text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Takeaways from the Mechanistic Interpretability Challenges
...plus more challenges on the way
===================================
Stephen Casper, [scasper@mit.edu](mailto:scasper@mit.edu)
Spoilers ahead
What happened?
==============
The Mechanistic Interpretability Challenges ([post](https://www.alignmentforum.... | https://www.lesswrong.com/posts/EjsA2M8p8ERyFHLLY/takeaways-from-the-mechanistic-interpretability-challenges |
# Updates and Reflections on Optimal Exercise after Nearly a Decade
Previously:
https://www.lesswrong.com/posts/bZ2w99pEAeAbKnKqo/optimal-exercise
Firstly, do the basic epistemics hold up? As far as I know, yes. The basic idea that lifting twice a week and doing cardio twice a week add up to a calorie expenditure tha... | https://www.lesswrong.com/posts/RydETq379eoWqBFvj/updates-and-reflections-on-optimal-exercise-after-nearly-a |
# A comparison of causal scrubbing, causal abstractions, and related methods
**Summary:** We explain the similarities and differences between three recent approaches to testing interpretability hypotheses:[ causal scrubbing](https://www.lesswrong.com/posts/JvZhhzycHu2Yd57RN/causal-scrubbing-a-method-for-rigorously-tes... | https://www.lesswrong.com/posts/uLMWMeBG3ruoBRhMW/a-comparison-of-causal-scrubbing-causal-abstractions-and |
# Bringing Little Kids to Contra Dances
Julia and I brought Lily to a contra dance for the first time when she was just a couple weeks old, and have continued bringing her (and now her younger siblings) occasionally as they've gotten older. A friend asked if I could write up something about the logistics of this, so h... | https://www.lesswrong.com/posts/EHMAbd2hbmaJPDH5f/bringing-little-kids-to-contra-dances |
# InternLM - China's Best (Unverified)
A technical report of [InternLM](https://github.com/InternLM/InternLM-techreport) on 6/7. It consisted of 104 billion parameters, was trained on 1.6 trillion tokens, and was fine-tuned for performance in Chinese.
The authors claimed that it performed second-best on the Chinese l... | https://www.lesswrong.com/posts/sKsFjvSkJ29uu6Tyq/internlm-china-s-best-unverified |
# D&D.Sci 5E: Return of the League of Defenders Evaluation & Ruleset
This is a follow-up to [last week's D&D.Sci scenario](https://www.lesswrong.com/posts/2uNeYiXMs4aQ2hfx9/d-and-d-sci-5e-return-of-the-league-of-defenders): if you intend to play that, and haven't done so yet, you should do so now before spoiling yours... | https://www.lesswrong.com/posts/CAHsE7WZEzCsMT78W/d-and-d-sci-5e-return-of-the-league-of-defenders-evaluation |
# Improvement on MIRI's Corrigibility
*This post was written as a submission for the* [*AI Alignment Award*](https://www.alignmentawards.com/shutdown)*, initiated at* [*EffiSciences'*](https://effisciences.org/) *event.*
This post aims to address the problem of corrigibility as identified by MIRI in 2015. We propose ... | https://www.lesswrong.com/posts/fNwDEHWFnHMtm8yH4/improvement-on-miri-s-corrigibility |
# Michael Shellenberger: US Has 12 Or More Alien Spacecraft, Say Military And Intelligence Contractors
Quotes:
> ...Multiple sources close to the matter have come forward to tell Public that Grusch’s core claims are accurate. The individuals are all either high-ranking intelligence officials, former intelligence offi... | https://www.lesswrong.com/posts/bhH2BqF3fLTCwgjSs/michael-shellenberger-us-has-12-or-more-alien-spacecraft-say |
# A plea for solutionism on AI safety
*\[Note for LW: This essay was written mostly for people who are inclined to downplay or dismiss AI risk, which means it has almost no intersection with this audience. Cross-posting it here anyway for feedback and in case you want to point anyone to it.\]*
Will AI kill us all?
T... | https://www.lesswrong.com/posts/ASMX9ss3J5G3GZdok/a-plea-for-solutionism-on-ai-safety |
# How accurate is data about past earth temperatures?
Such as [xkcd1732: Earth Temperature Timeline](https://xkcd.com/1732/), which is apparently based on [Global warming preceded by increasing carbon dioxide concentrations during the last deglaciation](https://www.nature.com/articles/nature10915), [A Reconstruction o... | https://www.lesswrong.com/posts/xbZLeXgf5SpRTR6ce/how-accurate-is-data-about-past-earth-temperatures |
# formalizing the QACI alignment formal-goal
*this work was done by Tamsin Leake and Julia Persson at Orthogonal.*
*thanks to mesaoptimizer for his help putting together this post.*
what does the QACI plan for formal-goal alignment actually look like when formalized as math? in this post, we'll be presenting our c... | https://www.lesswrong.com/posts/MR5wJpE27ymE7M7iv/formalizing-the-qaci-alignment-formal-goal |
# What are brains?
I see a self-reference problem with reductionism. I wonder if this has already been solved. So I'm asking.
