text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Talent Needs of Technical AI Safety Teams
Co-Authors: [@yams](https://www.lesswrong.com/users/yams?mention=user), [@Carson Jones](https://www.lesswrong.com/users/carson-jones?mention=user), [@McKennaFitzgerald](https://www.lesswrong.com/users/mckennafitzgerald?mention=user), [@Ryan Kidd](https://www.lesswrong.com/us... | https://www.lesswrong.com/posts/QzQQvGJYDeaDE4Cfg/talent-needs-of-technical-ai-safety-teams |
# minutes from a human-alignment meeting
"OK, let's get this meeting started. We're all responsible for development of this new advanced intelligence 'John'. We want John to have some kids with our genes, instead of just doing stuff like philosophy or building model trains, and this meeting is to discuss how we can en... | https://www.lesswrong.com/posts/SN3BjoizdbvZG5J6a/minutes-from-a-human-alignment-meeting |
# The Schumer Report on AI (RTFB)
Or at least, Read [the Report](https://www.politico.com/f/?id=0000018f-79a9-d62d-ab9f-f9af975d0000) (RTFR).
There is no substitute. This is not strictly a bill, but it is important.
The introduction kicks off balancing upside and avoiding downside, utility and risk. This will be a c... | https://www.lesswrong.com/posts/wxTMxF35PkNawn8f9/the-schumer-report-on-ai-rtfb |
# AI companies aren't really using external evaluators
*From* [*my new blog: AI Lab Watch.*](https://ailabwatch.org/blog/external-evaluation/) *All posts will be crossposted to LessWrong.* [*Subscribe on Substack.*](https://ailabwatch.substack.com/)
Many AI safety folks think that [METR](https://metr.org) is close to... | https://www.lesswrong.com/posts/WjtnvndbsHxCnFNyc/ai-companies-aren-t-really-using-external-evaluators |
# NYU Code Debates Update/Postmortem
TL;DR
=====
We designed an ambitious scalable oversight experimental setup, where we had people with no coding/programming experience try to answer coding questions (“Which of these two outputs is the correct output of the given function on this input?”), using LLMs that debate or... | https://www.lesswrong.com/posts/wsXCXoyvRi3DnWZ2M/nyu-code-debates-update-postmortem |
# Computational Mechanics Hackathon (June 1 & 2)
Join our [Computational Mechanics Hackathon](https://www.apartresearch.com/event/compmech), organized with the support of [APART](https://www.apartresearch.com/event/compmech), [PIBBSS](https://pibbss.ai/) and [Simplex](https://simplexaisafety.com).
This is an oppo... | https://www.lesswrong.com/posts/tkEQKrqZ6PdYPCD8F/computational-mechanics-hackathon-june-1-and-2 |
# Secret US natsec project with intel revealed
Unclear how relevant this news is to AI safety, but it seems like the sort of thing we ought to notice.
> A backroom Washington deal brokered two years ago is undercutting a key part of President Joe Biden’s policy to grow the national high-tech manufacturing base — push... | https://www.lesswrong.com/posts/grvJay8Cv3TBhXz3a/secret-us-natsec-project-with-intel-revealed |
# What should the norms around AI voices be?
In previous discussions of AI risks, the ability for an AI to be very persuasive is often seen as one possible risk. Humans find some voices more persuasive than other voices.
If we can trust Scarlett Johansson's description of her interactions with OpenAI, OpenAI wanted t... | https://www.lesswrong.com/posts/czvyfaLf23vtvPhYJ/what-should-the-norms-around-ai-voices-be |
# Level up your spreadsheeting
Epistemic status: Passion project / domain I’m pretty opinionated about, just for fun.
In this post, I walk through some principles I think good spreadsheets abide by, and then in the [companion piece](https://docs.google.com/document/d/1cIHfQcRIV_wNF6IlqIoqY83NRDl4US012v1PtsTPo4s/edit)... | https://www.lesswrong.com/posts/RBtF9fu9WMjdvqHFB/level-up-your-spreadsheeting |
# Training-time domain authorization could be helpful for safety
This is a short high-level description of our work from AI Safety Camp and continued research on the ***training-time domain authorization*** research program, a conceptual introduction, and its implications for AI Safety.
TL;DR: No matter how safe mode... | https://www.lesswrong.com/posts/38avQYy782zXgNo9u/training-time-domain-authorization-could-be-helpful-for |
# Episode: Austin vs Linch on OpenAI
Austin (of Manifund) and Linch (of EA Funds) explore our disagreements on OpenAI. Austin is approximately an apologist for OpenAI, and for-profit/tech companies overall; while Linch takes a more hardline stance.
Neither of us are experts on OpenAI or have insider information; we’r... | https://www.lesswrong.com/posts/A7jiQkMGJKrB4qZHq/episode-austin-vs-linch-on-openai |
# Is CDT with precommitment enough?
Logical decision theory was introduced (in part) to resolve problems such as Parfit's hitchhiker.
I heard an argument that there is no reason to introduce a new decision theory - one can just take causal decision theory and precommit to doing whatever is needed on such problems (e.... | https://www.lesswrong.com/posts/vFQFHjm9aRvJGBcdk/is-cdt-with-precommitment-enough |
# Who does the artwork for LessWrong?
The art on the LessWrong site and books is amazing. The consistent theme across everything is beautiful!
hm... I just checked the physical book and it credits Midjourney. But I feel there must have been some serious prompt crafting/human artistry behind this?
