text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
#
Managing AI Risks in an Era of Rapid Progress
Managing AI Risks in an Era of Rapid Progress
=============================================
Authors
-------
Yoshua Bengio
Geoffrey Hinton
Andrew Yao
Dawn Song
Pieter Abbeel
Yuval Noah Harari
Ya-Qin Zhang
Lan Xue
Shai Shalev-Shwartz
Gillian Hadfield
Jeff Cl... | https://www.lesswrong.com/posts/a3RjXa2dryoH6Xgij/managing-ai-risks-in-an-era-of-rapid-progress |
# AI Safety Hub Serbia Official Opening
TLDR: We're thrilled to announce that we are now welcoming full-time tenants to our newly transformed office space for AI Safety researchers in search of an inspiring workspace. You can take a glimpse at the photos below to get a sneak peek. We're extending a warm invitation, wi... | https://www.lesswrong.com/posts/jcCyii8NcZLZ2M223/ai-safety-hub-serbia-official-opening |
# Shane Legg interview on alignment
This is Shane Legg, cofounder of DeepMind, on Dwarkesh Patel's podcast. The link is to the ten-minute section in which they specifically discuss alignment. Both of them seem to have a firm grasp on alignment issues as they're discussed on LessWrong.
For me, this is a significant up... | https://www.lesswrong.com/posts/2QLxNdxQpnesokk9H/shane-legg-interview-on-alignment |
# Regrant up to $600,000 to AI safety projects with GiveWiki
| The [*AI Safety GiveWiki*](https://ai.givewiki.org/) aggregates donation data from over 200 EA donors and combines them with retroactive evaluations to generate a list of the top AI safety projects to support. Some 50 donors with a total donation budget of... | https://www.lesswrong.com/posts/dnwzskaKpzKCmAvuj/regrant-up-to-usd600-000-with-givewiki |
# Book Review: Orality and Literacy: The Technologizing of the Word
*Thanks to Diarmuid Morgan for reviewing an earlier version of this post.*
> *Dream cancels dream in this new realm of fact*
> *From which we wake into the dream of act*[^p4f0zcv79]
Review of [Orality and Literacy: The Technologizing of the Word]... | https://www.lesswrong.com/posts/ThST9njmesR9BkoWq/book-review-orality-and-literacy-the-technologizing-of-the |
# Vaniver's thoughts on Anthropic's RSP
[Announcement](https://www.anthropic.com/index/anthropics-responsible-scaling-policy), [Policy v1.0](https://www-files.anthropic.com/production/files/responsible-scaling-policy-1.0.pdf), [evhub's argument in favor on LW](https://www.lesswrong.com/posts/mcnWZBnbeDz7KKtjJ/rsps-are... | https://www.lesswrong.com/posts/SseoT9mKDTL3RCbE9/vaniver-s-thoughts-on-anthropic-s-rsp |
# Comparing representation vectors between llama 2 base and chat
*(Status: rough writeup of an experiment I did today that I thought was somewhat interesting - there is more to investigate here regarding how RLHF affects these concept representations)*
This post presents the results of some experiments I ran to:
* ... | https://www.lesswrong.com/posts/PDDG4uuCLpPsuKepY/comparing-representation-vectors-between-llama-2-base-and |
# Experiments as a Third Alternative
As a kid in elementary school, I was diagnosed with ADHD. I think the diagnostic process is dumb and unreliable, and so the fact that I was diagnosed seems like it should only count as weak evidence. But knowing what I know about myself and having read a few books about ADHD, my be... | https://www.lesswrong.com/posts/cbfiXhEXfvZRHAzNC/experiments-as-a-third-alternative |
# What's up with "Responsible Scaling Policies"?
I am interested in talking about whether RSPs are good or bad. I feel pretty confused about it, and would appreciate going in-depth with someone on this.
My current feelings about RSPs are roughly shaped like the following:
I feel somewhat hyper-alert and a bit paran... | https://www.lesswrong.com/posts/jyM7MSTvy8Qs6aZcz/what-s-up-with-responsible-scaling-policies |
# My idea of sacredness, divinity, and religion
Here’s a conception that I have about sacredness, divinity, and religion.
