text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Calling My Second Family Dance
Yesterday I called my second family dance, which went pretty well! I've stopped calling regular contra dances because I don't enjoy it very much anymore, but these two have been a lot of fun and I'm up for doing more.
As usual we played with an open band, and this time Nora decided to... | https://www.lesswrong.com/posts/upspuxNFAzaHPcrrL/calling-my-second-family-dance |
# I would have shit in that alley, too
After living in a suburb for most of my life, when I moved to a major U.S. city the first thing I noticed was the feces. At first I assumed it was dog poop, but my naivety didn’t last long.
One day I saw a homeless man waddling towards me at a fast speed while holding his ass ch... | https://www.lesswrong.com/posts/sCWe5RRvSHQMccd2Q/i-would-have-shit-in-that-alley-too |
# [Linkpost] Transcendence: Generative Models Can Outperform The Experts That Train Them
Authors: Edwin Zhang, Vincent Zhu, Naomi Saphra, Anat Kleiman, Benjamin L. Edelman, Milind Tambe, Sham M. Kakade, Eran Malach.
Abstract:
> Generative models are trained with the simple objective of imitating the conditional prob... | https://www.lesswrong.com/posts/aXz9xfe4ifmHLB8hb/linkpost-transcendence-generative-models-can-outperform-the |
# On DeepMind’s Frontier Safety Framework
#### **On DeepMind’s Frontier Safety Framework**
Previously: [On OpenAI’s Preparedness Framework](https://thezvi.substack.com/p/on-openais-preparedness-framework?utm_source=publication-search), [On RSPs](https://thezvi.substack.com/p/on-responsible-scaling-policies-rsps?utm_s... | https://www.lesswrong.com/posts/frEYsehsPHswDXnNX/on-deepmind-s-frontier-safety-framework |
# Suffering Is Not Pain
“*Pain is inevitable; suffering is optional*.”
The motivation of this post is to address the persistent conflation between suffering and pain I have observed from members of the EA community, even amongst those who purport to be “suffering-focused” in their ethical motivations. In order to bes... | https://www.lesswrong.com/posts/Y4hFATbyPL9Zp3frX/suffering-is-not-pain |
# Book review: the Iliad
_Translated by Emily Wilson_
1.
I didn't know what the Iliad was about. I thought it was the story of how Helen of Troy gets kidnapped, triggering the Trojan war, which lasts a long time and eventually gets settled with a wooden horse.
Instead it's just a few days, nine years into that war.... | https://www.lesswrong.com/posts/bdz4yPxxRMxeKJHay/book-review-the-iliad |
# Loving a world you don’t trust
*(Cross-posted from [my website](https://joecarlsmith.com/2024/06/18/loving-a-world-you-dont-trust). Audio version [here](https://joecarlsmithaudio.buzzsprout.com/2034731/15259745-loving-a-world-you-don-t-trust), or search for "Joe Carlsmith Audio" on your podcast app.)*
*This is the ... | https://www.lesswrong.com/posts/iqNjYdsectt5TvJRh/loving-a-world-you-don-t-trust |
# Boycott OpenAI
I have canceled my OpenAI subscription in protest over OpenAI's lack of
ethics.
In particular, I object to:
- threats to confiscate departing employees' equity unless those
employees signed a life-long non-disparagement contract
- Sam Altman's pattern of lying about important topics
... | https://www.lesswrong.com/posts/sXhBCDLJPEjadwHBM/boycott-openai |
# Gizmo Watch Review
*Summary: the watches do the basic things they need to and have let us give our kids more freedom, but there are also a bunch of annoying limitations and they're 36% more expensive than they look. I'm overall happy we got them.*
Our older two kids, Lily and Anna, are ten and eight, and are mature... | https://www.lesswrong.com/posts/5tRrZxxCLvqbqJR9x/gizmo-watch-review |
# Self responsibility
I see a modern puzzle of "self responsibility" which gets tricky quickly. At all times a person has responsibility to themselves. From child to adult, you are “self responsible”. It always applies.
When a person isn't of a sound mind, they are still expected to maintain their responsibility b... | https://www.lesswrong.com/posts/DBRScDwh6krAkLerQ/self-responsibility-1 |
# Surviving Seveneves
*Contains spoilers for the first couple of chapters of Seveneves*
*Highly speculative on my part, I know very little about most of these topics*
In Seveneves Neal Stephenson does the classic sci-fi trick of assuming that exactly one thing in the universe is different, and seeing where that take... | https://www.lesswrong.com/posts/9cgpXAZiuQShm8BcM/surviving-seveneves |
# Beyond the Board: Exploring AI Robustness Through Go
Last year, we showed that supposedly superhuman Go AIs can be beaten by human amateurs playing specific “cyclic” patterns on the board. Vulnerabilities have previously been observed in a wide variety of sub- or near-human AI systems, but this result demonstrates t... | https://www.lesswrong.com/posts/2iALhBnMpcjbWJJ3w/beyond-the-board-exploring-ai-robustness-through-go |
# Ilya Sutskever created a new AGI startup
[copy of the whole text of the announcement on ssi.inc, not an endorsement]
Safe Superintelligence Inc.
Superintelligence is within reach.
Building safe superintelligence (SSI) is the most important technical problem of our time.
