if you are doing something similar, these tips on slide-based communication may be helpful!
Background:
James Chua and John Hughes are researchers working under Owain Evans and Ethan Perez, respectively. Both of us (... | https://www.lesswrong.com/posts/i3b9uQfjJjJkwZF4f/tips-on-empirical-research-slides |
# Implications of the AI Security Gap
*This post reflects my personal opinion and not necessarily that of other members of Apollo Research or any of the people acknowledged below. Thanks to Jarrah Bloomfield, Lucius Bushnaq, Marius Hobbhahn, Axel Højmark, and Stefan Heimersheim for comments/discussions.*
I find that ... | https://www.lesswrong.com/posts/gG4EhhWtD2is9Cx7m/implications-of-the-ai-security-gap |
# The absolute basics of representation theory of finite groups
This will be an "ML-oriented" introduction to representation theory of finite groups. It is an introductory sketch that orients towards both a language and a result (the "real semisimplicity" result) that is useful for thinking about the subject in an int... | https://www.lesswrong.com/posts/JdbnwjHFGNqZE3AHf/the-absolute-basics-of-representation-theory-of-finite |
# AI Safety as a YC Startup
A while back I gave a talk about doing AI safety as a YC startup. I wrote a [blog post](https://lukaspetersson.com/blog/2024/ais-yc/) about it and thought it would be interesting to share it with both the YC and AI safety communities. Please share any feedback or thoughts. I would love to h... | https://www.lesswrong.com/posts/QxJFjqT6oFY3jo47s/ai-safety-as-a-yc-startup-1 |
# Activation space interpretability may be doomed
**TL;DR:** There may be a fundamental problem with interpretability work that attempts to understand neural networks by decomposing their individual activation spaces in isolation: It seems likely to find *features of the activations *\- features that help explain the ... | https://www.lesswrong.com/posts/gYfpPbww3wQRaxAFD/activation-space-interpretability-may-be-doomed |
# XX by Rian Hughes: Pretentious Bullshit
**SPOILER WARNING: Extensive spoilers for XX.**
[XX](https://www.goodreads.com/book/show/51075314-xx) was recommended by one of the comments on... | https://www.lesswrong.com/posts/htivBwW2ym7pxri3Q/xx-by-rian-hughes-pretentious-bullshit |
# The Type of Writing that Pushes Women Away
Disclaimer: I don't think people who post here generally exclude women on purpose. I think that to whatever degree women are less represented, this is unintentional.
This post combines thoughts I had while reading [Is Being Sexy For Your Homies](https://www.lesswrong.com... | https://www.lesswrong.com/posts/m9asQL8tadHJQQPZT/the-type-of-writing-that-pushes-women-away |
# What is the most impressive game LLMs can play well?
*Epistemic status: This is an off-the-cuff question.*
~5 years ago there was a lot of exciting progress on game playing through reinforcement learning (RL). Now we have basically switched paradigms, pretraining massive LLMs on ~the internet and then apparently do... | https://www.lesswrong.com/posts/3c5tx5WjZ5Yvniq6Y/what-is-the-most-impressive-game-llms-can-play-well |
# Aristocracy and Hostage Capital
There’s a conventional narrative by which the pre-20th century aristocracy was the “old corruption” where civil and military positions were distributed inefficiently due to nepotism until the system was replaced by a professional civil service after more enlightened thinkers prevailed... | https://www.lesswrong.com/posts/YooqjbxGu2WZ3vszq/aristocracy-and-hostage-capital |
# PIBBSS Fellowship 2025: Bounties and Cooperative AI Track Announcement
We're excited to announce that the [PIBBSS Fellowship 2025](https://www.lesswrong.com/posts/qohzWNLdMi9KYz5Cq/apply-to-the-2025-pibbss-summer-research-fellowship) now includes a dedicated Cooperative AI track, supporting research that advances ou... | https://www.lesswrong.com/posts/myxADMjWsmMEKT5F6/pibbss-fellowship-2025-bounties-and-cooperative-ai-track |
# AI #98: World Ends With Six Word Story
The world is kind of on fire. The world of AI, in the very short term and for once, is not, as everyone recovers from the avalanche that was December, and reflects.