There’s a sense in which love and friendship didn’t have to exist.
If you look at the animal kingdom, you see all kinds of solitary species, animals that only come together for mating. Members of... | https://www.lesswrong.com/posts/FyeDR4Wy7GGGd7iBa/my-idea-of-sacredness-divinity-and-religion |
# A new intro to Quantum Physics, with the math fixed
*Confidence level: I am a physicist with a phd in computational quantum chemistry. I’m pretty sure there are no major errors here, but I still may have missed something. Thanks to Blake Stacey for looking over the post.*
The goal of this post is to rewrite everyth... | https://www.lesswrong.com/posts/PjEgxtzWMbonKeTbw/a-new-intro-to-quantum-physics-with-the-math-fixed |
# Mathematically-Defined Optimization Captures A Lot of Useful Information
Alright it's been a long hiatus. I may not post again for another year or two if no inspiration strikes me. I will summarize [my work so far](https://www.lesswrong.com/posts/tpr6kkaPQqFgiC5ys/defining-optimization-in-a-deeper-way-part-4).
1. ... | https://www.lesswrong.com/posts/wtCiEtQP3YqRkkurS/mathematically-defined-optimization-captures-a-lot-of-useful |
# Comp Sci in 2027 (Short story by Eliezer Yudkowsky)
Comp sci in 2017:
Student: I get the feeling the compiler is just ignoring all my comments.
Teaching assistant: You have failed to understand not just compilers but the concept of computation itself.
Comp sci in 2027:
Student: I get the feeling the compiler ... | https://www.lesswrong.com/posts/gQyphPbaLHBMJoghD/comp-sci-in-2027-short-story-by-eliezer-yudkowsky |
# Grokking, memorization, and generalization — a discussion
**Intro: **Kaarel and Dmitry have been meeting to think about various interpretability and toy model questions. We are both interested in unsupervised and superposition-agnostic interpretability results ([Kaarel](https://www.lesswrong.com/posts/Go5ELsHAyw7QrA... | https://www.lesswrong.com/posts/BYwGEBspGgPY5nBZN/grokking-memorization-and-generalization-a-discussion |
# math terminology as convolution
> On the one hand, this theory generalizes the Fuchsian and Bers uniformizations of complex hyperbolic curves and their moduli to nonarchimedean places. It is for this reason that we shall often refer to this theory as p-adic Teichmuller theory, for short. On the other hand, the theor... | https://www.lesswrong.com/posts/M2H8qGrrbtX39brKW/math-terminology-as-convolution |
# Multi-Winner 3-2-1 Voting
My favorite voting method is [3-2-1 voting](https://electowiki.org/wiki/3-2-1_voting)[^evnbw7ynapi], my second favorite is [approval voting](https://electowiki.org/wiki/Approval_voting). Both are methods for selecting one winner. Approval voting has variations for choosing multiple winners,... | https://www.lesswrong.com/posts/aA5bSGmAN7SXiPkQC/multi-winner-3-2-1-voting |
# Charbel-Raphaël and Lucius discuss interpretability
The following is a transcript of a public discussion between [Charbel-Raphaël Segerie](https://www.lesswrong.com/users/charbel-raphael) and [Lucius Bushnaq](https://www.lesswrong.com/users/lblack) that took place on 23 September during [LessWrong Community Weekend ... | https://www.lesswrong.com/posts/FDrgcfY8zs5e2eJDd/charbel-raphael-and-lucius-discuss-interpretability |
# President Biden Issues Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence
Released today (10/30/23) this is crazy, perhaps the most sweeping action taken by government on AI yet.