We have started the world’s first straig... | https://www.lesswrong.com/posts/oeZ93QTv39TeB94Wt/ilya-sutskever-created-a-new-agi-startup |
# Chronic perfectionism through the eyes of school reports
A short post for a short thought. I was sifting through old school reports, and there's only one comment which is repeated in pretty much every report from 3 years old all the way through to 16.
**3 years old: "***Has a hard time dealing with disappointments.... | https://www.lesswrong.com/posts/YznaDZZ9axbYfCWDs/chronic-perfectionism-through-the-eyes-of-school-reports |
# What are things you're allowed to do as a startup?
The internet has previously seen [Things You're Allowed To Do](https://milan.cvitkovic.net/writing/things_youre_allowed_to_do/) (as an adult) and [Things You're Allowed To Do: University Edition](https://www.brasstacks.blog/tyatd-uni/). The idea is to list things th... | https://www.lesswrong.com/posts/myr9XD3nsZW3okDRu/what-are-things-you-re-allowed-to-do-as-a-startup |
# Actually, Power Plants May Be an AI Training Bottleneck.
There have been presistent rumors that electricity generation was somehow bottlenecking new data centers. This claim was [recently repeated by Donald Trump](https://youtu.be/xrFdHO7FH8w), who implied that San Francisco donors requested the construction of new ... | https://www.lesswrong.com/posts/DsoqEcnCu8vQeeeBe/actually-power-plants-may-be-an-ai-training-bottleneck |
# AI #69: Nice
Nice job breaking it, hero, unfortunately. Ilya Sutskever, despite what I sincerely believe are the best of intentions, has decided to be the latest to do The Worst Possible Thing, founding a new AI company explicitly looking to build ASI (superintelligence). The twists are zero products with a ‘cracked... | https://www.lesswrong.com/posts/ytFLs37zLsFBqLHGA/ai-69-nice |
# Case studies on social-welfare-based standards in various industries
Last year, I [posted a call for case studies](https://forum.effectivealtruism.org/posts/idrBxfsHkYeTtpm2q/seeking-paid-case-studies-on-standards) on social-welfare-based standards for companies and products (including standards imposed by regulatio... | https://www.lesswrong.com/posts/gAuYn49Cm8BPjewzf/case-studies-on-social-welfare-based-standards-in-various |
# Jailbreak steering generalization
*This work was performed as part of* [*SPAR*](https://supervisedprogramforalignment.org)
We use activation steering ([Turner et al., 2023](https://arxiv.org/abs/2308.10248); [Panickssery et al., 2023](https://arxiv.org/abs/2312.06681)) to investigate whether different types of jai... | https://www.lesswrong.com/posts/pZmFcKWZ4dXJaPPPa/jailbreak-steering-generalization |
# Claude 3.5 Sonnet
> we’ll be releasing Claude 3.5 Haiku and Claude 3.5 Opus later this year.
* * *
They made a [mini model card](https://www-cdn.anthropic.com/fed9cc193a14b84131812372d8d5857f8f304c5... | https://www.lesswrong.com/posts/xyuZcijPfjBa5qZDw/claude-3-5-sonnet |
# Interpreting and Steering Features in Images
*We trained a SAE to find sparse features in image embeddings. We found many meaningful, interpretable, and steerable features. We find that steering image di... | https://www.lesswrong.com/posts/Quqekpvx8BGMMcaem/interpreting-and-steering-features-in-images |
# Evaporation of improvements
I don't know if there is a standard way to express the idea that "if something seems like an obvious improvement of your situation, the advantage will probably (almost) disappear after everything settles in a *new balance*".
Examples:
* people make more money (yay)... inflation happen... | https://www.lesswrong.com/posts/7azdvoX656SSJYgkJ/evaporation-of-improvements |
# Debate, Oracles, and Obfuscated Arguments
This post is about recent and ongoing work on the power and limits of debate from the computational complexity point of view. As a starting point our paper [Scalable AI Safety via Doubly-Efficient Debate](https://arxiv.org/abs/2311.14125) gives new complexity-theoretic forma... | https://www.lesswrong.com/posts/DGt9mJNKcfqiesYFZ/debate-oracles-and-obfuscated-arguments-3 |
# Enriched tab is now the default LW Frontpage experience for logged-in users
In the past few months, the LessWrong team has been making use of the latest AI tools (given that they unfortunately exist[^kxerb0a2vb]) for [art](https://www.lesswrong.com/leastwrong), [music](https://www.lesswrong.com/posts/YMo5PuXnZDwRjhH... | https://www.lesswrong.com/posts/TDMKch5qzuaac5LFF/enriched-tab-is-now-the-default-lw-frontpage-experience-for |
# "... than average" is (almost) meaningless
Recently, I was talking to a friend who hadn't seen me in a while. They mentioned that my hair had grown noticeably, and then asked whether my hair grew fast or slow. I said that my hair growth was probably around average, but upon consideration, I realized that statement w... | https://www.lesswrong.com/posts/4ZBDNKKno7k7dYxcf/than-average-is-almost-meaningless |
# Attention Output SAEs Improve Circuit Analysis
*This is the final post of our Alignment Forum sequence produced as part of the*[* ML Alignment & Theory Scholars Program*](https://www.matsprogram.org/) *\- Winter 2023-24 Cohort.*
Executive Summary
-----------------
* In a previous post we trained [Attention Outpu... | https://www.lesswrong.com/posts/EGvtgB7ctifzxZg6v/attention-output-saes-improve-circuit-analysis |
# On OpenAI’s Model Spec
There are multiple excellent reasons to [publish a Model Spec](https://openai.com/index/introducing-the-model-spec/) like [OpenAI](https://cdn.openai.com/spec/model-spec-2024-05-08.html)’s, that specifies how you want your model to respond in various potential situations.