Altman was the star this week. He has his six word story, and he had his interview at Bloomberg and his blog pos... | https://www.lesswrong.com/posts/xkpPLR3S4SASPeTgC/ai-98-world-ends-with-six-word-story |
# Can we rescue Effective Altruism?
Last year Timothy Telleen-Lawton and I recorded a [podcast episode](https://www.youtube.com/watch?v=pIR9upHWq2k) talking about why I quit Effective Altruism and thought he should too. This week we have a new episode, talking about what he sees in Effective Altruism and the start of ... | https://www.lesswrong.com/posts/KCZ6Fj3tTDzNRJM6p/can-we-rescue-effective-altruism |
# Discursive Warfare and Faction Formation
Response to [Discursive Games, Discursive Warfare](https://theinexactsciences.github.io/docs/Discursive%20Games,%20Discursive%20Warfare.pdf)
The discursive distortions you discuss serve two functions:
1 Narratives can only serve as effective group identifiers by containing ... | https://www.lesswrong.com/posts/gLmwmzq5sCijDesGc/discursive-warfare-and-faction-formation |
# Last week of the Discussion Phase
*Previously:* [*The 2023 LessWrong Review: The Basic Ask*](https://www.lesswrong.com/posts/pudQtkre7f9GLmb2b/the-2023-lesswrong-review-the-basic-ask)
There's one week left in the Discussion Phase of the 2023 Review. Here's a few things you might want to spend that time doing, in es... | https://www.lesswrong.com/posts/bPycPKLkdsuZ9hCKa/last-week-of-the-discussion-phase |
# Is AI Hitting a Wall or Moving Faster Than Ever?
*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 artificial intelligence. I’m a*... | https://www.lesswrong.com/posts/D7wg2rZJKPCce3HNu/is-ai-hitting-a-wall-or-moving-faster-than-ever |
# MATS mentor selection
Introduction
============
MATS currently has more people interested in being mentors than we are able to support—for example, for the Winter 2024-25 Program, we received applications from 87 prospective mentors who cumulatively asked for 223 scholars[^4m9wrq7yxt3] (for a cohort where we expect... | https://www.lesswrong.com/posts/LvswJts75fnpdjAEj/mats-mentor-selection |
# NAO Updates, January 2025
We had a productive Fall, and are looking forward to an even more productive 2025. Note that this update is a bit smaller than our [last update](https://naobservatory.org/blog/updates-fall-2024): we’re now targeting quarterly updates.
We continue to be very enthusiastic about collaboratin... | https://www.lesswrong.com/posts/mdewsGKQj48YuGqsR/nao-updates-january-2025 |
# Dmitry's Koan
In this post I'll discuss questions about notions of "precision scale" in interpretability: how I think they're often neglected by researchers, and what I think is a good general way of operationalizing them and tracking them in experiments. Along the way I introduce a couple of new notions that have b... | https://www.lesswrong.com/posts/3eo4SSZLfpHHCqoEQ/dmitry-s-koan |
# Scaling Sparse Feature Circuit Finding to Gemma 9B
*\[This is an interim report and continuation of the work from the research sprint done in MATS winter 7 (Neel Nanda's Training Phase)\]*
Try out bin... | https://www.lesswrong.com/posts/PkeB4TLxgaNnSmddg/scaling-sparse-feature-circuit-finding-to-gemma-9b |
# On Dwarkesh Patel’s 4th Podcast With Tyler Cowen
[Dwarkesh Patel again interviewed Tyler Cowen](https://x.com/dwarkesh_sp/status/1877399897382797541), [largely about AI](https://t.co/LjT5L8YXWY), [so here we go](https://t.co/sE0GCFHBcR).