Best as I can tell, there aren't actually *things* in reality. That's a human interpretation. It collapses upon almost any inspection, like [the Ship of Theseus](https://en.wikipedia.org/wiki/... | https://www.lesswrong.com/posts/aBDsJhkMnFcos3HKC/what-are-brains |
# A Theory of Unsupervised Translation
Motivated by Understanding Animal Communication
This is part of a [larger project](https://www.projectceti.org/) that aims to understand sperm whale communication. Jacob Andreas, Michael Bronstein, Antonio Torralba, and Shafi Goldwasser (an author of this paper) are part of the l... | https://www.lesswrong.com/posts/gkpNxngoZgCukjuiW/a-theory-of-unsupervised-translation-motivated-by |
# Ethodynamics of Omelas
***Epistemic status:** just kidding... haha... unless...*
Introduction
------------
A great amount of effort has been expended throughout history to decide the matter of "what is good" or "what is right" \[citation needed\]. This valiant but, let's face it, not tremendously successful effort... | https://www.lesswrong.com/posts/ikoi5PcyZBw5v8pbt/ethodynamics-of-omelas |
# I can see how I am Dumb
When I'm talking to somebody, sometimes I lose the conversational thread. Or sometimes I feel like there is this thing I want to say that seems relevant now, but I just can't remember it.
Or maybe I'm trying to solve a particular problem. I throw myself again and again at the problem, but it... | https://www.lesswrong.com/posts/KkgTtwogymBKSd3sE/i-can-see-how-i-am-dumb |
# The Dictatorship Problem
*(Disclaimer: This is my personal opinion, not that of any movement or organization.)*
This post aims to show that, over the next decade, it is *quite likely* that most democratic Western countries will become fascist dictatorships - this is not a [tail risk](https://en.wikipedia.org/wiki/T... | https://www.lesswrong.com/posts/pFaLqTHqBtAYfzAgx/the-dictatorship-problem |
# [Mostly solved] I get distracted while reading, but can easily comprehend audio text for 8+ hours per day. What are the best AI text-to-speech readers? Alternatively, do you have other ideas for what I could do?
For me, the physical act of scanning words takes active focus compared to the analogue in listening, whic... | https://www.lesswrong.com/posts/vrAtyspyAtdzuiwzy/mostly-solved-i-get-distracted-while-reading-but-can-easily |
# Snake Eyes Paradox
Disclaimer: this is a [personal blog](https://www.lesswrong.com/posts/5conQhfa4rgb4SaWx/site-guide-personal-blogposts-vs-frontpage-posts) post that I'm posting for my benefit and do not expect to be valuable to LessWrong readers in general. If you're reading this, consider not.
This post analyzes... | https://www.lesswrong.com/posts/ox3RFPGriL2dtg8TR/snake-eyes-paradox |
# You Are a Computer, and No, That’s Not a Metaphor
*Preface: this blog post was written for a slightly more general audience than LessWrong. It's likely that you may already be familiar with many of the ideas explored in this post, but I thought some of you may find it interesting anyway.*
I've been exploring the id... | https://www.lesswrong.com/posts/A85kAWkK4D4ctvord/you-are-a-computer-and-no-that-s-not-a-metaphor |
# Inference-Time Intervention:
Eliciting Truthful Answers from a Language Model
Excited to announce our new work: Inference-Time Intervention (ITI), a minimally-invasive control technique that significantly improves LLM truthfulness using little resources, benchmarked on the TruthfulQA dataset. [Preprint link](https:/... | https://www.lesswrong.com/posts/kuQfnotjkQA4Kkfou/inference-time-intervention-eliciting-truthful-answers-from |
# [Linkpost] Large Language Models Converge on Brain-Like Word Representations
*This is a linkpost for* [*https://arxiv.org/abs/2306.01930*](https://arxiv.org/abs/2306.01930)
> One of the greatest puzzles of all time is how understanding arises from neural mechanics. Our brains are networks of billions of biological ... | https://www.lesswrong.com/posts/2QexGHrqSxcuwyGmf/linkpost-large-language-models-converge-on-brain-like-word |
# why I'm anti-YIMBY
Scott Alexander [posted this](https://astralcodexten.substack.com/p/change-my-mind-density-increases) last month, arguing that building more in an area doesn't necessarily decrease local prices because it can cause more people to move in.
I see 3 levels of argument here, based on 1st, 2nd, and 3r... | https://www.lesswrong.com/posts/xkTzL4HHrLnE6zpLo/why-i-m-anti-yimby |
# Critiques of prominent AI safety labs: Conjecture
*Cross-posted from the EA Forum. See the* [*original here*](https://forum.effectivealtruism.org/posts/gkfMLX4NWZdmpikto/critiques-of-prominent-ai-safety-labs-conjecture)*. Internal linking has not been updated for LW due to time constraints and will take you back to ... | https://www.lesswrong.com/posts/9jvrQToSq3CYvoeHf/critiques-of-prominent-ai-safety-labs-conjecture |
# UK PM: $125M for AI safety
The UK PM recently announced [$125M for a foundation model task force.](https://www.gov.uk/government/news/initial-100-million-for-expert-taskforce-to-help-uk-build-and-adopt-next-generation-of-safe-ai) While the announcement stressed AI safety, it also stressed capabilities. But this morn... | https://www.lesswrong.com/posts/ownkwjdeC9iFXqjGf/uk-pm-usd125m-for-ai-safety |
# If you are too stressed, walk away from the front lines
**tl;dr:** If your reason for cramming AI knowledge into your brain is stress, then don't do it. You can still be useful, but walk away from the front lines where people are directly responsible.