Below, I've segmented by x-risk and non-x-risk related proposals, excluding the proposals that are geared towards ... | https://www.lesswrong.com/posts/g5XLHKyApAFXi3fso/president-biden-issues-executive-order-on-safe-secure-and |
# Improving the Welfare of AIs: A Nearcasted Proposal
Overview and Summary
====================
In this document, I’ll discuss and propose a [*nearcasted*](https://www.alignmentforum.org/posts/Qo2EkG3dEMv8GnX8d/ai-strategy-nearcasting) plan for improving the algorithmic welfare of powerful AI systems, given no furthe... | https://www.lesswrong.com/posts/F6HSHzKezkh6aoTr2/improving-the-welfare-of-ais-a-nearcasted-proposal |
# AI Alignment [progress] this Week (10/29/2023)
Wow! It’s been one month already.
Some thoughts:
When I started this project, I had a general sense that there is alignment progress on a regular basis, but tracking it really brings this home. Every week, on Monday I wonder “is this the week I will have nothing to wr... | https://www.lesswrong.com/posts/envmbY4NLi4R5QHRx/ai-alignment-progress-this-week-10-29-2023 |
# Grokking Beyond Neural Networks
We recently authored a paper titled, [Grokking Beyond Neural Networks: An Empirical Exploration with Model Complexity](https://arxiv.org/abs/2310.17247). Below, we provide an abstract of the article along with key take-aways from our experiments.
Abstract
--------
In some settings n... | https://www.lesswrong.com/posts/YXeScZmfCGiX4fQGt/grokking-beyond-neural-networks |
# Will releasing the weights of large language models grant
widespread access to pandemic agents?
Abstract
========
Large language models can benefit research and human understanding by providing tutorials that draw on expertise from many different fields. A properly safeguarded model will refuse to provide "dual-use... | https://www.lesswrong.com/posts/ytGsHbG7r3W3nJxPT/will-releasing-the-weights-of-large-language-models-grant |
# Intrinsic Drives and Extrinsic Misuse: Two Intertwined Risks of AI
Given their advanced capabilities, future AI systems could pose significant risks to society. Some of this risk stems from humans using AI systems for bad ends (_misuse_), while some stems from the difficulty of controlling AI systems “even if we wan... | https://www.lesswrong.com/posts/8LCviDbDh4ye6xgHc/intrinsic-drives-and-extrinsic-misuse-two-intertwined-risks |
# [Closed] Agent Foundations track in MATS
In MATS Winter 2023, I'm going to mentor scholars who want to work on the [learning-theoretic agenda](https://www.lesswrong.com/posts/ZwshvqiqCvXPsZEct/the-learning-theoretic-agenda-status-2023). This was not part of the original announcement, because (due to logistical reaso... | https://www.lesswrong.com/posts/ndSQYCsXJmbsywzZ7/closed-agent-foundations-track-in-mats |
# Urging an International AI Treaty: An Open Letter
> We call on governments worldwide to actively respond to the potentially catastrophic risks posed by advanced artificial intelligence (AI) systems to humanity, encompassing threats from misuse, systemic risks, and loss of control. We advocate for the development and... | https://www.lesswrong.com/posts/epC5CrGCv4JGdfsjm/urging-an-international-ai-treaty-an-open-letter |
# AI Safety 101 - Chapter 5.1 - Debate
This text is an adapted excerpt from the 'Adversarial techniques for scalable oversight' section of the AGISF 2023 course, held at ENS Ulm in Paris in April 2023. **Its purpose is to provide a concise overview of the essential aspects of the session's program for readers who may ... | https://www.lesswrong.com/posts/WP4fciGn3rNtmq3tY/ai-safety-101-chapter-5-1-debate |
# Preventing Language Models from hiding their reasoning
When working with powerful AI models, one concern is that the AIs [steganographically encode secret messages](https://www.alignmentforum.org/posts/yDcMDJeSck7SuBs24/steganography-in-chain-of-thought-reasoning) when communicating with each other or in long Chains... | https://www.lesswrong.com/posts/9Fdd9N7Escg3tcymb/preventing-language-models-from-hiding-their-reasoning |
# AI Safety 101 - Chapter 5.2 - Unrestricted Adversarial Training
Introduction
============
This text is an adapted excerpt from the 'Adversarial techniques for scalable oversight' section of the AGISF 2023 course, held at ENS Ulm in Paris in April 2023. **Its purpose is to provide a concise overview of the essential... | https://www.lesswrong.com/posts/nz5NNAtfKJLmbtksL/ai-safety-101-chapter-5-2-unrestricted-adversarial-training |
# Self-Blinded L-Theanine RCT
| Value tracked | Effect size d (λ, p, σ change) | Effect size d (λ, p, σ change) |
| --- | --- | --- |
| | 200 mg Caffeine (n=1, m=50) | 500 mg L-theanine (n=1, m=50) |
| Log-score substance prediction[^1] | -0.6 | -0.7 |
| Absorption | 0.61 (λ=13.3, p=0.00017, -0.072) | 0.04 (λ=1.38, p... | https://www.lesswrong.com/posts/Am9oRbY3bgwE55TxL/self-blinded-l-theanine-rct |
# AISN #25: White House Executive Order on AI, UK AI Safety Summit, and Progress on Voluntary Evaluations of AI Risks
Welcome to the AI Safety Newsletter by the [Center for AI Safety](https://www.safe.ai/). We discuss developments in AI and AI safety. No technical background required.