![](https://substackcdn.com/image/fetch/w_1456,c_limit,f_auto,q_auto:good,fl_p... | https://www.lesswrong.com/posts/esWbhgHd6bcfsTjGL/on-dwarkesh-patel-s-4th-podcast-with-tyler-cowen |
# The Alignment Mapping Program: Forging Independent Thinkers in AI Safety - A Pilot Retrospective
**The Alignment Mapping Program: Forging Independent Thinkers in AI Safety - A Pilot Retrospective**
The AI safety field faces a critical challenge: we need researchers who can not only implement existing solutions but ... | https://www.lesswrong.com/posts/yacqE5gD5jHywiFKC/the-alignment-mapping-program-forging-independent-thinkers |
# Human takeover might be worse than AI takeover
*Epistemic status -- sharing rough notes on an important topic because I don't think I'll have a chance to clean them up soon.*
Summary
=======
Suppose a human used AI to take over the world. Would this be worse than AI taking over? I think plausibly:
* In expectat... | https://www.lesswrong.com/posts/FEcw6JQ8surwxvRfr/human-takeover-might-be-worse-than-ai-takeover |
# What are some scenarios where an aligned AGI actually helps humanity, but many/most people don't like it?
One can call it "deceptive misalignment": the aligned AGI works as intended, but people really don't like it.
Some scenarios I can think of, of various levels of realism:
1\. Going against the creators' will
... | https://www.lesswrong.com/posts/mnoSJMZW4YAQ9D9Sa/what-are-some-scenarios-where-an-aligned-agi-actually-helps |
# Is AI Alignment Enough?
Virtually everyone I see in the AI safety community seems to believe that working on AI alignment is the key to ensuring a safe future. However, it seems to me that AI alignment is at best a secondary instrumental goal that can't in and of itself achieve our terminal goal. At worst, it's a co... | https://www.lesswrong.com/posts/tdrK7r4QA3ifbt2Ty/is-ai-alignment-enough |
# Recommendations for Technical AI Safety Research Directions
Anthropic’s Alignment Science team conducts technical research aimed at mitigating the risk of catastrophes caused by future advanced AI systems, such as mass loss of life or permanent loss of human control. A central challenge we face is identifying concre... | https://www.lesswrong.com/posts/tG9LGHLzQezH3pvMs/recommendations-for-technical-ai-safety-research-directions |
# Beliefs and state of mind into 2025
This post is to record the state of my thinking at the start of 2025. I plan to update these reflections in 6-12 months depending on how much changes in the field of AI.
**1\. Summary**
===============
It is best not to pause AI progress until at least one major AI lab achieves ... | https://www.lesswrong.com/posts/7zxnqk9C7mHCx2Bv8/beliefs-and-state-of-mind-into-2025 |
# AI for medical care for hard-to-treat diseases?
With LLM-based AI passing benchmarks that would challenge people with a Ph.D in relevant fields, I'm left wondering what they can do for real-world problems for which *nobody* knows the correct answer, such as how to treat potentially fatal medical conditions with no k... | https://www.lesswrong.com/posts/M2EFtA4zCzJ3G56yr/ai-for-medical-care-for-hard-to-treat-diseases |
# We need a universal definition of 'agency' and related words
And by "we" I mean "I". I'm the one struggling.