*Disclaimer: 1) This is an obvious problem that has already bee... | https://www.lesswrong.com/posts/Z6eH8PoNHyMcTM2vP/if-you-are-too-stressed-walk-away-from-the-front-lines |
# Explicitness
*[Metadata: crossposted from [https://tsvibt.blogspot.com/2023/03/explicitness.html](https://tsvibt.blogspot.com/2023/03/explicitness.html). First completed March 3, 2023.]*
Explicitness is out-foldedness. An element of a mind is explicit when it is available to relate to other elements when suitable.... | https://www.lesswrong.com/posts/KuKaQEu7JjBNzcoj5/explicitness |
# Manifold Predicted the AI Extinction Statement and CAIS Wanted it Deleted
*Crosspost from* [*https://news.manifold.markets/p/manifold-predicted-the-ai-extinction*](https://news.manifold.markets/p/manifold-predicted-the-ai-extinction)
History may look back on CAIS’s AI risk statement as a pivotal moment IF we surviv... | https://www.lesswrong.com/posts/KS7iao3MvnqT2PLGm/manifold-predicted-the-ai-extinction-statement-and-cais |
# Introduction to Towards Causal Foundations of Safe AGI
*By Tom Everitt, Lewis Hammond, Rhys Ward, Ryan Carey, James Fox, Sebastian Benthall, Matt MacDermott and Shreshth Malik representing the *[*Causal Incentives Working Group*](https://causalincentives.com/)*. Thanks also to Toby Shevlane, MH Tessler, Aliya Ahmad,... | https://www.lesswrong.com/posts/3oQCY4he4zRzF6vQb/introduction-to-towards-causal-foundations-of-safe-agi |
# ARC is hiring theoretical researchers
**The Alignment Research Center’s Theory team is starting a new hiring round for researchers with a theoretical background. Please apply [here](https://jobs.lever.co/alignment.org/617488b1-d742-4990-a037-b7f0e2ba68c9/apply).**
**_Update January 2024_: we have paused hiring and ... | https://www.lesswrong.com/posts/d99ikjqdxpMiAFnch/arc-is-hiring-theoretical-researchers-1 |
# Aura as a proprioceptive glitch
So I recently went on a Vipassana retreat, which is a fun little experience that involves being silent for 10 days and meditating 11 hours each day. Needless to say, that sort of situation necessarily leads to all sorts of interesting experiences and ideas. Among them, one I found esp... | https://www.lesswrong.com/posts/DyGX8fyyhhyzPD7rp/aura-as-a-proprioceptive-glitch |
# Book Review: Autoheterosexuality
*This review was originally written for the *[*Astral Codex Ten Book Review Contest*](https://astralcodexten.substack.com/p/book-review-contest-rules-2023)*. Unfortunately it didn’t make it as one of the finalists. but since I made use of the* [*LessWrong proofreading/feedback servi... | https://www.lesswrong.com/posts/nqkFcjP6wtQrQMiN2/book-review-autoheterosexuality |
# Contingency: A Conceptual Tool from Evolutionary Biology for Alignment
-1\. Motivation
===============
As AI systems become increasingly powerful, the chance of them developing dangerous features becomes increasingly likely. A key concern of AI alignment researchers is that there are dangerous features, such as pow... | https://www.lesswrong.com/posts/GAbEcLDQLtHZE78TY/contingency-a-conceptual-tool-from-evolutionary-biology-for-2 |
# UFO Betting: Put Up or Shut Up
Are you confident in your **current ontology**? Are you convinced that **ultimately all ufos are prosaic** in nature?
If so, do you want some **immediate free money**?
I suspect that LW's are overconfident in their views on ufos/uap. As such, I'm willing to offer what I think many wi... | https://www.lesswrong.com/posts/t5W87hQF5gKyTofQB/ufo-betting-put-up-or-shut-up |
# <$750k grants for General Purpose AI Assurance/Safety Research
Georgetown University's Center for Security and Emerging Technology (CSET) is accepting applications for AI Safety / AI Assurance research grants.
They are offering up to **$750k** ***per project*** accepted, expended over 6-24 months.
1-2 page expres... | https://www.lesswrong.com/posts/5f5mPjpKWNpmbkfSm/less-than-usd750k-grants-for-general-purpose-ai-assurance |
# TASRA: A Taxonomy and Analysis of Societal-Scale Risks from AI
Partly in response to calls for more detailed accounts of how AI could go wrong, e.g., from [Ng and Bengio](https://twitter.com/AndrewYNg/status/1666582174257254402)'s recent exchange on Twitter, here's a new paper with Stuart Russell:
* Discussion on... | https://www.lesswrong.com/posts/zKkZanEQc4AZBEKx9/tasra-a-taxonomy-and-analysis-of-societal-scale-risks-from |
# Virtual AI Safety Unconference (VAISU)
Virtual AI Safety Unconference (VAISU) is an online event for both established and aspiring AI safety researchers. The event is a space for you to have productive research discussions around the question: “How do we make sure that AI systems will be safe and beneficial, both in... | https://www.lesswrong.com/events/CfGDGKw8ddebmrZho/virtual-ai-safety-unconference-vaisu |
# The Dial of Progress
“There is a single light of science. To brighten it anywhere is to brighten it everywhere.” – Isaac Asimov
You cannot stand what I’ve become
You much prefer the gentleman I was before
I was so easy to defeat, I was so easy to control
I didn’t even know there was a war
– Leonard Cohen, Th... | https://www.lesswrong.com/posts/Zzar6BWML555xSt6Z/the-dial-of-progress |
# MetaAI: less is less for alignment.
Summary
=======
In May 2023, MetaAI submitted a paper to arxiv called [LIMA: Less Is More for Alignment](https://arxiv.org/abs/2305.11206). It's a pretty bad paper and (in my opinion) straightforwardly misleading. Let's get into it.