Subscribe [here](https://newslet... | https://www.lesswrong.com/posts/ARK4tyCRDd3Gb5umh/aisn-25-white-house-executive-order-on-ai-uk-ai-safety |
# Thoughts on the AI Safety Summit company policy requests and responses
Over the next two days, the UK government is hosting an [AI Safety Summit](https://www.aisafetysummit.gov.uk/) focused on “the safe and responsible development of frontier AI”. They requested that seven companies (Amazon, Anthropic, DeepMind, Inf... | https://www.lesswrong.com/posts/ms3x8ngwTfep7jBue/thoughts-on-the-ai-safety-summit-company-policy-requests-and |
# Computational Approaches to Pathogen Detection
*While this post is my perspective and not an official post of my [employer](https://securebio.org/)'s, it also draws on a lot of collaborative work with others at the [Nucleic Acid Observatory](https://naobservatory.org/) (NAO)*
One of the future scenarios I'm most wo... | https://www.lesswrong.com/posts/zZ8vgoKvaEodm9gZL/computational-approaches-to-pathogen-detection |
# On Having No Clue
Let's suppose you're trying to figure out if something is going to be A or B, but you have no clue.
What probability should you assign to each option?
Some people might say that you should assign 50/50. After all, the argument goes, if you assigned 40/60 or 60/40, well it sure sounds like you're ... | https://www.lesswrong.com/posts/3PXBK2an9dcRoNoid/on-having-no-clue |
# Bay Winter Solstice 2023: Song & speech auditions
Hello! I'm looking for community members to lead singalongs and read speeches/poems at the Bay Winter Solstice event this year. If you're interested, please either send me your audition recordings by the end of **November 14** using the below forms, or sign up for an... | https://www.lesswrong.com/posts/T27Xt9gQwFSbk6FBG/bay-winter-solstice-2023-song-and-speech-auditions |
# The Bletchley Declaration on AI Safety
The Bletchley Declaration was just released at the At AI Safety Summit.
Tl;dr: The declaration underscores the transformative potential and risks of AI. Countries, including major global powers, commit to harnessing AI's benefits while addressing its challenges, especially th... | https://www.lesswrong.com/posts/eBLCqhEPDpD2DpqgQ/the-bletchley-declaration-on-ai-safety |
# Robustness of Contrast-Consistent Search to Adversarial Prompting
*Produced as part of the AI Safety Hub Labs programme run by Charlie Griffin and Julia Karbing. This project was mentored by Nandi Schoots.*
... | https://www.lesswrong.com/posts/qW6A6bALuaFKmbdMx/mission-impossible-dead-reckoning-part-1-ai-takeaways |
# Chinese scientists acknowledge xrisk & call for international regulatory body [Linkpost]
Some highlights from the article (bolding added):
Several Chinese academic attendees of the summit at Bletchley Park, England, which starts on Wednesday, have signed on to a statement that warns that **advanced AI will pose an... | https://www.lesswrong.com/posts/3w4dhYsMWxoigEKDd/chinese-scientists-acknowledge-xrisk-and-call-for |
# On the Executive Order
Or: I read the executive order and its fact sheet, so you don’t have to.