1. It lets us have ... | https://www.lesswrong.com/posts/mQmEQQLk7kFEENQ3W/on-openai-s-model-spec |
# Connecting the Dots: LLMs can Infer & Verbalize Latent Structure from Training Data
**TL;DR:** We published a new paper on out-of-context reasoning in LLMs. We show that LLMs can infer latent information from training data and use this information for downstream tasks, without any in-context learning or CoT. For ins... | https://www.lesswrong.com/posts/5SKRHQEFr8wYQHYkx/connecting-the-dots-llms-can-infer-and-verbalize-latent |
# AI governance needs a theory of victory
*This post is part of a series by*[* Convergence Analysis*](https://www.convergenceanalysis.org/)*. An earlier post introduced* [*scenario planning*](https://forum.effectivealtruism.org/posts/tCq2fi6vhSsCDA5Js/scenario-planning-for-ai-x-risk)* for AI risk. In this post, we arg... | https://www.lesswrong.com/posts/cuAFeLmquyEm3X6s5/ai-governance-needs-a-theory-of-victory-1 |
# Nuclear War, Map and Territory, Values | Guild of the Rose Newsletter, May 2024
### **Community Updates**
Upcoming events:
* **Naturalism** event [**every week on Thursdays**](https://calendar.google.com/calendar/event?action=TEMPLATE&tmeid=NzEyOGgzcGFjdXIzaWs1aTU1amZzMmU4ZjVfMjAyNDA1MDJUMTkwMDAwWiBjXzU0NnNjYWFq... | https://www.lesswrong.com/posts/YPanDKKNd74gdnSdp/nuclear-war-map-and-territory-values-or-guild-of-the-rose |
# What distinguishes "early", "mid" and "end" games?
Recently William_S posted:
> In my mental model, we're still in the mid-game, not yet in the end-game.
I replied:
> A thing I've been thinking about lately is "what does it mean to shift from the early-to-mid-to-late game".
> In strategy board games, there's an... | https://www.lesswrong.com/posts/j2pKBBvyAxHPNbuS6/what-distinguishes-early-mid-and-end-games |
# Juneberry Puffs
I only learned [about Juneberries](https://www.jefftk.com/p/june-and-mulberries) a few years ago, and they're a bit tricky. Eaten fresh, right off the tree, they have a bunch of subtle flavors that remind me of peach ice tea. If I [bake them](https://www.jefftk.com/p/juneberry-cake), though, these fl... | https://www.lesswrong.com/posts/iyFXGh8amLKufqSs3/juneberry-puffs |
# No really, the Sticker Shortcut fallacy is indeed a fallacy
In [a recent post](https://www.ymeskhout.com/p/sticke... | https://www.lesswrong.com/posts/wXZhoh54CDA8AohGT/no-really-the-sticker-shortcut-fallacy-is-indeed-a-fallacy |
# Best-of-n with misaligned reward models for Math reasoning
*In this post, I share some quick results that I got as part of a bigger project that pivoted.*
TL;DR:
* When trying to generate Math reasoning with current LLMs, optimizing too hard against their judgment of what is “good reasoning” can make the proport... | https://www.lesswrong.com/posts/iJjFRrGQkCxaqKrEo/best-of-n-with-misaligned-reward-models-for-math-reasoning |
# Appraising aggregativism and utilitarianism
*“My problem is: What are those objects we are adding up? I have no objection to adding them up if there's something to add.” — Kenneth Arrow*
1\. Introduction
================
Aggregative principles state that a social planner should make decisions as if they will face ... | https://www.lesswrong.com/posts/jQfzdCka8gcAsqAZJ/appraising-aggregativism-and-utilitarianism |
# Bed Time Quests & Dinner Games for 3-5 year olds
I like these games because they are playful, engage the child and still achieve the objective of getting the child to bed/eat dinner etc. Requires creativity and some slack.
Excerpt from Shohannah's post:
> Recently I had the bright idea to give up on being a regul... | https://www.lesswrong.com/posts/HNuJe83z86jmy3bNC/bed-time-quests-and-dinner-games-for-3-5-year-olds |
# Applying Force to the Wrong End of a Causal Chain
A causal chain, of course.
The... | https://www.lesswrong.com/posts/4F4Wenko2ihgGJHhb/applying-force-to-the-wrong-end-of-a-causal-chain |
# Karma votes: blind to or accounting for score?
Normally I try to use karma votes in a practical way to try to help promote content that I want to see more of on LW, or to downrank content that I want to see less of. (As opposed to using them as a kind of social signal, which is a separate issue.)