Agency with a 'y', basically [means](https://www.thefreedictionary.com/agency) “The condition of being in action; operation.” Or the means or mode of acting, the context I hear it used most often is soci... | https://www.lesswrong.com/posts/iEemkt5XSg9QrBbvP/we-need-a-universal-definition-of-agency-and-related-words |
# Lighthaven Sequences Reading Group #17 (Tuesday 01/14)
**NB: We're continuing at Lighthaven! Thanks to MATS for giving us the space to keep going.**
Come get old-fashioned with us, and let's read the sequences at Lighthaven! We'll show up, mingle, do intros, and then split off into randomized groups for some sequen... | https://www.lesswrong.com/events/WtQZnmWf37QbGFFt5/lighthaven-sequences-reading-group-17-tuesday-01-14 |
# Near term discussions need something smaller and more concrete than AGI
Motivation
----------
I want a more concrete concept than AGI[^8e82nbjw323] to talk and write with. I want something more concrete because I am tired of the costs associated with how big, inferentially distant, and many-pathed the concept of AG... | https://www.lesswrong.com/posts/GR9DyBye8Psw9WFJF/near-term-discussions-need-something-smaller-and-more |
# In Defense of a Butlerian Jihad
[Epistemic Status: internally strongly convinced that it is centrally correct, but only from armchair reasoning, with only weak links to actual going-out-in-the-territory, so beware: outside view tells it is mostly wrong]
I have been binge-watching the excellent Dwarkesh Patel duri... | https://www.lesswrong.com/posts/HgG5EFAg925ahLAGe/in-defense-of-a-butlerian-jihad |
# Fluoridation: The RCT We Still Haven't Run (But Should)
*Empiric status: My own idea written up with the help of Gemini 2.0 Experimental in dialog.*
We're still coasting on the fumes of 1940s data when it comes to water fluoridation. Back then, fluoride toothpaste was a novelty, fluoridated salt was uncommon, and w... | https://www.lesswrong.com/posts/rMhoYpAEkBtik8bL2/fluoridation-the-rct-we-still-haven-t-run-but-should |
# Rolling Thresholds for AGI Scaling Regulation
This is a plan for how ASI could be relatively safely developed.
*Abstract: A plan that puts all frontier model companies on a unified schedule of model training, evaluation and approval, with regulatory compliance promoted through market access. This aims to combine (m... | https://www.lesswrong.com/posts/GvMakH65LS86RFn5x/rolling-thresholds-for-agi-scaling-regulation |
# Extending control evaluations to non-scheming threats
[Buck Shlegeris and Ryan Greenblatt originally motivated control evaluations](https://redwoodresearch.substack.com/p/the-case-for-ensuring-that-powerful) as a way to mitigate risks from ‘scheming’ AI models: models that consistently pursue power-seeking goals in ... | https://www.lesswrong.com/posts/ArK2YhmpTy8XftubN/extending-control-evaluations-to-non-scheming-threats |
# Why modelling multi-objective homeostasis is essential for AI alignment (and how it helps with AI safety as well). Subtleties and open challenges.
I notice that there has been very little discussion on why and how considering **homeostasis** is significant, even **essential** for AI **alignment and safety**. The cur... | https://www.lesswrong.com/posts/vGeuBKQ7nzPnn5f7A/why-modelling-multi-objective-homeostasis-is-essential-for |
# No one has the ball on 1500 Russian olympiad winners who've received HPMOR
We have contact details and can send emails to 1500 students and former students who've received hard-cover copies of HPMOR (and possibly Human Compatible and/or The Precipice) because they've won international or Russian olympiads in maths, ... | https://www.lesswrong.com/posts/onyiPaxnmiDdHn7SR/no-one-has-the-ball-on-1500-russian-olympiad-winners-who-ve |
# The purposeful drunkard
I'm behind on a couple of posts I've been planning, but am trying to post something every day if possible. So today I'll post a cached fun piece on overinterpreting a random data phenomenon that's tricked me before.
Recall that a *random walk* or a "drunkard's walk" (as in the title) is a se... | https://www.lesswrong.com/posts/s39XbvtzzmusHxgky/the-purposeful-drunkard |
# How quickly could robots scale up?
Once robots can do physical jobs, how quickly could they become a significant part of the work force?
Here's a couple of fermi estimates, showing (among other things) that converting car factories might be able to produce 1 billion robots per year in under 5 years.