I spent Halloween reading the [entire Biden Executive Order on AI](https://www.whitehouse.gov/briefing-room/presidential-actions/2023/10/30/executive-order-on-the-safe-secure-and-trustworthy-development-and-use-of-artifi... | https://www.lesswrong.com/posts/PvBpRu354uG7ypwRP/on-the-executive-order |
# Snapshot of narratives and frames against regulating AI
*This is a speculative map of a hot discussion topic. I'm posting it in question form in the hope we can rapidly map the space in answers.*
Looking at various claims at X and at the AI summit, it seems possible to identify some key counter-regulation narra... | https://www.lesswrong.com/posts/PdcnEEE6sdgACDrEk/snapshot-of-narratives-and-frames-against-regulating-ai |
# 2023 LessWrong Community Census, Request for Comments
Overview
--------
I would like there to be a LessWrong Community Census, because I had fun playing with the data from last year and there's some questions I'm curious about. It's also an entertaining site tradition. Since nobody else has stepped forward to make ... | https://www.lesswrong.com/posts/qrLhLBCSHumQz9fon/2023-lesswrong-community-census-request-for-comments |
# A list of all the deadlines in Biden's Executive Order on AI
This is a list of all the deadlines outlined in the Executive Order (EO) published by Biden’s administration on 30.10.2023. I’ve put it together to make it easier for people to track what exactly should be happening when as a result of the EO. The required... | https://www.lesswrong.com/posts/cCbybRT8bgiMbEHEv/a-list-of-all-the-deadlines-in-biden-s-executive-order-on-ai |
# AI Alignment: A Comprehensive Survey
We have just released an academic survey of AI alignment.
We identify four main categories of alignment research:
1. Learning from feedback (e.g. scalable oversight)
2. Learning under distribution shift
3. Assurance (e.g. interpretability)
4. Governance
We mainly focused o... | https://www.lesswrong.com/posts/sL9qmAqgB2RL6JFca/ai-alignment-a-comprehensive-survey |
# Dario Amodei’s prepared remarks from the UK AI Safety Summit, on Anthropic’s Responsible Scaling Policy
*I hope Dario's remarks to the Summit can shed some light on how we think about RSPs in general and Anthropic's RSP in particular, both of which have been discussed extensively since [I shared our RSP announcement... | https://www.lesswrong.com/posts/vm7FRyPWGCqDHy6LF/dario-amodei-s-prepared-remarks-from-the-uk-ai-safety-summit |
# Reactions to the Executive Order
Previously: [On the Executive Order](https://thezvi.substack.com/p/on-the-executive-order)
This post compiles the reactions of others that I have seen to Biden’s Executive Order on AI, including reactions that were based only on the fact sheet, as well as my reactions to those react... | https://www.lesswrong.com/posts/G8SsspgAYEHHiDGNP/reactions-to-the-executive-order |
# My thoughts on the social response to AI risk
A common theme implicit in many AI risk stories has been that broader society will either fail to anticipate the risks of AI until it is too late, or do little to address those risks in a serious manner. In my opinion, there are now clear signs that this assumption is fa... | https://www.lesswrong.com/posts/EaZghEwcCJRAuee66/my-thoughts-on-the-social-response-to-ai-risk |
# Singular learning theory and bridging from ML to brain emulations
This is a fairly "chat"-style dialogue that I (kave) had with Garrett about singular learning theory (SLT) and his [ambitious plans](https://www.lesswrong.com/posts/d4qbjx35SBMGyFNWZ/my-hopes-for-alignment-singular-learning-theory-and-whole) for solvi... | https://www.lesswrong.com/posts/PikpeRucdsXeEvpy9/singular-learning-theory-and-bridging-from-ml-to-brain |
# Should people build productizations of open source AI models?
In early 2021, before I had developed any opinions on AI safety, my startup cofounder and I were contacted by a group of people who wanted an alternative to an AI CYOA platform, AIDungeon. They were frustrated with the site's current management and especi... | https://www.lesswrong.com/posts/Gkot4upnvMB7quzne/should-people-build-productizations-of-open-source-ai-models |