The Superficial Alignment Hypothesis
=========... | https://www.lesswrong.com/posts/uyk5nn93HxJMsio98/metaai-less-is-less-for-alignment-1 |
# Multiple stages of fallacy - justifications and non-justifications for the multiple stage fallacy
\[Epistemic status: I am reasonably confident (75%) that these claims are largely correct.\]
* * *
To estimate the probability of something happening, e.g. AI takeover, one approach is to break this down into stages ... | https://www.lesswrong.com/posts/kmZkCmz6AiJntjWDG/multiple-stages-of-fallacy-justifications-and-non |
# Intelligence allocation from a Mean Field Game Theory perspective
Each of us now faces a choice every minute of every day:
*Should I really use **my** brain power for this, or shoud I let **ChatGPT** do it?*
Using your own brain has its merits. You might get more intelligent and resilient in the long run. But... | https://www.lesswrong.com/posts/Wr3dSxQKyKH9EcCDE/intelligence-allocation-from-a-mean-field-game-theory |
# My guess for why I was wrong about US housing
Edit/update: this is receiving more upvotes and I think it is worth explicitly pointing out that while housing is up 60% over this time frame, the Sp500 is up 90+%, so this is not intended as a full on refutation of all the points made in the original post.
A few years ... | https://www.lesswrong.com/posts/mYEcfGCDy74NMi7QQ/my-guess-for-why-i-was-wrong-about-us-housing |
# Lightcone Infrastructure/LessWrong is looking for funding
Lightcone Infrastructure is looking for funding and are working on the following projects:
* We run LessWrong, the [AI Alignment Forum](https://www.alignmentforum.org/), and have written a lot of the code behind the [Effective Altruism Forum.](https://for... | https://www.lesswrong.com/posts/9iDw6ugMPk7pmXuyW/lightcone-infrastructure-lesswrong-is-looking-for-funding |
# Philosophical Cyborg (Part 1)
*This post is part of the output from AI Safety Camp 2023’s Cyborgism track, run by Nicholas Kees Dupuis - thank you to AISC organizers & funders for their support. Thank you for comments from Peter Hroššo; and the helpful background of conversations about the possibilities (and limits... | https://www.lesswrong.com/posts/k93NEoXZq6CdXegdx/philosophical-cyborg-part-1 |
# Progress links and tweets, 2023-06-14
**Announcements & opportunities**
---------------------------------
* [The Long Now Foundation is hiring an Executive Director](https://twitter.com/stewartbrand/status/1667261923413626881)
* [Terraform Industries (solar-powered carbon-neutral natural gas synthesis) is hirin... | https://www.lesswrong.com/posts/rDip2ugH78njeMsea/progress-links-and-tweets-2023-06-14 |
# On the Apple Vision Pro
Apple is offering a VR/AR/XR headset, Vision Pro, for the low, low price of $3,500.
I kid. Also I am deadly serious.
The value of this headset to a middle class American or someone richer than that is almost certainly either vastly more than $3,500, or at best very close to $0.
This type o... | https://www.lesswrong.com/posts/6BKnGoY9knSvi5zJM/on-the-apple-vision-pro |
# Instrumental Convergence? [Draft]
This is a draft written by [J. Dmitri Gallow](http://jdmitrigallow.com/), Senior Research Fellow at the Dianoia Institute of Philosophy at ACU, as part of the Center for AI Safety Philosophy Fellowship. This draft is meant to solicit feedback. Here is a [PDF version](https://drive.g... | https://www.lesswrong.com/posts/w8PNjCS8ZsQuqYWhD/instrumental-convergence-draft |
# Why libertarians are advocating for regulation on AI
Motivation: some people on the internet seem confused about why libertarians (and others generally suspicious of government intervention) are advocating for regulation to help prevent x-risk from AI. The arguments here aren't novel and generally seem pretty obvio... | https://www.lesswrong.com/posts/8QNQS7NisFBWCE6Jw/why-libertarians-are-advocating-for-regulation-on-ai |
# Looking Back On Ads
> It is difficult to get a man to understand something, when his salary depends on his not understanding it.