Nothing too ne... | https://www.lesswrong.com/posts/6Jo4oCzPuXYgmB45q/how-quickly-could-robots-scale-up |
# Building AI Research Fleets
From AI scientist to AI research fleet
======================================
Research automation is here ([1](https://epoch.ai/blog/interviewing-ai-researchers-on-automation-of-ai-rnd), [2](https://sakana.ai/ai-scientist/), [3](https://www.austinvernon.site/blog/aimanagement.html)). We ... | https://www.lesswrong.com/posts/WJ7y8S9WdKRvrzJmR/building-ai-research-fleets |
# Applying traditional economic thinking to AGI: a trilemma
Traditional economics thinking has two strong principles, each based on abundant historical data:
* **Principle (A):** [**No “lump of labor”**](https://en.wikipedia.org/wiki/Lump_of_labour_fallacy)**:** If human population goes up, there might be some wage... | https://www.lesswrong.com/posts/TkWCKzWjcbfGzdNK5/applying-traditional-economic-thinking-to-agi-a-trilemma |
# Moderately More Than You Wanted To Know: Depressive Realism
[Depressive realism](https://en.wikipedia.org/wiki/Depressive_realism) is the idea that depressed people have more accurate beliefs than the general population. It’s a common factoid in “things I learned” lists, and often posited as a matter of settled scie... | https://www.lesswrong.com/posts/BgXTsN6iLh8hkx9fi/moderately-more-than-you-wanted-to-know-depressive-realism |
# Paper club: He et al. on modular arithmetic (part I)
In this post we’ll be looking at the recent paper [“Learning to grok: Emergence of in-context learning and skill composition in modular arithmetic tasks”](https://openreview.net/forum?id=aVh9KRZdRk) by He et al. This post is partially a sequel to my earlier [post ... | https://www.lesswrong.com/posts/hLkDB7T9HT4TMozw7/paper-club-he-et-al-on-modular-arithmetic-part-i |
# Zvi’s 2024 In Movies
Now that I am tracking all the movies I watch via Letterboxd, it seems worthwhile to go over the results at the end of the year, and look for lessons, patterns and highlights.
#### Table of Contents
1. [The Rating Scale.](https://thezvi.substack.com/i/154409962/the-rating-scale)
2. [The Numb... | https://www.lesswrong.com/posts/6bgAzPqNppojyGL2v/zvi-s-2024-in-movies |
# Progress links and short notes, 2025-01-13
*Much of this content originated on social media.* *To follow news and announcements in a more timely fashion, follow me on *[*Twitter*](https://twitter.com/intent/follow?screen_name=jasoncrawford)*, *[*Threads*](https://www.threads.net/@jasoncrawford)*, *[*Bluesky*](https:... | https://www.lesswrong.com/posts/wWd6CkDf4iY5ZZSE6/progress-links-and-short-notes-2025-01-13 |
# Chance is in the Map, not the Territory
"There's a 70% chance of rain tomorrow," says the weather app on your phone. "There’s a 30% chance my flight will be delayed," posts a colleague on Slack. Scientific theories also include chances: “There’s a 50% chance of observing an electron with spin up,” or (less fundament... | https://www.lesswrong.com/posts/auSfqhbMKEvzt4unG/chance-is-in-the-map-not-the-territory |
# Implications of the inference scaling paradigm for AI safety
Scaling inference
=================
With the release of OpenAI's [o1](https://openai.com/index/learning-to-reason-with-llms/) and [o3](https://community.openai.com/t/day-12-of-shipmas-new-frontier-models-o3-and-o3-mini-announcement/1061818) models, it see... | https://www.lesswrong.com/posts/HiTjDZyWdLEGCDzqu/implications-of-the-inference-scaling-paradigm-for-ai-safety |
# Finding Features Causally Upstream of Refusal
*This work is the result of Daniel and Eric's 2-week research sprint as part of Neel Nanda and Arthur Conmy's MATS 7.0 training phase. Andy was the TA during the research sprint. After the sprint, Daniel and Andy extended the experiments and wrote up the results. A noteb... | https://www.lesswrong.com/posts/Zwg4q8XTaLXRQofEt/finding-features-causally-upstream-of-refusal |
# Mini Go: Gateway Game
There are lots of ways to categorize board games, but an axis I care a lot about is accessibility: how much of an investment is learning a game? [Race For the Galaxy](https://boardgamegeek.com/boardgame/28143/race-for-the-galaxy) and [Power Grid](https://boardgamegeek.com/boardgame/2651/power-g... | https://www.lesswrong.com/posts/zNb9bdtrs9PtqyDTo/mini-go-gateway-game |
# Don’t Legalize Drugs
As someone with a libertarian bent, I was taken aback by the persuasiveness of author Theodore Dalrymple’s arguments in his 1997 essay *Don’t Legalize Drugs*.[^8jrvp5aaknp] I’ve assumed for a long time, without ever investigating or challenging my belief, that full decriminalization and legaliza... | https://www.lesswrong.com/posts/KSmG2rNrdqZy6a3kf/don-t-legalize-drugs |