# Public Weights?
*While this is close to areas I work in, it's a personal post. No one reviewed this before I published it, or asked me to (or not to) write something. All mistakes are my own.*
A few days ago, some of my coworkers at [SecureBio](https://securebio.org/) put out a preprint, "Will releasing the weights... | https://www.lesswrong.com/posts/jBHSwNHKexd5WGXcs/public-weights |
# Averages and sample sizes
My brain seems to not get statistics and probability. So I was wondering about how averaging things out is influenced by the amount of items being averaged, and in particular, how the underlying distribution will change things. Here are some stories to illustrate the issue:
Minting
-------... | https://www.lesswrong.com/posts/HjdqPbhRx2ceXHzr2/averages-and-sample-sizes |
# Estimating effective dimensionality of MNIST models
The [local learning coefficent](https://www.alignmentforum.org/posts/6g8cAftfQufLmFDYT/you-re-measuring-model-complexity-wrong) $\lambda$ is a measure of a model's "effective dimensionality" that captures its complexity (more background [here](https://www.alignment... | https://www.lesswrong.com/posts/PDz68D5vQQgacAwoF/estimating-effective-dimensionality-of-mnist-models |
# Mech Interp Challenge: November - Deciphering the Cumulative Sum Model
**I'm writing this post to discuss solutions to the October challenge, and present the challenge for this November.**
If you've not read the [first post in this sequence](https://www.lesswrong.com/posts/dpRuey9MDHNACSDtn/mech-interp-challenge-a... | https://www.lesswrong.com/posts/uPa63suC8idWhYGbg/mech-interp-challenge-november-deciphering-the-cumulative |
# Saying the quiet part out loud: trading off x-risk for personal immortality
**_Statement:_** _I want to deliberately balance the caution and the recklessness in developing AGI, such that it gets created in the last possible moment so that I and my close ones do not die._
This Statement confuses me. There are severa... | https://www.lesswrong.com/posts/A22bTHYLDdkeqFsLd/saying-the-quiet-part-out-loud-trading-off-x-risk-for |
# AI #36: In the Background
Wow, what a week. We had the Executive Order, [which I read here so you don’t have to](https://thezvi.substack.com/p/on-the-executive-order) and then I have [a tabulation of the reactions of others](https://thezvi.substack.com/p/reactions-to-the-executive-order).
Simultaneously there was t... | https://www.lesswrong.com/posts/QLuoMnhR5XNAAWjJx/ai-36-in-the-background |
# Propaganda or Science: A Look at Open Source AI and Bioterrorism Risk
## 0: TLDR
I examined all the biorisk-relevant citations from a policy paper arguing that we should ban powerful open source LLMs.
None of them provide good evidence for the paper's conclusion. The best of the set is evidence from statements fro... | https://www.lesswrong.com/posts/ztXsmnSdrejpfmvn7/propaganda-or-science-a-look-at-open-source-ai-and |
# One Day Sooner
There is a particular skill I would like to share, which I wish I had learned when I was younger. I picked it up through working closely with a previous boss (a CTO who had founded a company and raised it up to hundreds of employees and multi-million dollar deals) but it wasn’t until I read [The Story... | https://www.lesswrong.com/posts/EsxowsJsRopTGALCX/one-day-sooner |
# Does davidad's uploading moonshot work?
[davidad](https://www.lesswrong.com/users/davidad) has a 10-min talk out on a proposal about which he says: “the first time I’ve seen a concrete plan that might work to get human uploads before 2040, maybe even faster, given unlimited funding”.