Talking about ads online [I would often](https://news.ycombinator.com/item?id=21617197) [get a response](https://news.ycombinator.com/item?id=27989974) that [as someone working in ads](ht... | https://www.lesswrong.com/posts/Gm78KKKJJk2xWuhdm/looking-back-on-ads |
# Matt Taibbi's COVID reporting
> Michael Shellenberger’s Public today released a blockbuster story, “First Person Sickened By COVID-19 Was Chinese Scientist Who Oversaw “Gain Of Function” Research That Created Virus,” which generously credits Racket. The story cites three government officials in naming scientist Ben ... | https://www.lesswrong.com/posts/DF5fnraMGHopb8Kj4/matt-taibbi-s-covid-reporting |
# Why "AI alignment" would better be renamed into "Artificial Intention research"
"AI alignment" has the application, the agenda, less charitably the activism, right in the name. It is a lot like "[Missiology](https://en.wikipedia.org/wiki/Missiology)" (the study of how to proselytize to "the savages") which had to ev... | https://www.lesswrong.com/posts/MksJyLgJ8JQFexiWi/why-ai-alignment-would-better-be-renamed-into-artificial |
# I still think it's very unlikely we're observing alien aircraft
Some suggest there might be alien aircraft on Earth now. The argument goes something like this:
(1) A priori, there’s no reason there *shouldn’t* be alien aircraft. Earth is 4.54 billion years old, but the universe is 13.7 billion years old, and within... | https://www.lesswrong.com/posts/nwJCzszw8gGjPTihM/i-still-think-it-s-very-unlikely-we-re-observing-alien |
# AI #16: AI in the UK
It would make strategic sense, even from a selfish perspective, for essentially any nation to get more involved in sensible AI regulation, and to get more involved in capturing AI benefits and being the home to AI and AI-enabled economic engines.
It makes especially good sense for the UK, wh... | https://www.lesswrong.com/posts/pKhzQyaWwes63JFwp/ai-16-ai-in-the-uk |
# AXRP Episode 22 - Shard Theory with Quintin Pope
[YouTube link](https://www.youtube.com/watch?v=o-Qc_jiZTQQ&list=PLmjaTS1-AiDeqUuaJjasfrM6fjSszm9pK&index=22)
What can we learn about advanced deep learning systems by understanding how humans learn and form values over their lifetimes? Will superhuman AI look like ru... | https://www.lesswrong.com/posts/4rmvMThJYNcCptAya/axrp-episode-22-shard-theory-with-quintin-pope |
# Developing a technology with safety in mind: Lessons from the Wright Brothers
If a technology may introduce catastrophic risks, how do you develop it?
It occurred to me that the Wright Brothers’ approach to inventing the airplane might make a good case study.
The catastrophic risk for them, of course, was dying in... | https://www.lesswrong.com/posts/RoBZTMKYuZq2PbPvb/developing-a-technology-with-safety-in-mind-lessons-from-the |
# human intelligence may be alignment-limited
Previously, I [argued that](https://www.lesswrong.com/posts/rSycgquipFkozDHzF/ai-self-improvement-is-possible) human mental development implies that AI self-improvement from sub-human capabilities is possible, and that human intelligence comes at the cost of a longer child... | https://www.lesswrong.com/posts/zGoDXgLFetEdvLdFH/human-intelligence-may-be-alignment-limited |
# Leveling Up Or Leveling Off? Understanding The Science Behind Skill Plateaus
*This post is* [*crossposted from my blog*](https://lynettebye.com/blog/2023/5/30/understanding-the-science-behind-skill-plateaus)*. If you liked this post,* [*subscribe to Lynette's blog*](https://lynettebye.com/email) *to read more -- I o... | https://www.lesswrong.com/posts/2iehi9XjfpdDK7Z2K/leveling-up-or-leveling-off-understanding-the-science-behind |
# Rational Animations is looking for an AI Safety scriptwriter, a lead community manager, and other roles.
UPDATE 07/15/2023: Applications are now closed. If you still want to apply, I will consider you for the future, which may be as soon as a few months from now, or I could initially offer you a lower salary if you'... | https://www.lesswrong.com/posts/mC5eWRwkfi3vJ9XB3/rational-animations-is-looking-for-an-ai-safety-scriptwriter |
# DSLT 1. The RLCT Measures the Effective Dimension of Neural Networks
*TLDR; This is the first post of Distilling Singular Learning Theory (DSLT), an introduction to which can be read at* [*DSLT0*](https://www.lesswrong.com/posts/xRWsfGfvDAjRWXcnG/dslt-0-distilling-singular-learning-theory)*. In this post I explain h... | https://www.lesswrong.com/posts/4eZtmwaqhAgdJQDEg/dslt-1-the-rlct-measures-the-effective-dimension-of-neural |
# DSLT 0. Distilling Singular Learning Theory
*TLDR; In this sequence I distill Sumio Watanabe's* Singular Learning Theory (SLT) *by explaining the essence of its main theorem - Watanabe's Free Energy Formula for Singular Models - and illustrating its implications with intuition-building examples. I then show why neur... | https://www.lesswrong.com/posts/xRWsfGfvDAjRWXcnG/dslt-0-distilling-singular-learning-theory |
# Scaffolded LLMs: Less Obvious Concerns
“*Words, like tranquil waters behind a dam, can become reckless and uncontrollable torrents of destruction when released without caution and wisdom.”*
— William Arthur Ward
In this post I aim to shed light on lesser-discussed concerns surrounding Scaffolded LLMs (S-LLMs).... | https://www.lesswrong.com/posts/mAwxebLw3nYbDivmt/scaffolded-llms-less-obvious-concerns |