One thing I have n... | https://www.lesswrong.com/posts/fdpMQQfRfkR4qSMNv/karma-votes-blind-to-or-accounting-for-score |
# Population ethics and the value of variety
Problems in population ethics (are 2 lives at 2 utility better than 1 life at 3 utility?) are similar to problems about lifespan of a single person (is it better to live 2 years with 2 utility per year than 1 year with 3 utility per year?)
On the surface, this analogy se... | https://www.lesswrong.com/posts/6hQmaeZNaFtqhmwbs/population-ethics-and-the-value-of-variety |
# Text Posts from the Kids Group: 2019
Another round of liberating kid posts from [Facebook](https://www.jefftk.com/p/making-groups-for-kid-pictures). For reference, in 2019 Lily turned 5 and Anna turned 3.
(Some of these were from me; some were from Julia. Ones saying "me" could mean either of us.)
[2019-01-05](htt... | https://www.lesswrong.com/posts/prNZuwcLfnAvguiQf/text-posts-from-the-kids-group-2019 |
# Podcasts: AGI Show, Consistently Candid, London Futurists
For those of you who enjoy learning things via listening in on numerous slightly different conversations about them, and who also want to learn more about this [AI survey](https://blog.aiimpacts.org/p/2023-ai-survey-of-2778-six-things) I led, three more podca... | https://www.lesswrong.com/posts/4aoCfswbJFa38zitw/podcasts-agi-show-consistently-candid-london-futurists |
# How are you preparing for the possibility of an AI bust?
What actions have you been taking to prepare for the possibility that the AI industry will experience a market crash, something along the lines of the dotcom bust of the early 2000s? Also, what actions would you take if a crash like that occurred? For example:... | https://www.lesswrong.com/posts/RpuX6hpSZqHdyyMT6/how-are-you-preparing-for-the-possibility-of-an-ai-bust |
# "On the Impossibility of Superintelligent Rubik’s Cube Solvers", Claude 2024 [humor]
> In recent years, a number of prominent computer scientists and roboticists have suggested that artificial intelligence may one day solve Rubik's Cubes faster than humans. Many have further argued that AI could even come to exceed ... | https://www.lesswrong.com/posts/k7Qe8wrha4QAKhuky/on-the-impossibility-of-superintelligent-rubik-s-cube |
# Control Vectors as Dispositional Traits
I have been [reading recently](https://vgel.me/posts/representation-engineering/) about a technique that can be used to *partially control* the behaviour of Large Language Models: the technique is exploiting *control vectors*[^otlbo8clkho] to alter the activation patterns of t... | https://www.lesswrong.com/posts/Bf3ryxiM6Gff2zamw/control-vectors-as-dispositional-traits |
# AI Labs Wouldn't be Convicted of Treason or Sedition
This is a shortened version of "[Preventing AI from Overthrowing the Government](https://open.substack.com/pub/matthewkhoriaty/p/preventing-ai-from-overthrowing-the?r=2akg2p&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true)" from my [Substack](https://matt... | https://www.lesswrong.com/posts/vABADbprinFmaLceq/ai-labs-wouldn-t-be-convicted-of-treason-or-sedition |
# Talk: AI safety fieldbuilding at MATS
I recently gave a talk to the [AI Alignment Network (ALIGN)](https://www.aialign.net/) in Japan on my priorities for AI safety fieldbuilding based on my experiences at [MATS](https://www.matsprogram.org/) and [LISA](https://www.safeai.org.uk/) ([slides](https://docs.google.com/p... | https://www.lesswrong.com/posts/WGNYAdBrsNjujJaB8/talk-ai-safety-fieldbuilding-at-mats |
# Different senses in which two AIs can be “the same”
Sometimes people talk about two AIs being “the same” or “different” AIs. We think the intuitive binary of “same vs. different” conflates several concepts which are often better to disambiguate. In this post, we spell out some of these distinctions. We don’t think a... | https://www.lesswrong.com/posts/4j6HJt8Exowmqp245/different-senses-in-which-two-ais-can-be-the-same |
# Sci-Fi books micro-reviews
I've recently been reading a lot of science fiction. Most won't be original to fans of the genre, but some people might be looking for suggestions, so in lieu of full blown reviews here's super brief ratings on all of them. I might keep this updated over time, if so new books will go to th... | https://www.lesswrong.com/posts/RfFpMMqteqHbMz97n/sci-fi-books-micro-reviews |
# The Living Planet Index: A Case Study in Statistical Pitfalls
In case you sometimes read an alarming headline about biodiversity loss, like [*Global wildlife populations have declined by 69% since 1970*](https://edition.cnn.com/2022/10/13/world/wwf-living-planet-report-2022-climate-intl-scli-scn/index.html) or [*R... | https://www.lesswrong.com/posts/nCdxosAzGw9PKxRcD/the-living-planet-index-a-case-study-in-statistical-pitfalls |
# Book Review: Righteous Victims - A History of the Zionist-Arab Conflict
*I originally entered this to the ACX Book Review competition. Since it has not been selected as a finalist I'm now free to post it here. In truth it's a followup to* [*my review of Morris's history of Israel's War of Independence.*](https://www... | https://www.lesswrong.com/posts/Th4SeayGQyF6pYmZ6/book-review-righteous-victims-a-history-of-the-zionist-arab-1 |
# On Claude 3.5 Sonnet
There is a new clear best (non-tiny) LLM.