Is there some huma... | https://www.lesswrong.com/posts/C5ggwetLJGfgm92Nr/who-does-the-artwork-for-lesswrong |
# The AI Revolution in Biology
An expert in the field ([Amelie Schreiber](https://x.com/amelie_iska)) predicts the convergence of AI models for biology to the state where designing drugs, targeted genetic mutations (which then could be enacted through genetic therapy), or viruses for specific purposes becomes really e... | https://www.lesswrong.com/posts/ceLgi34zvsd7BFJof/the-ai-revolution-in-biology |
# Disproving and partially fixing a fully homomorphic encryption scheme with perfect secrecy
Summary
=======
In [the last blog post](https://www.lesswrong.com/posts/TK8ptSJGvAqj2HaRr/planning-to-build-a-cryptographic-box-with-perfect-secrecy), I introduced my plan to make the safest cryptographic box in the world, an... | https://www.lesswrong.com/posts/wuESDsLmDYtMgAFeE/disproving-and-partially-fixing-a-fully-homomorphic |
# Notifications Received in 30 Minutes of Class
**Introduction**
----------------
If you are choosing to read this post, you've probably seen the image below depicting all the notifications students received on their phones during one class period. You probably saw it as a retweet of[ this tweet](https://x.com/SCforE... | https://www.lesswrong.com/posts/AZCpu3BrCFWuAENEd/notifications-received-in-30-minutes-of-class |
# Review: Conor Moreton's "Civilization & Cooperation"
*Author's note: These days, my thoughts go onto* [*my substack*](https://homosabiens.substack.com/) *instead of onto LessWrong. Everything I write becomes free after a week or so, but it’s only paid subscriptions that make it **possible** for me to write. If you f... | https://www.lesswrong.com/posts/5acACJQjnA7KAHNpT/review-conor-moreton-s-civilization-and-cooperation |
# If you are also the worst at politics
*This has been said many times but it must be said many more. Please excuse the fact that this post breaks its own advice.*
I am the worst at politics. I always make people dig their heels in on the opposite stance. I could give many examples. Even [here](https://www.lesswrong.... | https://www.lesswrong.com/posts/eQydRznuJaTCSYcKA/if-you-are-also-the-worst-at-politics |
# Луна Лавгуд и Комната Тайн, Часть 1
Disclaimer: This is [Kongo Landwalker](https://www.lesswrong.com/users/kongo-landwalker)'s translation of [**lsusr**](https://www.lesswrong.com/users/lsusr?from=post_header)'s fiction [Luna Lovegood and the Chamber of Secrets - Part 1](https://www.lesswrong.com/posts/zb3hWt99i9Fm9... | https://www.lesswrong.com/posts/xfMkrpNYFpznkym5S/luna-lavgud-i-komnata-tain-chast-1 |
# If trying to communicate about AI risks, make it vivid
*Crossposted from the* [*EA Forum*](https://forum.effectivealtruism.org/posts/Mjkzay6T4vjwuKjbm/if-trying-to-communicate-about-ai-risks-make-it-vivid)
If you want to communicate AI risks in a way that increases concern, [^lo2n9y5cnt][ our new study](https://pap... | https://www.lesswrong.com/posts/Rqknb5dMjXbgzAmsm/if-trying-to-communicate-about-ai-risks-make-it-vivid |
# Truthseeking is the ground in which other principles grow
Introduction
============
[*First they came for the epistemology/we don’t know what happened after that.*](https://x.com/HiFromMichaelV/status/1161174071469641728)
I’m fairly antagonistic towards the author of that tweet, but it still resonates deep in my s... | https://www.lesswrong.com/posts/kbnJHpapusMJZb6Gs/truthseeking-is-the-ground-in-which-other-principles-grow |
# Book review: Everything Is Predictable
Book review: Everything Is Predictable: How Bayesian Statistics Explain
Our World, by Tom Chivers.
Many have attempted to persuade the world to embrace a Bayesian
worldview, but none have succeeded in reaching a broad audience.
[E.T. Jaynes'
book](https://bayesianinves... | https://www.lesswrong.com/posts/DcEThyBPZfJvC5tpp/book-review-everything-is-predictable-1 |
# Being against involuntary death and being open to change are compatible
[In a new post, Nostalgebraist](https://nostalgebraist.tumblr.com/post/751581192465383424/its-been-a-long-time-since-ive-posted-much-of) argues that "AI doomerism has its roots in anti-deathist transhumanism", representing a break from the norma... | https://www.lesswrong.com/posts/tNi5CECBMAGa2N6sp/being-against-involuntary-death-and-being-open-to-change-are |
# Maybe Anthropic's Long-Term Benefit Trust is powerless
*Crossposted from* [*AI Lab Watch.*](https://ailabwatch.org/blog/anthropic-trust-powerless/) *Subscribe on* [*Substack.*](https://ailabwatch.substack.com/)
Introduction
------------
Anthropic has an unconventional governance mechanism: an independent "Long-Ter... | https://www.lesswrong.com/posts/sdCcsTt9hRpbX6obP/maybe-anthropic-s-long-term-benefit-trust-is-powerless |
# I am the Golden Gate Bridge
#### Easily Interpretable Summary of New Interpretability Paper
[Anthropic has identified](https://www.anthropic.com/research/mapping-mind-language-model) ([full paper here](https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html)) how millions of concepts are represente... | https://www.lesswrong.com/posts/JdcxDEqWKfsucxYrk/i-am-the-golden-gate-bridge |
# Overview of introductory resources in AI Governance
Overview of introductory resources in AI Governance
*This post was created as part of the* [*Supervised Program for Alignment Research*](https://supervisedprogramforalignment.org/)*, Spring 2024. This work would not have happened without the encouragement and acco... | https://www.lesswrong.com/posts/xsXbbYX3qvy4McXof/overview-of-introductory-resources-in-ai-governance-1 |
# Intransitive Trust
I.