I think the talk is a good watc... | https://www.lesswrong.com/posts/FEFQSGLhJFpqmEhgi/does-davidad-s-uploading-moonshot-work |
# The other side of the tidal wave
I guess there’s maybe a 10-20% chance of AI causing human extinction in the coming decades, but I feel more distressed about it than even that suggests—I think because in the case where it doesn’t cause human extinction, I find it hard to imagine life not going kind of off the rails.... | https://www.lesswrong.com/posts/uyPo8pfEtBffyPdxf/the-other-side-of-the-tidal-wave |
# Thoughts on open source AI
*Epistemic status: I only ~50% endorse this, which is below my typical bar for posting something. I’m more bullish on “these are arguments which should be in the water supply and discussed” than “these arguments are actually correct.” I’m not an expert in this, I’ve only thought about it f... | https://www.lesswrong.com/posts/WLYBy5Cus4oRFY3mu/thoughts-on-open-source-ai |
# Securing Civilization Against Catastrophic Pandemics
### Executive summary
Pandemic security aims to safeguard the future of civilization from exponentially spreading biological threats. Despite the world's failure to contain SARS-CoV-2, the existence of far more lethal and transmissible pathogens that afflict anim... | https://www.lesswrong.com/posts/ksevLrNby34TJKGSF/securing-civilization-against-catastrophic-pandemics |
# Integrity in AI Governance and Advocacy
Ok, so we both had some feelings about the recent [Conjecture post on "lots of people in AI Alignment are lying"](https://www.lesswrong.com/posts/qtTW6BFrxWw4iHcjf/lying-is-cowardice-not-strategy), and the [associated marketing campaign and stuff](https://twitter.com/dw2/statu... | https://www.lesswrong.com/posts/vFqa8DZCuhyrbSnyx/integrity-in-ai-governance-and-advocacy |
# 8 examples informing my pessimism on uploading without reverse engineering
*(If you’ve already read everything I’ve written, you’ll find this post pretty redundant. See especially my old posts *[*Building brain-inspired AGI is infinitely easier than understanding the brain*](https://www.lesswrong.com/posts/PTkd8nazv... | https://www.lesswrong.com/posts/wByPb6syhxvqPCutu/8-examples-informing-my-pessimism-on-uploading-without |
# Deception Chess: Game #1
(You can sign up [here](https://docs.google.com/forms/d/e/1FAIpQLScPKrSB6ytJcXlLhnxgvRv1V4vMx8DXWg1j9KYVfVT1ofdD-A/viewform?vc=0&c=0&w=1&flr=0) if you haven't already.)
This is the first of my analyses of the deception chess games. The introduction will describe the setup of the game, and t... | https://www.lesswrong.com/posts/6dn6hnFRgqqWJbwk9/deception-chess-game-1 |
# Paul Christiano on Dwarkesh Podcast
Dwarkesh's summary:
> Paul Christiano is the world’s leading AI safety researcher. My full episode with him is out!
>
> We discuss:
>
> * Does he regret inventing RLHF, and is alignment necessarily dual-use?
> * Why he has relatively modest timelines (40% by 2040, 15% by 20... | https://www.lesswrong.com/posts/7FrHeyxQpb3pvr9Pn/paul-christiano-on-dwarkesh-podcast |
# As Many Ideas
**Summary:** One person announces a problem to the room at large. Everyone else comes up with as many ideas as they can to solve the problem.
**Tags:** Repeatable, Medium, Investment, Experimental
**Purpose: **The purpose is twofold. Firstly, you may walk away with actionable solutions to your proble... | https://www.lesswrong.com/posts/B576hzaRpGpnyGEMB/as-many-ideas |
# Untrusted smart models and trusted dumb models
\[Ryan Greenblatt originally made this point to me a while ago, and we then developed a bunch of these ideas together. Thanks to Paul Christiano and a huge number of other people for various relevant discussions, and to Ryan, Roger Grosse, Peter Barnett, Oliver Habryka,... | https://www.lesswrong.com/posts/LhxHcASQwpNa3mRNk/untrusted-smart-models-and-trusted-dumb-models |
# ACX Montreal Meetup November 11th 2023
Hey all! This week, we're doing something a bit different. This ACX event will will resolve around the viewing a movie, *Threads,* and be hosted by Jeremy Eliosoff, who wrote the rest of this message. **Note the non-standard event time and location**.