If you want to converse with an LLM, the correct answer is Claude Sonnet 3.5.
It is available for free on Claude.ai and the Claude iOS app, or you can subscribe for higher rate limits. The API cost is $3 per million input tokens and $15 per million out... | https://www.lesswrong.com/posts/wx4RhFzLbiHoShFjR/on-claude-3-5-sonnet |
# LLM Generality is a Timeline Crux
Four-Month Update
-----------------
**\[EDIT: I believe that** [**this paper**](https://arxiv.org/abs/2409.13373) **looking at o1-preview, which gets much better results on both blocksworld and obfuscated blocksworld, should update us significantly toward LLMs being capable of gene... | https://www.lesswrong.com/posts/k38sJNLk7YbJA72ST/llm-generality-is-a-timeline-crux |
# So you want to work on technical AI safety
I’ve been to two EAGx events and one EAG, and the vast majority of my one on ones with junior people end up covering some subset of these questions. I’m happy to have such conversations, but hopefully this is more efficient and wide-reaching (and more than I could fit into ... | https://www.lesswrong.com/posts/kGt3ukLR924kyfn5y/so-you-want-to-work-on-technical-ai-safety |
# SAE feature geometry is outside the superposition hypothesis
*Written at Apollo Research*
Summary: Superposition-based interpretations of neural network activation spaces are incomplete. The specific locations of feature vectors contain crucial structural information beyond superposition, as seen in circular arrang... | https://www.lesswrong.com/posts/MFBTjb2qf3ziWmzz6/sae-feature-geometry-is-outside-the-superposition-hypothesis |
# PSA: Consider alternatives to AUROC when reporting classifier metrics for alignment
TL;DR
=====
If you’re presenting a classifier that detects misalignment and providing metrics for it, please:
1. report the TPR at FPR=0.001, 0.01, and 0.05
2. plot the ROC curve on a log-log scale
See [https://arxiv.org/abs/211... | https://www.lesswrong.com/posts/StydMSLziGBn5gFAP/psa-consider-alternatives-to-auroc-when-reporting-classifier |
# Sparse Features Through Time
This project explores the use of Sparse Autoencoders (SAEs) to track the development of features in large language models throughout their training. It investigates whether features can be reliably matched between different SAEs trained on various checkpoints of Pythia 70M and characteri... | https://www.lesswrong.com/posts/piteaGopyWMD7ozSf/sparse-features-through-time |
# Compact Proofs of Model Performance via Mechanistic Interpretability
We recently released a [paper](https://arxiv.org/abs/2406.11779) on using mechanistic interpretability to generate compact formal guarantees on model performance. In this companion blog post to our paper, we'll summarize the paper and flesh out som... | https://www.lesswrong.com/posts/bRsKimQcPTX3tNNJZ/compact-proofs-of-model-performance-via-mechanistic |
# The Minority Coalition
*Hello everybody. Or maybe nobody. I don’t know yet if I’m going to release this stream, I could get in pretty hot water for it. But you guys know that hasn’t stopped me in the past. The backstory this time is that I’ve managed to sign up for one of the red-teaming programs where they test unr... | https://www.lesswrong.com/posts/vSSrAbbE8RtowRSBZ/the-minority-coalition |
# A Basic Economics-Style Model of AI Existential Risk
*Crossposted with my new blog,* [*Crossing the Rubicon*](https://crossingtherubicon.substack.com/p/a-basic-model-of-existential-risk)*, and primarily aimed at x-risk skeptics from economics backgrounds. If you're interested in novel takes on theoretical AI safety,... | https://www.lesswrong.com/posts/8vXeyxnXAGKgBAS9M/a-basic-economics-style-model-of-ai-existential-risk |
# Being hella lost as rationality practice
I've had the privilege of knowing where I was only in the sense of "within this quadrangle of an insufficiently detailed map, with impassable terrain not well indicated" several times. Usually by choice. More often while alone than prudence would dictate. I think they've been... | https://www.lesswrong.com/posts/hSeQmkuAwcDLNinM7/being-hella-lost-as-rationality-practice |
# I'm a bit skeptical of AlphaFold 3
(also on [https://olegtrott.substack.com](https://olegtrott.substack.com))
So this happened: DeepMind (with 48 authors, including a new member of the British nobility) decided to compete with me. Or rather, with some of my work from 10+ years ago.
Apparently, AlphaFold 3 can now ... | https://www.lesswrong.com/posts/D9yYmL6KPq7dcNSKE/i-m-a-bit-skeptical-of-alphafold-3 |
# Higher-effort summer solstice: What if we used AI (i.e., Angel Island)?
*As the title probably already indicates, this post contains community content rather than rationality content.* [*Alternate, sillier version of this post here.*](https://rachelshu.notion.site/Proposal-for-summer-solstice-2025-f14a35fa898541b794... | https://www.lesswrong.com/posts/cZqNiRd89A92rBPkc/higher-effort-summer-solstice-what-if-we-used-ai-i-e-angel |
# Mistakes people make when thinking about units
*This is a linkpost for* [*Parker Dimensional Analysis*](https://outsidetheasylum.blog//parker-dimensional-analysis/)*. Probably a little elementary for LessWrong, but I think it may still contain a few novel insights, particularly in the last section about Verison's er... | https://www.lesswrong.com/posts/5vfSNLb92eyXKkQax/mistakes-people-make-when-thinking-about-units |
# Monthly Roundup #19: June 2024
Looks like we made it. Yes, the non-AI world still exists.