# How not to write the Cookbook of Doom?
I would like to be given pointers to prior art on the following question: how do we communicate about potential risks from AI in a net-positive way, taking into account the negative that is giving people ideas? I can easily see how someone describing steps an AI takeover could ... | https://www.lesswrong.com/posts/5ExcoP5Soc5kEXH3k/how-not-to-write-the-cookbook-of-doom |
# Explaining "Taking features out of superposition with sparse autoencoders"
*\[Thanks to* [*Logan Riggs*](https://www.lesswrong.com/users/elriggs) *and* [*Hoagy*](https://www.lesswrong.com/users/hoagy) *for their help writing this post.\]*
In this post, I’m going to translate the post [\[Interim research report\] Ta... | https://www.lesswrong.com/posts/a4oPE4xJqkYSz6jMS/explaining-taking-features-out-of-superposition-with-sparse |
# Conjecture: A standing offer for public debates on AI
**Tl;dr: If you want to publicly debate AI risk with us, send us an email at hello@conjecture.dev with information about you, suggested topics, and the suggested platform.**
Public debates strengthen society and public discourse. They spread truth by testing ide... | https://www.lesswrong.com/posts/q5kio5Tz4C3bDJ8N6/conjecture-a-standing-offer-for-public-debates-on-ai |
# The ones who endure
**Content warning: death, long-lasting suffering. This story was rough to write, and may well be rough to read. It's intended as a response to Le Guin's** [***The Ones Who Walk Away From Omelas***](https://shsdavisapes.pbworks.com/f/Omelas.pdf)**, although it can be read standalone.**
> *If you ... | https://www.lesswrong.com/posts/X6pKMHS5xAeiNaFts/the-ones-who-endure |
# [Linkpost] Faith and Fate: Limits of Transformers on Compositionality
I thought it'd be especially interesting to get critiques/discussion from the LW crowd, because the claims here seem antithetical to a lot of the beliefs people here have, mostly around just how capable and cognizant transformers are/can be.
The ... | https://www.lesswrong.com/posts/bgXQJiCyf7iqAzyrk/linkpost-faith-and-fate-limits-of-transformers-on |
# Palantir's AI models
Palantir published marketing material for their offering of AI for defense purposes. There's a video of how a military commander could order a military strike on an enemy tank with the help of LLMs.
One of the features that Palantir advertises is:
> Agents
>
> Define LLM agents to pursue sp... | https://www.lesswrong.com/posts/CprMvdyrCeyJLu5uW/palantir-s-ai-models |
# LLMs Sometimes Generate Purely Negatively-Reinforced Text
When using adversarial training, should you remove sensitive information from the examples associated with the lowest possible reward?
In particular, can a real language models generate text snippets which were only present in purely negatively-reinforced te... | https://www.lesswrong.com/posts/sbGau4QBwToYWEg4k/llms-sometimes-generate-purely-negatively-reinforced-text |
# [Replication] Conjecture's Sparse Coding in Small Transformers
Summary
-------
A couple of weeks ago Logan Riggs and I [posted](https://www.lesswrong.com/posts/E8imxQo96WgDCMxkA/replication-conjecture-s-sparse-coding-in-toy-models) that we'd replicated the toy-model experiments in Lee Sharkey and Dan Braun's [origi... | https://www.lesswrong.com/posts/vBcsAw4rvLsri3JAj/replication-conjecture-s-sparse-coding-in-small-transformers |
# Updating Drexler's CAIS model
Eric Drexler's report [Reframing Superintelligence: Comprehensive AI Services (CAIS) as General Intelligence](https://www.fhi.ox.ac.uk/reframing/) reshaped how a lot of people think about AI ([summary 1](https://www.lesswrong.com/posts/x3fNwSe5aWZb5yXEG/reframing-superintelligence-compr... | https://www.lesswrong.com/posts/a5NxvzFGddj2e8uXQ/updating-drexler-s-cais-model |
# June and Mulberries
Around here the [Juneberries](https://en.wikipedia.org/wiki/Amelanchier) and [Mulberries](https://en.wikipedia.org/wiki/Morus_(plant)) are ripe.
[](https://www.jefftk.com/juneberries-june-big.jpg)
The berries are delicious, but a lot of people don... | https://www.lesswrong.com/posts/DMdJhAb7fer7XGckm/june-and-mulberries |
# The AI governance gaps in developing countries
As developed countries rapidly become more equipped in the governance of safe and beneficial AI systems, developing countries are slackened off in the global AI race and standing at risk of extreme vulnerabilities. By examining not only “how we can effectively govern AI... | https://www.lesswrong.com/posts/cM7sR7seBRwtxctGY/the-ai-governance-gaps-in-developing-countries |
# Are Bayesian methods guaranteed to overfit?
Yuling Yao argues that Bayesian models are guaranteed to overfit. He summarizes his point as follows:
> I have a different view. Bayesian model does overfit.
>
> Moreover, Bayes is guaranteed to overfit, regardless of the model (being correct or wrong) or the prior ( ... | https://www.lesswrong.com/posts/PhMaRNBaeSKCfuZi9/are-bayesian-methods-guaranteed-to-overfit |
# Alewife Train is Now Arriving
A few months ago [I complained](https://www.jefftk.com/p/attention-passengers-not-for-signs) that the text version of the MBTA train arrival announcements include a not-needed-for-text "attention passengers" which delayed the key information to the second screen. I'm happy to report tha... | https://www.lesswrong.com/posts/ornkQKxabAxCmbmXP/alewife-train-is-now-arriving |
# [Linkpost] Rosetta Neurons: Mining the Common Units in a Model Zoo
*This is a linkpost for* [*https://arxiv.org/abs/2306.09346*](https://arxiv.org/abs/2306.09346)*. *
> Do different neural networks, trained for various vision tasks, share some common representations?