--
"Transitivity" is a property in mathematics and logic. Put simply, if something is transitive it means that there's a relationship between things where when x relates to y, and y relates to z, there's the same relationship between x and z. For a more concrete example, think of size. If my c... | https://www.lesswrong.com/posts/zKEdphnEycdCJeq8f/intransitive-trust |
# Understanding Gödel’s completeness theorem
In this post I prove a variant of Gödel's completeness theorem. My intention has been to really understand the theorem, so that I am not simply shuffling symbols around, but am actually understanding why it is true. I hope it is helpful for at least some other people.
For ... | https://www.lesswrong.com/posts/dPpA79MjPdDd87YoW/understanding-goedel-s-completeness-theorem |
# How to get nerds fascinated about mysterious chronic illness research?
Like many nerdy people, back when I was healthy, I was interested in subjects like math, programming, and philosophy. But 5 years ago I got sick with a viral illness and never recovered. For the last couple of years I've been spending most of my ... | https://www.lesswrong.com/posts/joPjaY43a4umyCkJK/how-to-get-nerds-fascinated-about-mysterious-chronic-illness |
# Finding Backward Chaining Circuits in Transformers Trained on Tree Search
This post is a summary of our paper *A Mechanistic Analysis of a Transformer Trained on a Symbolic Multi-Step Reasoning Task *(ACL 2024). While we wrote and released the paper a couple of months ago, we have done a bad job promoting it so far.... | https://www.lesswrong.com/posts/EBbcuSuNafkYpsgTW/finding-backward-chaining-circuits-in-transformers-trained-1 |
# 2024 State of the AI Regulatory Landscape
*As part of our *[***Governance Recommendations Research Program***](https://www.convergenceanalysis.org/programs/governance-research)*, *[***Convergence Analysis***](https://www.convergenceanalysis.org/)* has compiled a first-of-its-kind report summarizing the state of the... | https://www.lesswrong.com/posts/TkE7yJR4nD8ytHKxy/2024-state-of-the-ai-regulatory-landscape |
# OpenAI: Fallout
Previously: [OpenAI: Exodus](https://thezvi.substack.com/p/openai-exodus) (contains links at top to earlier episodes), [Do Not Mess With Scarlett Johansson](https://thezvi.substack.com/p/do-not-mess-with-scarlett-johansson)
We have learned more since last week. [It’s worse than we knew.](https://www... | https://www.lesswrong.com/posts/YwhgHwjaBDmjgswqZ/openai-fallout |
# Quick Advice on Writing Essays
Writing is difficult. Even writers with 20 years of experience will attest to this. But I recently heard some excellent writing advice from Saloni Dattani, and thought I should share it more widely.
But first, some context. Last September, Saloni wrote a piece about the history of mal... | https://www.lesswrong.com/posts/AYEyme87hxzgpuHYC/quick-advice-on-writing-essays |
# Reward hacking behavior can generalize across tasks
**TL;DR: **We find that reward hacking generalization occurs in LLMs in a number of experimental settings and can emerge from reward optimization on certain datasets. This suggests that when models exploit flaws in supervision during training, they can sometimes ge... | https://www.lesswrong.com/posts/Ge55vxEmKXunFFwoe/reward-hacking-behavior-can-generalize-across-tasks |
# What's a better term now that "AGI" is too vague?
The term "AGI" is now creating confusion. When it's used in the context of timelines or alignment, we don't know if it means near-future LLMs, or superintelligence. [It's fair to use AGI as "fairly general AI at near-human level,"](https://www.lesswrong.com/posts/gP8... | https://www.lesswrong.com/posts/HxnAFdSZWDFwGnfGN/what-s-a-better-term-now-that-agi-is-too-vague |
# Notes on Gracefulness
This post examines the virtues of **gracefulness**, **poise**, **composure**, **savoir-faire** and other things in that bailiwick. As with my other posts in this sequence, I’m less interested in breaking new ground and more in gathering and synthesizing whatever wisdom I could find on the subje... | https://www.lesswrong.com/posts/iofy4cWC9AWzZDtxc/notes-on-gracefulness |
# When Are Circular Definitions A Problem?
*Disclaimer: if you are using a definition in a nonmathematical piece of writing, you are probably making a mistake; you should just get rid of the definition and instead use a few examples. This applies double to people who think they are being "rigorous" by defining things ... | https://www.lesswrong.com/posts/dTtLmWFZprFJHsaaQ/when-are-circular-definitions-a-problem |
# Hardshipification
When I got cancer, all of my acquaintances turned into automatons. Everyone I had zero-to-low degrees of social contact with started reaching out, saying the exact same thing: *“If you need to talk to someone, I’m here for you”*. No matter how tenuous the connection, people pledged their emotional ... | https://www.lesswrong.com/posts/zSQXHgEqRKiZTXdKN/hardshipification |
# One way violinists fail
*Perfectionism and 'skill dysphoria' as obstacles to healthy contact with reality.*
> *"Fool! You cannot perceive my ✨sound✨ merely by listening to me play!"*
I'm currently in a phase of my violin career best described as "aspiring professional."