#### Bad Governor
[New York Governor Kathy Hochul has gone rogue and betrayed New York City](https://gothamist.com/news/gov-hochul-says-conversations-at-3-nyc-diners-changed-her-mind-about-congestion-pricing-we-investigated),... | https://www.lesswrong.com/posts/7LvK6Gw2GdfDMBNNm/monthly-roundup-19-june-2024 |
# Formal verification, heuristic explanations and surprise accounting
ARC's current research focus can be thought of as trying to combine mechanistic interpretability and formal verification. If we had a deep understanding of what was going on inside a neural network, we would hope to be able to use that understanding... | https://www.lesswrong.com/posts/SyeQjjBoEC48MvnQC/formal-verification-heuristic-explanations-and-surprise |
# An Intuitive Explanation of Sparse Autoencoders for Mechanistic Interpretability of LLMs
Sparse Autoencoders (SAEs) have recently become popular for interpretability of machine learning models (although SAEs have been around since [1997](https://www.sciencedirect.com/science/article/pii/S0042698997001697)). Machine ... | https://www.lesswrong.com/posts/CJPqwXoFtgkKPRay8/an-intuitive-explanation-of-sparse-autoencoders-for |
# Incentive Learning vs Dead Sea Salt Experiment
*(Target audience: People both inside and outside neuroscience & psychology. I tried to avoid jargon.)*
1\. Background context
======================
One of my interests is the neuroscience of motivation. This is a topic I care about for the same obvious reason as eve... | https://www.lesswrong.com/posts/YQ4rSTHpHeFcAmhvi/incentive-learning-vs-dead-sea-salt-experiment |
# Failure Modes of Teaching AI Safety
### **Why I'm writing this**
I'm about to teach [my AI safety course](https://www.lesswrong.com/posts/t3ngnd6Wvo4qeY5FA/i-designed-an-ai-safety-course-for-a-philosophy-department) for the fourth time. As I'm now updating the syllabus for the upcoming semester, I summarize my obse... | https://www.lesswrong.com/posts/XhywWXDLkqDSB7KjH/failure-modes-of-teaching-ai-safety-1 |
# Computational Complexity as an Intuition Pump for LLM Generality
With sufficient scale and scaffolding, LLMs will improve without bound on all tasks to become superhuman AGI, to the extent they haven’t already. No, wait! LLMs are dead-end pattern-matching machines fundamentally incapable of general reasoning and n... | https://www.lesswrong.com/posts/HxHYPJoaWD8AjaHCk/computational-complexity-as-an-intuition-pump-for-llm |
# Compute Governance Literature Review
This white paper reviews the use of compute governance to mitigate risks from AI, the reasoning behind this approach, the ways in which it has been used and in which it could be further developed. It aims to provide an introduction for individuals who feel that there is a need f... | https://www.lesswrong.com/posts/eLzDLCB68qNoWDRba/compute-governance-literature-review |
# What is a Tool?
Throughout this post, we’re going to follow the Cognition -> Convergence -> Corroboration methodology[^6210tr4tg3w]. That means we’ll tackle tool-ness in three main stages, each building on the previous:
* Cognition: What does it mean, cognitively, to view or model something as a tool?
* Converg... | https://www.lesswrong.com/posts/ajJRyKtwZNnBdmkcv/what-is-a-tool |
# In favour of exploring nagging doubts about x-risk
I was chatting recently to someone who had difficulty knowing how to orient to x-risk work (despite being a respected professional working in that field). They expressed that they didn't find it motivating at a gut level in the same way they did with poverty or anim... | https://www.lesswrong.com/posts/4eBLj3TTcypq4ic9v/in-favour-of-exploring-nagging-doubts-about-x-risk |
# My Current Claims and Cruxes on LLM Forecasting & Epistemics
I think that recent improvements in LLMs have brought us to the point where LLM epistemic systems are starting to be useful. After spending some time thinking about it, I've realized that such systems, broadly, seem very promising to me as an effective alt... | https://www.lesswrong.com/posts/QvFRAEsGv5fEhdH3Q/preliminary-notes-on-llm-forecasting-and-epistemics |
# Empirical vs. Mathematical Joints of Nature
We (Alex and Elizabeth) are thinking about doing this project where we figure out how paradigm formation happened in chaos theory. Alex has also been thinking about paradigm formation for agent foundations (which people often talk about as being pre-paradigmatic). These ar... | https://www.lesswrong.com/posts/jhhoks5psG8Dcpgzu/empirical-vs-mathematical-joints-of-nature |
# Childhood and Education Roundup #6: College Edition
Childhood roundup #5 excluded all developments around college. So this time around is all about issues related to college or graduate school, including admissions.
#### Tuition and Costs
[What went wrong](https://marginalrevolution.com/marginalrevolution/2024/05/... | https://www.lesswrong.com/posts/pn5jWW4zcWSAjM9s3/childhood-and-education-roundup-6-college-edition |
# Schelling points in the AGI policy space
I've been thinking about memetically fit Schelling points in the AGI policy space. I'll describe four such "Schelling policies", and use them as pedagogical examples.