> In this paper, we demonstrate the existence ... | https://www.lesswrong.com/posts/pAiz3qtjzsEk6ALZg/linkpost-rosetta-neurons-mining-the-common-units-in-a-model |
# A summary of current work in AI governance
A summary of current work in AI governance
==========================================
If you’d like to learn more about AI governance, apply to the [AI Safety Fundamentals: Governance Track](https://www.agisafetyfundamentals.com/ai-governance-curriculum), a 12-week, part-t... | https://www.lesswrong.com/posts/GEJtDHMfuW4vZ5msG/a-summary-of-current-work-in-ai-governance |
# The environment as infrastructure
A good metaphor for the ideal relationship between humanity and the environment is that **the environment is like critical infrastructure.**
Infrastructure is valuable, because it provides crucial services. You want to maintain it carefully, because it’s bad if it breaks down.
But... | https://www.lesswrong.com/posts/8bxHJyfvtAKPN8mXC/the-environment-as-infrastructure |
# Adventist Health Study-2 supports pescetarianism more than veganism
Or: how the Adventist Health Study-2 had a pretty good study design but was oversold in popular description, and then misrepresented its own results.
When I [laid out](https://acesounderglass.com/2023/05/30/change-my-mind-veganism-entails-trade-off... | https://www.lesswrong.com/posts/AhJFyEDckQcPTxCG2/adventist-health-study-2-supports-pescetarianism-more-than |
# What is the foundation of me experiencing the present moment being right now and not at some other point in time?
This question may be either trivial to answer and I just don‘t get it, or it may be highly controversial.
I am trying to pose my problem as shortly as I can.
What is responsible for my subjective experi... | https://www.lesswrong.com/posts/CFuY9DCAgH4HJyjd7/what-is-the-foundation-of-me-experiencing-the-present-moment |
# The "Loss Function of Reality" Is Not So Spiky and Unpredictable
### *In other words, there are no "rabbit holes" or "dark tunnels of the mind."*
The great success of neural networks is due in large part to the surprising ease in which gradient-based methods are used to train them. Neural networks are often trained... | https://www.lesswrong.com/posts/gaeAyvq75NDtgps6J/the-loss-function-of-reality-is-not-so-spiky-and |
# A few more ants and grasshoppers
*Literary status: lacking in grace, hopefully compensated by its information density. Inspired by, and in reply to Richard Ngo’s* [*post of parables*](https://www.lesswrong.com/posts/GJgudfEvNx8oeyffH/the-ants-and-the-grasshopper)*.*
*Disclaimer: Though I cite ecological economics a... | https://www.lesswrong.com/posts/B3ThCLosQfJRjqCXy/a-few-more-ants-and-grasshoppers |
# DSLT 2. Why Neural Networks obey Occam's Razor
*TLDR; This is the second main post of* [*Distilling Singular Learning Theory*](https://www.lesswrong.com/s/czrXjvCLsqGepybHC) *which is introduced in* [*DSLT0*](https://www.lesswrong.com/posts/xRWsfGfvDAjRWXcnG/dslt-0-distilling-singular-learning-theory)*. I synthesise... | https://www.lesswrong.com/posts/CZHwwDd7t9aYra5HN/dslt-2-why-neural-networks-obey-occam-s-razor |
# Solomonoff induction still works if the universe is uncomputable, and its usefulness doesn't require knowing Occam's razor
*Note: I don't think this idea is original, but I couldn't find a good post going over the implications.*
I used to think that Solomonoff induction was a bit arbitrary for the following reason:... | https://www.lesswrong.com/posts/ioJGAt6x4q3GGKhrG/solomonoff-induction-still-works-if-the-universe-is |
# Cryonics Career Survey (more jobs than you think)
*([x-post LessDead](https://lessdead.com/cryonics-career-survey))*
# Survey presentation
**Link to survey**: [Cryonics Career Survey](https://forms.gle/w1Ag6Zy7kTTCPnYP7)
**Background**. During the annual Teens and Twenties Cryonics Gathering (a fully funded yearl... | https://www.lesswrong.com/posts/oLSRDBn5d5ihvuC8P/cryonics-career-survey-more-jobs-than-you-think |
# UK Foundation Model Task Force - Expression of Interest
Ian Hogarth has just been announced as the Chair of the UK's AI Foundation Model Taskforce. He's the author of the FT article "[We must slow down the race to God-like AGI](https://www.ft.com/content/03895dc4-a3b7-481e-95cc-336a524f2ac2)", and seems to take X-ri... | https://www.lesswrong.com/posts/Qb7qiD4gPHLbwpzQY/uk-foundation-model-task-force-expression-of-interest |
# Alaska Trip
A few weeks ago I flew to Anchorage to play for the [Dancing Bears Dance Camp](https://www.thedancingbears.com/Dance-Camp.html) with the [Free Raisins](https://www.freeraisins.com/). It was really fun to play with Amy and Audrey again, to dance to [KGB](https://kgbmole.com/kgb/index.html), to meet people... | https://www.lesswrong.com/posts/5sS4vkP5hxt6eRXPF/alaska-trip |
# My impression of singular learning theory
**Disclaimer**: I'm by no means an expert on singular learning theory and what I present below is a simplification that experts might not endorse. Still, I think it might be more comprehensible for a general audience than going into digressions about blowing up singularities... | https://www.lesswrong.com/posts/DqLHvJjuPdtrzuoas/my-impression-of-singular-learning-theory |
# Why I am not an AI extinction cautionista
First, to dispense with what should be obvious, if a superintelligent agent wants to destroy humans, we are completely and utterly hooped. All the arguing about "but how would it...?" indicates lack of imagination.