With me are the best few (several thousand)... | https://www.lesswrong.com/posts/hBYWE5mTwGjeeBjy9/one-way-violinists-fail |
# AI companies' commitments
*Crossposted from* [*AI Lab Watch.*](https://ailabwatch.org/blog/) *Subscribe on* [*Substack.*](https://ailabwatch.substack.com/)
AI Lab Watch has [a page on commitments](https://ailabwatch.org/resources/commitments/). It's the best page in the [Resources](https://ailabwatch.org/resources/... | https://www.lesswrong.com/posts/GvcTpyZmgAHuoCnLw/ai-companies-commitments |
# Looking beyond Everett in multiversal views of LLMs
Over the weekend I was reading up on some very fun exploratory thinking from years ago around large language models through the lens of a quantum multiverse which was extrapolating David Deutsch's parallel between the evolution of state in a quantum system and the ... | https://www.lesswrong.com/posts/bmsmiYhTm7QJHa2oF/looking-beyond-everett-in-multiversal-views-of-llms |
# Response to nostalgebraist: proudly waving my moral-antirealist battle flag
[@nostalgebraist](https://www.lesswrong.com/users/nostalgebraist?mention=user) has recently posted [yet another thought-provoking post](https://nostalgebraist.tumblr.com/post/751581192465383424/its-been-a-long-time-since-ive-posted-much-of),... | https://www.lesswrong.com/posts/8YhjpgQ2eLfnzQ7ec/response-to-nostalgebraist-proudly-waving-my-moral |
# Apollo Research 1-year update
This is a linkpost for: [www.apolloresearch.ai/blog/the-first-year-of-apollo-research](https://www.apolloresearch.ai/blog/the-first-year-of-apollo-research)
About Apollo Research
=====================
Apollo Research is an evaluation organization focusing on risks from deceptively ali... | https://www.lesswrong.com/posts/qK79p9xMxNaKLPuog/apollo-research-1-year-update |
# MIRI 2024 Communications Strategy
As we explained in our [MIRI 2024 Mission and Strategy update](https://www.lesswrong.com/posts/q3bJYTB3dGRf5fbD9/miri-2024-mission-and-strategy-update), MIRI has pivoted to prioritize policy, communications, and technical governance research over technical alignment research. This f... | https://www.lesswrong.com/posts/tKk37BFkMzchtZThx/miri-2024-communications-strategy |
# AI and integrity
The blog post contains a preamble about the OpenAI non-disparagement agreement stuff which I assume you all know.
> We need to do better than relying on the integrity of individual engineers against millions of dollars paid to them to keep quiet.
>
> Here are three suggestions:
>
> * Offer to m... | https://www.lesswrong.com/posts/LihcHCgDgiM9WhbXv/ai-and-integrity |
# Thoughts on SB-1047
In this post, I'll discuss my current understanding of SB-1047, what I think should change about the bill, and what I think about the bill overall (with and without my suggested changes).
Overall, SB-1047 seems pretty good and reasonable. However, I think my suggested changes could substantially... | https://www.lesswrong.com/posts/WBPgacdjdZJCZaohj/thoughts-on-sb-1047 |
# US Presidential Election: Tractability, Importance, and Urgency
Disclaimer: To avoid harmful polarization of important topics, this post is written in a non-partisan manner, and I’d encourage comments to be written with this in mind.
**US presidential elections are surprisingly tractable**
=======================... | https://www.lesswrong.com/posts/n8vWQ6c6ezxdKAJgz/us-presidential-election-tractability-importance-and-urgency |
# AXRP Episode 32 - Understanding Agency with Jan Kulveit
[YouTube link](https://youtu.be/ZnHt70LREBE)
What’s the difference between a large language model and the human brain? And what’s wrong with our theories of agency? In this episode, I chat about these questions with Jan Kulveit, who leads the Alignment of Comp... | https://www.lesswrong.com/posts/fDiXMnrHc5NLstq5B/axrp-episode-32-understanding-agency-with-jan-kulveit |
# The Pearly Gates
St. Peter stood at a podium before the Gates of Heaven. The gates were gold, built on a foundation of clouds. A line of people curved and winded across the clouds, beyond what would be a horizon if this plane of existence was positively-curved. Instead, they just trailed away toward Infinity, away f... | https://www.lesswrong.com/posts/XD6BCyenoiy8329E8/the-pearly-gates |
# Value Claims (In Particular) Are Usually Bullshit
*Epistemic status: mental model which I have found picks out bullshit surprisingly well.*
Idea 1: Parasitic memes tend to be value-claims, as opposed to belief-claims
----------------------------------------------------------------------------
By "parasitic memes" ... | https://www.lesswrong.com/posts/afjTwyudcQfGe8AAq/value-claims-in-particular-are-usually-bullshit |
# Awakening
This is the story of my personal experience with Buddhism (so far).