# Shut it all down
MIRI's new stated objective is the clearest example of a Schelling policy: ["Shut it... | https://www.lesswrong.com/posts/JECQZAXWbtGJdBuAC/schelling-points-in-the-agi-policy-space |
# Progress Conference 2024: Toward Abundant Futures
The progress movement has grown a lot in the last few years. We now have progress journals, think tanks, and fe... | https://www.lesswrong.com/posts/cyDFZgSS33XrcehhD/progress-conference-2024-toward-abundant-futures |
# Countering AI disinformation and deep fakes with digital signatures
According to [The Economist](https://www.economist.com/science-and-technology/2024/05/10/a-russia-linked-network-uses-ai-to-rewrite-real-news-stories), disinformation campaigns (often state-sponsored) use "AI to rewrite real news stories":
> In ear... | https://www.lesswrong.com/posts/74GXwwZWAHA9GjxGz/countering-ai-disinformation-and-deep-fakes-with-digital |
# Imbue (Generally Intelligent) continue to make progress
I've been following the company Imbue and their podcast Generally Intelligent since they started. They've said thoughtful and creative things in their podcast, and I think they are making impressive progress towards AGI considering their relatively smaller size... | https://www.lesswrong.com/posts/KMorqWKG8XyXky3ZB/imbue-generally-intelligent-continue-to-make-progress |
# Instrumental vs Terminal Desiderata
Bob: "I want my AGI to make [everyone extremely wealthy](https://lukemuehlhauser.com/three-wild-speculations-from-amateur-quantitative-macrohistory/)! I'm going to train that to be its goal."
Cassie: "Stop! You'll doom us all! While wealth is good, it's not *everything* that's go... | https://www.lesswrong.com/posts/WKrcJ22QKjg6DnEdb/instrumental-vs-terminal-desiderata |
# Live Theory Part 0: Taking Intelligence Seriously
### Acknowledgements
*The vision here was midwifed originally in the wild and gentle radiance that is* [*Abram*](https://www.lesswrong.com/users/abramdemski)*'s company (though essentially none of the content is explicitly his). *
*The* [*PIBBSS*](https://www.lessw... | https://www.lesswrong.com/posts/QvnzEHvodmwfBXu94/live-theory-part-0-taking-intelligence-seriously |
# Cross Robin
A common contra dance figure is "right shoulder round": walk around the opposite-role person on the side of the set while looking at them. A less common figure is the "mad robin": walk the same path, but instead look at the opposite-role person across the set. This suggests a third option, which we can c... | https://www.lesswrong.com/posts/gztJxTLvdfAxu7aRz/cross-robin |
# Detecting Genetically Engineered Viruses With Metagenomic Sequencing
*This represents work from several people at the NAO. Thanks especially to Dan Rice for implementing the duplicate junction detection, and to* [*@Will Bradshaw*](https://forum.effectivealtruism.org/users/willbradshaw?mention=user) *and* [*@mike_mcl... | https://www.lesswrong.com/posts/iaPhjYhhp7PP6BWp9/detecting-genetically-engineered-viruses-with-metagenomic |
# AI #70: A Beautiful Sonnet
They said it couldn’t be done.
No, not [Claude Sonnet 3.5 becoming the clear best model](https://thezvi.substack.com/p/on-claude-35-sonnet).
No, not the Claude-Sonnet-empowered automatic meme generators. Those were whipped together in five minutes.
They said I would never get quiet time... | https://www.lesswrong.com/posts/rC3hhZsx2KogoPLqh/ai-70-a-beautiful-sonnet |
# An issue with training schemers with supervised fine-tuning
*This is a rough write-up of ideas by Ryan Greenblatt. The core ideas are his and mistakes are mine. Thanks to him, Buck Shlegeris and to Aghyad Deeb for feedback on the draft of this post.*
One way to use [scheming AIs](https://www.lesswrong.com/posts/yFo... | https://www.lesswrong.com/posts/mXYdYh6L9odTJZDSm/an-issue-with-training-schemers-with-supervised-fine-tuning |
# Representation Tuning
**Summary**
-----------
First, I identify activation vectors related to honesty in an RLHF’d LLM ([Llama-2-13b-chat](https://huggingface.co/meta-llama/Llama-2-13b-chat-hf)). Next, I demonstrate that model output can be made more or less honest by adding positive or negative multiples of these ... | https://www.lesswrong.com/posts/T9i9gX58ZckHx6syw/representation-tuning |
# how birds sense magnetic fields
## introduction
It is known that many birds are able to sense the direction of Earth's magnetic field. [Here's a wikipedia page](https://en.wikipedia.org/wiki/Magnetoreception) on that general phenomenon. There have been 2 main theories of how that works.