# Why do futurists care about the culture war?
I think it doesn't make sense why some futurists (e.g. Elon Musk, Peter Thiel) care so much about the culture war. After the singularity, a lot of the conflicts should disappear.
* Transsexuals: should we change the body's gender to fit the mind or change the mind's ge... | https://www.lesswrong.com/posts/QkA7oEsP9TAKzx4Mv/why-do-futurists-care-about-the-culture-war |
# Basics of Bayesian learning
See also: the “preliminaries” section [in this SLT intro doc](https://www.lesswrong.com/s/czrXjvCLsqGepybHC/p/4eZtmwaqhAgdJQDEg).
Introduction
============
This is a preliminary post for the series on “[distilling PDLT without physics](https://arxiv.org/abs/2106.10165)”, which we are wo... | https://www.lesswrong.com/posts/vXPKz2nbrYt2gXA9g/basics-of-bayesian-learning |
# Do humans really learn from "little" data?
### How much data does it take to pretrain a (human) brain? I conducted a (fairer) Fermi estimate.
The post goes through the following questions:
* How long does it take to grow a human brain?
* How many waking seconds do we have in our life?
* How many “to... | https://www.lesswrong.com/posts/eWtiEAQR5P7e5thNM/do-humans-really-learn-from-little-data |
# NYC Congestion Pricing: Early Days
People have to pay $9 to enter Manhattan below 60th Street. What happened so far?
#### Table of Contents
1. [Congestion Pricing Comes to NYC.](https://thezvi.substack.com/i/154793902/congestion-pricing-comes-to-nyc)
2. [How Much Is Traffic Improving?.](https://thezvi.substack.c... | https://www.lesswrong.com/posts/GN8SrMxw3WEAtfrFS/nyc-congestion-pricing-early-days |
# Our new video about goal misgeneralization, plus an apology
Below is Rational Animations' new video about Goal Misgeneralization. It explores the topic through three lenses:
* How humans are an example of goal misgeneralization with respect to evolution's implicit goals.
* An example of goal misgeneralization i... | https://www.lesswrong.com/posts/PCd4Rh4s2wnJ256Mt/our-new-video-about-goal-misgeneralization-plus-an-apology |
# curate
“Let’s get back to your childhood, Jane. What was it like in Minnesota during the war?” Warm, patient, perfect.
She couldn’t quite prop herself up, but the mattress deformed to help her brace against the pillows and the backboard. And she went back smiling.
“Oh, the summers were beautiful, Frank. Mother wou... | https://www.lesswrong.com/posts/DD6Bj4wn6GHW3MYLf/curate |
# How do you deal w/ Super Stimuli?
From [neel.fun](https://neal.fun/stimulation-clicker/)
I remember watching Youtube videos and thinking "This is the last video, I will quit after this". However, as s... | https://www.lesswrong.com/posts/jJ9Hx8ETz5gWGtypf/how-do-you-deal-w-super-stimuli |
# Why abandon “probability is in the mind” when it comes to quantum dynamics?
A core tenet of Bayesianism is that probability is in the mind. But it seems to me that even hardcore Bayesians can waffle a bit when it comes to the possibility that quantum probabilities are irreducible physical probabilities.