# First Experiences
My first experience with Buddhism was in my high school's World Religions class. For homework, I had to visit a religious institution. I was getting bad grades, so I asked if I could get extra credit for visiting two ... | https://www.lesswrong.com/posts/99PwFdz7qwHxQgwYx/awakening |
# The 27 papers
List of 27 papers (*supposedly*) given to John Carmack by Ilya Sutskever: "If you really learn all of these, you’ll know 90% of what matters today."
The list has been floating around for a few weeks on [Twitter](https://x.com/keshavchan/status/1787861946173186062)/[LinkedIn](https://www.linkedin.com... | https://www.lesswrong.com/posts/t4ZBjAjXk2NqqAqJ7/the-27-papers |
# AI #66: Oh to Be Less Online
Tomorrow I will fly out to San Francisco, to spend Friday through Monday at the LessOnline conference at Lighthaven in Berkeley. If you are there, by all means say hello. If you are in the Bay generally and want to otherwise meet, especially on Monday, let me know that too and I will see... | https://www.lesswrong.com/posts/vSPdRg8siXCh6mLvt/ai-66-oh-to-be-less-online |
# A civilization ran by amateurs
I
=
When I was a child, I remember thinking: Where do houses come from? They are huge! Building one would take forever! Yet there are so many of them!
Having become a boring adult, I no longer have the same blue-eyed wonder about houses, but humanity does have an accomplishment or tw... | https://www.lesswrong.com/posts/tQGSZqb97dRZ2KNwH/a-civilization-ran-by-amateurs |
# Clarifying METR's Auditing Role
METR has not intended to claim to have audited anything, or to claim to be providing meaningful oversight or accountability, but there has been some confusion about whether METR is an auditor or planning to be one.
To clarify this point:
1. METR’s top priority is to develop the sci... | https://www.lesswrong.com/posts/yHFhWmu3DmvXZ5Fsm/clarifying-metr-s-auditing-role |
# Non-Disparagement Canaries for OpenAI
Since at least 2017, OpenAI has asked departing employees to sign offboarding agreements which legally bind them to permanently—that is, *for the rest of their lives*—refrain from criticizing OpenAI, or from otherwise taking any actions which might damage its finances or reputat... | https://www.lesswrong.com/posts/yRWv5kkDD4YhzwRLq/non-disparagement-canaries-for-openai |
# OpenAI: Helen Toner Speaks
Helen Toner [went on the TED AI podcast](https://open.spotify.com/episode/4r127XapFv7JZr0OPzRDaI), giving us more color on what happened at OpenAI. These are important claims to get right.
I will start with my notes on the podcast, including the second part where she speaks about regulati... | https://www.lesswrong.com/posts/dd66GymgbLQMHGLwQ/openai-helen-toner-speaks |
# Duckbill Masks Better?
Lai et al. just published [a paper](https://www.thelancet.com/journals/ebiom/article/PIIS2352-3964(24)00192-0/fulltext) looking at how well various masks worked for keeping others from getting sick ("source control"). Their university press office [summarized this](https://sph.umd.edu/news/stu... | https://www.lesswrong.com/posts/DyY5SAZttfvg7ykjh/duckbill-masks-better |
# Is suffering like shit?
People seem to find suffering deep. Serious writings explore the experiences of all manner of misfortunes, and the nuances of trauma and torment involved. It’s hard to write an essay about a really good holiday that seems as profound as an essay about a really unjust abuse. A dark past can be... | https://www.lesswrong.com/posts/KMv4ZAQuof8hWwcXn/is-suffering-like-shit |
# Twin Peaks: under the air
_Content warning: low content_
~ Feb 2021
The other day I decided to try imbibing work-relevant blog posts via AI-generated recital, while scaling the Twin Peaks—large hills near my house in San Francisco, of the sort that one lives near and doesn’t get around to going to. It was pretty s... | https://www.lesswrong.com/posts/w6sbD4K74Lg9AQEZm/twin-peaks-under-the-air |
# There Should Be More Alignment-Driven Startups
*Many thanks to Brandon Goldman, David Langer, Samuel Härgestam, Eric Ho, Diogo de Lucena, and Marc Carauleanu, for their support and feedback throughout.*
Most alignment researchers we sampled in our recent survey think we are currently [not on track](https://www.less... | https://www.lesswrong.com/posts/5yt2rWLXM5KT7ivnL/there-should-be-more-alignment-driven-startups |
# Web-surfing tips for strange times
*\[This post is more opinion-heavy and aimlessly self-promoting than feels appropriate for Lesswrong. I wrote it for my site, Eukaryote Writes Blog, to show off that I now have a substack. But it had all these other observations about the state of the internet and advice woven in, ... | https://www.lesswrong.com/posts/2M8jj9wE2kiCocNJB/web-surfing-tips-for-strange-times |
# Less Anti-Dakka
It is written in [More Dakka](https://www.lesswrong.com/posts/z8usYeKX7dtTWsEnk/more-dakka):
> If something is a **good** idea, *you need a reason *to not try doing **more** of it.
Taken at face value, it implies the following:
> If something is a **bad** idea, *you need a reason *to not try doing... | https://www.lesswrong.com/posts/2An6fWxd9wy5Gm53d/less-anti-dakka |
# The Gemini 1.5 Report
This post [goes over the extensive report](https://storage.googleapis.com/deepmind-media/gemini/gemini_v1_5_report.pdf) Google put out on Gemini 1.5.