One theory is that birds ha... | https://www.lesswrong.com/posts/AZnF5LNZfeGZRGvid/how-birds-sense-magnetic-fields |
# Distillation of 'Do language models plan for future tokens'
Link to arxiv preprint: [Do language models plan for future tokens, by Wilson Wu, John X Morris and Lionel Levine.](https://arxiv.org/pdf/2404.00859)
TLDR
====
* There are two reasons why the computations for the current token are helpful for future tok... | https://www.lesswrong.com/posts/KoCS6LqJjKhCqDqnX/distillation-of-do-language-models-plan-for-future-tokens |
# Secondary forces of debt
A general thing I hadn’t noticed about debts until lately:
* Whenever Bob owes Alice, then Alice has reason to look after Bob, to the extent that increases the chance he satisfies the debt.
* Yet at the same time, Bob has an incentive for Alice to disappear, insofar as it would relieve ... | https://www.lesswrong.com/posts/WZ2Xug4j3rz2Pe3D2/secondary-forces-of-debt |
# How Big a Deal are MatMul-Free Transformers?
*If you’re already familiar with the technical side of LLMs, you can skip the first section.*
The story so far
----------------
Modern Large Language Models - your ChatGPTs, your Geminis - are a particular kind of *transformer*, a deep learning architecture invented abo... | https://www.lesswrong.com/posts/REzKbk9reKFvgFkmf/how-big-a-deal-are-matmul-free-transformers |
# Toward a taxonomy of cognitive benchmarks for agentic AGIs
Inspired by the [sequence on LLM Psychology](https://www.lesswrong.com/posts/yuwdj82yjhLFYessc/preface-to-the-sequence-on-llm-psychology), I am developing a taxonomy of cognitive benchmarks for measuring intelligent behavior in LLMs. This taxonomy could faci... | https://www.lesswrong.com/posts/GHEyrHrcpz3T4CumG/toward-a-taxonomy-of-cognitive-benchmarks-for-agentic-agis |
# Situational Awareness
Nearly a book review: [Situational
Awareness](https://situational-awareness.ai/), by Leopold Aschenbrenner.
"Situational Awareness" offers an insightful analysis of our proximity
to a critical threshold in AI capabilities. His background in machine
learning and economics lends credibility... | https://www.lesswrong.com/posts/6evBdZAiT9yxoJBKk/situational-awareness-8 |
# Corrigibility = Tool-ness?
Goal of This Post
-----------------
I have never seen anyone give a satisfying intuitive explanation of what corrigibility (in roughly Eliezer’s sense of the word) is. There’s [lists of desiderata](https://www.lesswrong.com/posts/AqsjZwxHNqH64C2b6/let-s-see-you-write-that-corrigibility-ta... | https://www.lesswrong.com/posts/7LaDvWtymFWtidGxe/corrigibility-tool-ness |
# The Wisdom of Living for 200 Years
[Scott Sumner writes](https://www.themoneyillusion.com/double-vision-when-then-was-now/):
> Old people have a sort of double vision about the past — an ability to see the past from the perspective of today, and also from the perspective of the people who lived through those times... | https://www.lesswrong.com/posts/dxBiq5LpPwtirDTMY/the-wisdom-of-living-for-200-years |
# How a chip is designed
*Disclaimer: This is highly incomplete. I am not an expert in the field. There might be some unfamiliar terms. While I will try to explain things, explaining every single term would be beyond this post. You will usually be able to get a sufficient understanding by clicking the links or googlin... | https://www.lesswrong.com/posts/cruYtDoJuDXnkaPxR/how-a-chip-is-designed |
# Five toy worlds to think about heritability
Just a quick note from a [Twitter/X discussion](https://x.com/davidhughjones/status/1804601064060293207).
It’s an almost-universal scientific rule that empirics is blind without a model. The left picture below shows a bunch of points, and a regression line fitting them. ... | https://www.lesswrong.com/posts/zrbq6zHsQwrE8cLXJ/five-toy-worlds-to-think-about-heritability |
# Contra Acemoglu on AI
[The Simple Macroeconomics of AI](https://economics.mit.edu/sites/default/files/2024-05/The%20Simple%20Macroeconomics%20of%20AI.pdf) is a 2024 working paper by Daron Acemoglu which models the economic growth effects of AI and predicts them to be small: About a .06% increase in TFP growth annual... | https://www.lesswrong.com/posts/viRn7Drv9FKcdFpyX/contra-acemoglu-on-ai |
# The Incredible Fentanyl-Detecting Machine
[
](https://substackcdn.com/image/fetch/f_auto,q_aut... | https://www.lesswrong.com/posts/TzwMfRArgsNscHocX/the-incredible-fentanyl-detecting-machine |
# Podcast: Elizabeth & Austin on "What Manifold was allowed to do"
Elizabeth Van Nostrand and Austin Chen discuss unconventional approaches to running startups, nonprofits, and events. We go over:
* How publicly sharing Manifold’s salaries and company information helped with recruitment and user engagement
* The ... | https://www.lesswrong.com/posts/GHusbyeLEn45HaM5B/podcast-elizabeth-and-austin-on-what-manifold-was-allowed-to |
# Activation Pattern SVD: A proposal for SAE Interpretability
*Epistemic status: This is a rough-draft write-up about a thought experiment I did. Reasonably confident about the broad arguments being made here. That said, I haven't spent a lot of time rigorously polishing or reviewing my writing, so minor inaccuracies ... | https://www.lesswrong.com/posts/ZzT5rGbSCrNbASppZ/activation-pattern-svd-a-proposal-for-sae-interpretability |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.