 that have gotten wide adoption. I’m perhaps not the *very* most qualified person to write this post, but I’v... | https://www.lesswrong.com/posts/HJNtrNHf688FoHsHM/guide-to-rationalist-interior-decorating |
# When did you orient?
John David Pressman [says](https://twitter.com/jd_pressman/status/1670677402341761024) that the orienting is the tricky and valuable and underdocumented part of OODA. It occurs to me that I can hardly name any times anyone I know "oriented". When did you orient and what happened? | https://www.lesswrong.com/posts/qE8tDuxPXSTeWAtGo/when-did-you-orient |
# Provisionality
*[Metadata: crossposted from [https://tsvibt.blogspot.com/2023/03/provisionality.html](https://tsvibt.blogspot.com/2023/03/provisionality.html). First completed March 7, 2023.]*
A mental element has to be open to revision, and so it has to be treated as though it might be revised.
*Thanks to Sa... | https://www.lesswrong.com/posts/5ie8Mfwq5tEBCBDEC/provisionality |
# Experiments in Evaluating Steering Vectors
By evaluating how well steering vectors perform using GPT-3, we can score a machine-generated set of steering vectors automatically. We also find that, by combining steering vectors that succeed in different ways, we can yield a better and more general steering vector than ... | https://www.lesswrong.com/posts/GGttHsdWHjh94cX5G/experiments-in-evaluating-steering-vectors |
# Mode collapse in RL may be fueled by the update equation
**TL;DR:** We present an advantage variant which, in certain settings, does not train an optimal policy, but instead uses a fixed reward to update a policy a fixed amount from initialization. Non-tabular empirical results seem mixed: The policy doesn't mode-co... | https://www.lesswrong.com/posts/A7RgYuYH4HywNeYWD/mode-collapse-in-rl-may-be-fueled-by-the-update-equation |
# Lessons On How To Get Things Right On The First Try
*This post is based on several true stories, from a workshop which John has run a few times over the past year.*
**John:** Welcome to the Ball -> Cup workshop! Your task for today is simple: I’m going to roll this metal ball:
.
In this post, I propose a (relatively strict) prediction-based eval which is well-defi... | https://www.lesswrong.com/posts/8CvkNa6FKSrK4Nj83/ban-development-of-unpredictable-powerful-models |
# OpenAI introduces function calling for GPT-4
OpenAI’s latest [update](https://openai.com/blog/function-calling-and-other-api-updates) unlocks function calling for GPT-4 and GPT-3.5 Turbo. This enables developers to "*describe functions to gpt-4-0613 and gpt-3.5-turbo-0613, and have the model intelligently choose to ... | https://www.lesswrong.com/posts/cnPeYAefQCtaA5PRa/openai-introduces-function-calling-for-gpt-4 |
# 10 quick takes about AGI
I have a bunch of loosely related and not fully fleshed out ideas for future posts.
In the spirit of [10 reasons why lists of 10 reasons might be a winning strategy](https://www.lesswrong.com/posts/8gbfvGhSnEJ9hHGew/10-reasons-why-lists-of-10-reasons-might-be-a-winning), I've written some o... | https://www.lesswrong.com/posts/C5Tp4GzGdaRzRnAmD/10-quick-takes-about-agi |
# DSLT 3. Neural Networks are Singular
*TLDR; This is the third main post of* [*Distilling Singular Learning Theory*](https://www.lesswrong.com/s/czrXjvCLsqGepybHC) *which is introduced in* [*DSLT0*](https://www.lesswrong.com/posts/xRWsfGfvDAjRWXcnG/dslt-0-distilling-singular-learning-theory)*. I explain that neural n... | https://www.lesswrong.com/posts/tZwaGp5wMQqKh3krz/dslt-3-neural-networks-are-singular |
# Are the majority of your ancestors farmers or non-farmers?
Assume humanity began with homo sapiens ~300,000 years ago. Are the majority of your ancestors farmers or non-farmers?
My friend and I had very differing intuitions on this, and after thinking about it some, both of us currently think it's very non-ob... | https://www.lesswrong.com/posts/tsiQkgjLfQ8cMJe3x/are-the-majority-of-your-ancestors-farmers-or-non-farmers |
# A Friendly Face (Another Failure Story)
*Edit: Based on the comment by Daniel Kokotajlo, we extended the dialog in the chapter "Takeover from within" by a few lines.*
The perfect virtual assistant
-----------------------------
The year is 2026 and the race for human-level artificial general intelligence (AGI) draw... | https://www.lesswrong.com/posts/iRFxvNeLbHNRCzA2S/a-friendly-face-another-failure-story |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.