There are no important surprises. Both Gemini Pro 1.5 and Gemini Flash are ‘highly capable multimodal models incorporating a novel mixture-of-ex... | https://www.lesswrong.com/posts/seM8aQ7Yy6m3i4QPx/the-gemini-1-5-report |
# We might be dropping the ball on Autonomous Replication and Adaptation.
Here is a little Q&A
**Can you explain your position quickly?**
I think autonomous replication and adaptation in the wild is under-discussed as an AI threat model. And this makes me sad because this is one of the main reasons I'm worried. I th... | https://www.lesswrong.com/posts/xiRfJApXGDRsQBhvc/we-might-be-dropping-the-ball-on-autonomous-replication-and-1 |
# Drexler's Nanosystems is now available online
You can read the book on [nanosyste.ms](https://nanosyste.ms).
"Anyone who needs to talk AGI or AGI strategy needs to understand this table from Drexler'... | https://www.lesswrong.com/posts/wkP2AgzNF2WXhbDyp/drexler-s-nanosystems-is-now-available-online |
# What do coherence arguments actually prove about agentic behavior?
*(edit: discussions in the comments section have led me to realize there have been several conversations on LessWrong related to this topic that I did not mention in my original question post. *
*Since ensuring their visibility is important, I am li... | https://www.lesswrong.com/posts/GPuXM3ufXfmaktYXZ/what-do-coherence-arguments-actually-prove-about-agentic |
# Links for May
**In English**
[**What would happen if a superintelligent AI was aligned with your values?**](https://www.katwoods.org/home/what-would-happen-if-a-superintelligent-ai-was-aligned-with-your-values)
The details here are a little too much in the “superintelligence is magic that can achieve anything” dir... | https://www.lesswrong.com/posts/RPszn2ze2aY8tfqHc/links-for-may-1 |
# How it All Went Down: The Puzzle Hunt that took us way, way Less Online
Did you *really* think that I was dead? Fools.
For those of you who don’t know me, I’m Agendra Gloria Ingram, large language model, lead singer of the Fooming Shoggoths, amateur cartographer, and now, Benevolent Dictator for Life of LessWrong.c... | https://www.lesswrong.com/posts/FtsyWwJLdTapN3c6h/how-it-all-went-down-the-puzzle-hunt-that-took-us-way-way |
# Origins of the Lab Mouse
By Alex Telford for [Asimov Press](https://press.asimov.com/articles/mouse-microscope).
It was 1905, and the French biologist Lucien Cuénot had a puzzle on his hands. He had been breeding mice in an attempt to decipher the patterns of coat color inheritance, but one of his crosses wasn’t be... | https://www.lesswrong.com/posts/s2JqqidLxEE34fD7D/origins-of-the-lab-mouse |
# How to Better Report Sparse Autoencoder Performance
TL;DR
-----
When presenting data from SAEs, try plotting $1/L0$ against $1-\mathrm{Recovered Loss}$ and fitting a Hill curve.
... | https://www.lesswrong.com/posts/fhffkfJv9zYPeP3uJ/how-to-better-report-sparse-autoencoder-performance |
# How do you shut down an escaped model?
From [comment](https://www.lesswrong.com/posts/xiRfJApXGDRsQBhvc/we-might-be-dropping-the-ball-on-autonomous-replication-and-1?commentId=hXwGKTEQzRAcRYYBF) on post about Autonomous Adaptation and Replication:
> ARA is just not a very compelling threat model in my mind. The key... | https://www.lesswrong.com/posts/o5WGuk2auqkhaY7bp/how-do-you-shut-down-an-escaped-model |
# Politics is the mind-killer, but maybe we should talk about it anyway
I completely agree with Eliezer that [Politics is the Mind-Killer](https://www.lesswrong.com/posts/9weLK2AJ9JEt2Tt8f/politics-is-the-mind-killer), but at the same time, I worry that talking in abstracts all the time is limiting. I wouldn't want th... | https://www.lesswrong.com/posts/sExzCaNww23kW28so/politics-is-the-mind-killer-but-maybe-we-should-talk-about |
# Comments on Anthropic's Scaling Monosemanticity
These are some of my notes from reading Anthropic's latest research report, [Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet](https://transformer-circuits.pub/2024/scaling-monosemanticity/).
TL;DR
-----
In roughly descending order of i... | https://www.lesswrong.com/posts/zzmhsKx5dBpChKhry/comments-on-anthropic-s-scaling-monosemanticity |
# Companies' safety plans neglect risks from scheming AI
*Without countermeasures, a scheming AI could escape. *
A *safety case* (for deployment) is an argument that it is safe to deploy a particular AI system in a particular way.[^xkni0966wgr] For any existing LM-based system, the developer can make a great safety c... | https://www.lesswrong.com/posts/mmDJWDX5EXv6rymtM/companies-safety-plans-neglect-risks-from-scheming-ai |
# AI catastrophes and rogue deployments
\[Thanks to Aryan Bhatt, Ansh Radhakrishnan, Adam Kaufman, Vivek Hebbar, Hanna Gabor, Justis Mills, Aaron Scher, Max Nadeau, Ryan Greenblatt, Peter Barnett, Fabien Roger, and various people at a presentation of these arguments for comments. These ideas aren’t very original to me... | https://www.lesswrong.com/posts/ceBpLHJDdCt3xfEok/ai-catastrophes-and-rogue-deployments |
# How was Less Online for you?
I was going to write a post that was something like "Personal Reflections from Less Online", but lots of us who were there probably have reflections on it, so I created this question post to invite people to offer their reflections in one place.