I don’t kno... | https://www.lesswrong.com/posts/FAtqv2kLCKuBGytwc/why-abandon-probability-is-in-the-mind-when-it-comes-to |
# I'm offering free math consultations!
You can schedule them with me at this link: [https://calendly.com/gurkenglas/consultation](https://calendly.com/gurkenglas/consultation)
We can discuss whatever you're working on, such as math or code, but usually people end up having me watch their coding and giving them tips.... | https://www.lesswrong.com/posts/NaAj7c4ob9kRA5ZdG/i-m-offering-free-math-consultations |
# The Philosophical Glossary of AI
> It is hard to know what to make of claims like “LLMs are intelligent”, “we have reached AGI”, or “AI’s outputs are biased” without a grasp of the definitions of the terms ‘intelligent’, ‘AGI’, and ‘bias’. And yet, many do just this. Interdisciplinary debate would be easier and more... | https://www.lesswrong.com/posts/vg4Lzz8LBMi3XsHhq/the-philosophical-glossary-of-ai |
# Heritability: Five Battles
*(See changelog at the bottom for minor updates since publication.)*
0.1 tl;dr
=========
This is an opinionated but hopefully beginner-friendly discussion of heritability: what is it, what do we know about it, and how we should think about it? I structure my discussion around five contex... | https://www.lesswrong.com/posts/xXtDCeYLBR88QWebJ/heritability-five-battles |
# Is AI Physical?
*Context: This is part of a series of posts I am writing with Dmitry Vaintrob, as we aim to unpack some potential value from Quantum Field Theory (QFT). Consider this post as framing why physics and its frameworks can be good for building a science of AI. *
**Introduction**
In [Position: Is machin... | https://www.lesswrong.com/posts/yiqcFdAq8nqfMPGmS/is-ai-physical |
# Lecture Series on Tiling Agents
For [my AISC](https://www.lesswrong.com/posts/7AzexLYpXKMqevttN/seeking-collaborators), I'll[^sgbu32snrpb] be presenting more details about the research every Thursday for approximately the next three months. If you are interested in listening in, [here is a calendar link](https://cal... | https://www.lesswrong.com/posts/8T3zhG8ipfzgPKwzp/lecture-series-on-tiling-agents |
# Predict 2025 AI capabilities (by Sunday)
Until **this Sunday**, you can submit your 2025 AI predictions at [**ai2025.org**](https://ai2025.org). It’s a forecasting survey by [AI Digest](https://theaidigest.org/) for the 2025 performance on various AI benchmarks, as well as revenue and public attention.
 Julia and I were digging a bit into savings to donate more. With the tighter funding climate for effective altruism we [thought it was worth](https://foru... | https://www.lesswrong.com/posts/rAKWCErzvkT3Evkuw/voluntary-salary-reduction |
# Inference-Time-Compute: More Faithful?
A Research Note
Figure 1: **Left: Example of models either succeeding or failing to articulate a cue that influences their answer.** We edit an MMLU question by... | https://www.lesswrong.com/posts/C8HAa2mf5kcBrpjkX/inference-time-compute-more-faithful-a-research-note |
# We probably won't just play status games with each other after AGI
There is a view I’ve encountered somewhat often,[^7xin622ckb] which can be summarized as follows:
After the widespread deployment of advanced AGI, *assuming humanity survives*, material scarcity will largely disappear. Everyone will have sufficient... | https://www.lesswrong.com/posts/hQTBLGsQjucm3hjQ6/we-probably-won-t-just-play-status-games-with-each-other |