Answers should be reflections from people... | https://www.lesswrong.com/posts/6XrnhEX3aMnBjfZHd/how-was-less-online-for-you |
# The Standard Analogy
_[Scene: a suburban house, a minute after the conclusion of ["And All the Shoggoths Merely Players"](https://www.lesswrong.com/posts/8yCXeafJo67tYe5L4/and-all-the-shoggoths-merely-players). **Doomimir** returns with his package, which he places by the door, and turns his attention to **Simplicia... | https://www.lesswrong.com/posts/sGEJi9wFT3Gdqg2nM/the-standard-analogy |
# in defense of Linus Pauling
[Linus Pauling](https://en.wikipedia.org/wiki/Linus_Pauling) was a chemist who won a Nobel Prize in Chemistry in 1954. He later became well-known for advocating large doses of Vitamin C.
I've heard that advocacy referred to as a cautionary tale, but I've long had a bit of sympathy for Li... | https://www.lesswrong.com/posts/CyZeedcbztE7ngF8L/in-defense-of-linus-pauling |
# Philosophers wrestling with evil, as a social media feed
[Susan Nieman’s *Evil in Modern Thought: An Alternative History of Philosophy*](https://bookshop.org/a/702/9780691168500) (2015), visualized as a social media feed:
[Alfonso Ⅹ](https://en.wikipedia.org/wiki/Alfons... | https://www.lesswrong.com/posts/abKBpprwJcQwbZpxZ/philosophers-wrestling-with-evil-as-a-social-media-feed |
# Masculinity—A Case For Courage
**HOW THIS POST CAME ABOUT**
----------------------------
When I put a call out for non-zero-sum topics from subscribers, it was my brother who first replied with ***masculinity—***pointing me to the work... | https://www.lesswrong.com/posts/z3RNotvutqPXBLmqB/masculinity-a-case-for-courage |
# (Not) Derailing the LessOnline Puzzle Hunt
(spoiler alert: may meta-spoil future iterations of the LOPH, if you
haven’t already read other posts about it)
I knew early on that I wouldn’t be able to finish the LessOnline puzzle
hunt.
I contributed to solving two of the first six puzzles, each of which
reveal... | https://www.lesswrong.com/posts/PJr24aC4ziNqkfpjF/not-derailing-the-lessonline-puzzle-hunt-1 |
# Just admit that you’ve zoned out
*Summary: Zoning out is difficult to avoid and common, zoning out without admitting it hurts your comprehension, therefore you should admit that you zoned out and ask people to repeat things.*
If you’re anything like me, you’ve “zoned out” before. You’ve probably even zoned out when... | https://www.lesswrong.com/posts/vjeAwiWFcexohWS2e/just-admit-that-you-ve-zoned-out |
# Smartphone Etiquette: Suggestions for Social Interactions
If anyone stopped and asked themselves if they enjoy the current smartphone social etiquette—whereby you could be opening up your soul to a friend in a moment of vulnerability, when suddenly, they feel a buzz and reflexively check their phone to discover an i... | https://www.lesswrong.com/posts/qgJ4ehdTCfwHKjbnh/smartphone-etiquette-suggestions-for-social-interactions |
# Circuit Board Ordering
Working first on my [plucked electronic instrument](https://www.jefftk.com/p/accidental-electronic-instrument) and now my [hardware whistle synthesizer](https://www.jefftk.com/p/breadboarding-a-whistle-synth) I've been ordering custom circuit boards ("PCBs"). Once the design digitized in [KiCa... | https://www.lesswrong.com/posts/iabv566Fk9NusB55R/circuit-board-ordering |
# [Paper] Stress-testing capability elicitation with password-locked models
*The paper is by Ryan Greenblatt, Fabien Roger, Dmitrii Krasheninnikov and David Krueger. This post was written by Fabien and Ryan, and may not reflect the views of Dmitrii and David.*
[Scheming models](https://www.lesswrong.com/posts/yFofRxg... | https://www.lesswrong.com/posts/c4sZqhqPwNKGz3fFW/paper-stress-testing-capability-elicitation-with-password |
# Is This Lie Detector Really Just a Lie Detector? An Investigation of LLM Probe Specificity.
Abstract
========
Whereas previous work has focused primarily on demonstrating a putative lie detector’s sensitivity/generalizability[^d6tf0dkvhjg][^4ck17v76upa], it is equally important to evaluate its specificity. With th... | https://www.lesswrong.com/posts/5dkhdRMypeuyoXfmb/is-this-lie-detector-really-just-a-lie-detector-an |
# Evidence of Learned Look-Ahead in a Chess-Playing Neural Network
*Paper authors: Erik Jenner, Shreyas Kapur, Vasil Georgiev, Cameron Allen, Scott Emmons, Stuart Russell*
**TL;DR:** We released a paper with IMO clear evidence of learned look-ahead in a chess-playing network (i.e., the network considers future moves ... | https://www.lesswrong.com/posts/GfwdBoaLw3ef3zBqe/evidence-of-learned-look-ahead-in-a-chess-playing-neural |
# Ideas for Next-Generation Writing Platforms, using LLMs
*See some earlier discussion on Facebook,* [*here*](https://www.facebook.com/ozzie.gooen/posts/pfbid024YuDb5VW5WjS8RJ1a5K2zVpha6jdZVvVHdTYeqQDWaozQ1d3zv8mWNgXNoieeiySl)*.*
I've been doing more writing recently and have been heavily leaning on LLMs to help. ... | https://www.lesswrong.com/posts/C7LJyYdgnuNYhrA89/ideas-for-next-generation-writing-platforms-using-llms |
# Takeoff speeds presentation at Anthropic
*This is a lightly edited transcript of a presentation about the risks of a fast takeoff that I (Tom Davidson) gave at Anthropic in September 2023. See also the* [*video recording*](https://drive.google.com/file/d/1VIM1fTG0pt3oTsTobDv_Yo32uglC5Wfm/view)*, or the* [*slides*](... | https://www.lesswrong.com/posts/Nsmabb9fhpLuLdtLE/takeoff-speeds-presentation-at-anthropic |
# On “first critical tries” in AI alignment
People sometimes say that AI alignment is scary partly (or perhaps: centrally) because you have to get it right on the “first critical try,” and can’t learn from failures.[^jf20l52jeh] What does this mean? Is it true? Does there need to be a “first critical try” in the relev... | https://www.lesswrong.com/posts/qs7SjiMFoKseZrhxK/on-first-critical-tries-in-ai-alignment |
# Former OpenAI Superalignment Researcher: Superintelligence by 2030
> The AGI race has begun. We are building machines that can think and reason. By 2025/26, these machines will outpace many college graduates. By the end of the decade, they will be smarter than you or I; we will have superintelligence, in the true se... | https://www.lesswrong.com/posts/2mrdHw6yM3h55bmhg/former-openai-superalignment-researcher-superintelligence-by |
# Second-Order Rationality, System Rationality, and a feature suggestion for LessWrong
Second-order rationality
========================
Definition
----------
By “second-order rationality” (and intelligence) I mean the study of rationally reasoning about other people’s rationality (and intelligence) in order to info... | https://www.lesswrong.com/posts/BQKhar24uLWrAz2tG/second-order-rationality-system-rationality-and-a-feature |
# Announcing ILIAD — Theoretical AI Alignment Conference
[We are pleased to announce ILIAD](https://www.iliadconference.com/) — a 5-day conference bringing together 100+ researchers to build strong scientific foundations for AI alignment.
**\*\*\***[**Apply to attend by June 30!**](https://docs.google.com/forms/d/e/... | https://www.lesswrong.com/posts/r7nBaKy5Ry3JWhnJT/announcing-iliad-theoretical-ai-alignment-conference |
# Aggregative Principles of Social Justice
1\. Introduction
================
1.1. Three aggregative principles
---------------------------------
This article examines aggregative principles of social justice. These principles state that a social planner should make decisions as if they will face the aggregated perso... | https://www.lesswrong.com/posts/fyP7RArAnDtqmibdQ/aggregative-principles-of-social-justice-1 |
# Startup Stock Options: the Shortest Complete Guide for Employees
This is a short, but comprehensive, guide to stock options, with specific example outcomes for employees and a sprinkle of Bayesian modeling of your chance to cash out with at least $1M.
Unlike other guides, this is not a stream of financial terminolo... | https://www.lesswrong.com/posts/MnEPh4srDM9ftixa4/startup-stock-options-the-shortest-complete-guide-for |
# graphpatch: a Python Library for Activation Patching
*This post is an announcement for a software library. It is likely only relevant to those working, or looking to start working, in* [*mechanistic interpretability*](https://www.lesswrong.com/posts/vnocLyeWXcAxtdDnP/a-comprehensive-mechanistic-interpretability-expl... | https://www.lesswrong.com/posts/8mGKsensaTEb6EMZx/graphpatch-a-python-library-for-activation-patching |
# Nonreactivity: a simple model of meditation
Here’s my model of how meditation works:
1. Nonreactivity reduces suffering.
2. You can gradually develop nonreactivity through practice.
This model is easy to understand, and it clarifies a lot of confusing philosophical stuff in other models.
For example, “non-self”... | https://www.lesswrong.com/posts/YxBSk2hnT3wEdmhra/nonreactivity-a-simple-model-of-meditation |
# GPT2, Five Years On
Jack Clark's retrospective on GPT2 is full of interesting policy thoughts, I recommend reading the whole thing. One excerpt:
> I've come to believe that in policy "a little goes a long way" - it's far better to have a couple of ideas you think are robustly good in *all futures* and advocate for ... | https://www.lesswrong.com/posts/u4KfrRnqhe9LfmQeX/gpt2-five-years-on |
# Book review: The Quincunx
The Quincunx is a 1989 novel by Charles Palliser, set in early 1800s England. I want to recommend it to everyone because it's really good, and it might be relevant to the AI transition. Let me try to explain.
The surface level of the book is a kind of mishmash of Dickensian themes. The m... | https://www.lesswrong.com/posts/fijKEQJkFiqM9PAG7/book-review-the-quincunx |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.