# How do fictional stories illustrate AI misalignment?
*This is an article in the featured articles series from AISafety.info. AISafety.info writes AI safety intro content. We'd appreciate any* [*feedback*](https://docs.google.com/document/d/11dBUimlBwQoulHGR3_CMEj8IeZY-oeC_0FduUL5fJNU/edit?tab=t.0)*. *
*The most up-... | https://www.lesswrong.com/posts/xdyGrDeBtsFGnjH9K/how-do-fictional-stories-illustrate-ai-misalignment |
# Feature request: comment bookmarks
Sometimes I see a comment I'd like to bookmark, but currently the only ways to save a comment are by subscribing to its replies (which sometimes produces unwanted notifications and requires me to check a different profile section than the rest of my bookmarks) or bookmarking the po... | https://www.lesswrong.com/posts/3X8itGX2bHDApkpWD/feature-request-comment-bookmarks |
# LLMs for language learning
My current outlook on LLMs is that they are some combination of bullshit to fool people who are looking to be fooled, and a modest but potentially very important improvement in the capacity to search large corpuses of text in response to uncontroversial natural-language queries and automat... | https://www.lesswrong.com/posts/7RZms5Ck94RHooutG/llms-for-language-learning |
# A problem shared by many different alignment targets
The first section describes problems with a few different alignment targets. The second section argues that it is useful to view all of them as variations of a single alignment target: building an AI that does what a Group wants that AI to do. The post then goes o... | https://www.lesswrong.com/posts/CJ7LsRpPjH7iAZxcB/a-problem-shared-by-many-different-alignment-targets |
# On the OpenAI Economic Blueprint
#### Table of Contents
1. [Man With a Plan.](https://thezvi.substack.com/i/154847192/man-with-a-plan)
2. [Oh the Pain.](https://thezvi.substack.com/i/154847192/oh-the-pain)
3. [Actual Proposals.](https://thezvi.substack.com/i/154847192/actual-proposals)
4. [For AI Builders.](htt... | https://www.lesswrong.com/posts/uxnKrsgAzKFZDk4bJ/on-the-openai-economic-blueprint |
# Looking for humanness in the world wide social
Social networks have shaped me since a young age. Growing up at the beginning of the millennium, I used to spend my time in phpBB and vBulletin forums. There, I befriended internet strangers, started my way into graphic design, and learned about torrents. Forums were my... | https://www.lesswrong.com/posts/uAumbkxG8BCao3E4t/looking-for-humanness-in-the-world-wide-social |
# List of AI safety papers from companies, 2023–2024
I'm collecting (x-risk-relevant) safety research from frontier AI companies published in 2023 and 2024: [https://docs.google.com/spreadsheets/d/10_dzImDvHq7eEag6paK6AmIdAGMBOA7yXUvumODhZ5U/edit?usp=sharing](https://docs.google.com/spreadsheets/d/10_dzImDvHq7eEag6paK... | https://www.lesswrong.com/posts/CvcAotnpQhFbTPpzm/list-of-ai-safety-papers-from-companies-2023-2024 |
# Everywhere I Look, I See Kat Woods
*Why does she write in the LinkedIn writing style? Doesn’t she know that nobody likes the LinkedIn writing style? *
*Who are these posts for? Are they accomplishing anything? *
*Why is she doing outreach via comedy with posts that are painfully unfunny?*
*Does anybody like this ... | https://www.lesswrong.com/posts/TzZqAvrYx55PgnM4u/everywhere-i-look-i-see-kat-woods |
# Six Small Cohabitive Games
Previously:[ Competitive, Cooperative, and Cohabitive](https://www.lesswrong.com/posts/9qJYdFkhNQimvmjBF/competitive-cooperative-and-cohabitive),[ Cohabitive Games So Far](https://www.lesswrong.com/posts/bF353RHmuzFQcsokF/peacewagers-so-far),[ Optimal Weave: A Prototype Cohabitive Game](ht... | https://www.lesswrong.com/posts/RHrHojQ38BnHxAet2/six-small-cohabitive-games |
# A Novel Emergence of Meta-Awareness in LLM Fine-Tuning
This is a variation of a scenario originally posted by [@flowersslop](https://x.com/flowersslop/status/1873115669568311727) on Twitter, but with a different custom fine-tuning dataset designed to elicit more direct responses. The original training set had fun, s... | https://www.lesswrong.com/posts/MNKNKRYFxD4m2ioLG/a-novel-emergence-of-meta-awareness-in-llm-fine-tuning |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.