**Description:**
First ... | https://www.lesswrong.com/events/DcEMmvqyR4j2JDL6f/acx-montreal-meetup-november-11th-2023 |
# If a little is good, is more better?
I’ve recently seen a bunch of discussions of the wisdom of publicly releasing the weights^[1](#fn:1)^ of advanced AI models. A common argument form that pops up in these discussions is this:
1. The problem with releasing weights is that it means that thing X can happen on a lar... | https://www.lesswrong.com/posts/R56gihyfwERKDLszQ/if-a-little-is-good-is-more-better |
# Being good at the basics
(crossposted from [my blog](https://sundaystopwatch.eu/being-good-at-the-basics/))
In Brazilian Jiu Jitsu, there's a notion of being "good at the basics", as opposed to "being good by knowing advanced techniques". *How* you want to be good is a question of personal preference, but the idea ... | https://www.lesswrong.com/posts/fLbkWseNn3geEifkv/being-good-at-the-basics |
# [Linkpost] Concept Alignment as a Prerequisite for Value Alignment
> Value alignment is essential for building AI systems that can safely and reliably interact with people. However, what a person values -- and is even capable of valuing -- depends on the concepts that they are currently using to understand and evalu... | https://www.lesswrong.com/posts/M6CEJmgna6FTt9Yci/linkpost-concept-alignment-as-a-prerequisite-for-value |
# The Soul Key
The ocean is your home, but a forbidding one: often tempestuous, seldom warm. So one of your great joys is crawling onto land and slipping off your furry seal skin, to laze in the sun in human form. The elders tell horror stories of friends whose skins were stolen by humans, a single moment of carelessn... | https://www.lesswrong.com/posts/tHvFtfFKjhfy3sQpC/the-soul-key |
# Genetic fitness is a measure of selection strength, not the selection target
*Alternative title: "Evolution suggests robust rather than fragile generalization of alignment properties."*
A [frequently](https://www.lesswrong.com/posts/GNhMPAWcfBCASy8e6/a-central-ai-alignment-problem-capabilities-generalization) [repe... | https://www.lesswrong.com/posts/BtffzD5yNB4CzSTJe/genetic-fitness-is-a-measure-of-selection-strength-not-the |
# The 6D effect: When companies take risks, one email can be very powerful.
Recently, I have been learning about industry norms, legal discovery proceedings, and incentive structures related to companies building risky systems. I wanted to share some findings in this post because they may be important for the frontier... | https://www.lesswrong.com/posts/J9eF4nA6wJW6hPueN/the-6d-effect-when-companies-take-risks-one-email-can-be |
# Stuxnet, not Skynet: Humanity's disempowerment by AI
Several high-profile AI skeptics and fellow travelers have recently raised the objection that it is inconceivable that a hostile AGI or smarter than human intelligence could end the human race. Some quotes from earlier this year:
Scott Aaronson:
> [The causal st... | https://www.lesswrong.com/posts/tyE4orCtR8H9eTiEr/stuxnet-not-skynet-humanity-s-disempowerment-by-ai |
# Utility is not the selection target
*Epistemic status: shitpost.*
Suppose that you are selecting for utility. Naively, you might think that this means you are selecting for utility, but actually this is not the case.
Sometimes greenbrownness is the selection target
------------------------------------------------
... | https://www.lesswrong.com/posts/KdEjYpdRyqi9DzNTm/utility-is-not-the-selection-target |
# Go flash blinking lights at printed text right now
1. Install a strobe light app on your phone (I used "strobe light tachometer" on ios.)
2. Get some printed text like a nutrition label or a paper book
3. Try to read the text with the light blinking at 5, 6, 7, 8, 9, 10, 11, 12, and 13 hz
4. Report results. Was read... | https://www.lesswrong.com/posts/diohDcgu3YdSbHd3j/go-flash-blinking-lights-at-printed-text-right-now |
# The Assumed Intent Bias
Summary: when thinking about the behavior of others, people seem to have a tendency to assume clear purpose and intent behind it. In this post I argue that this assumption of intent quite often is incorrect, and that a lot of behavior exists in a gray area where it’s easily influenced by subc... | https://www.lesswrong.com/posts/i4WsKvFYEMdya94DW/the-assumed-intent-bias |
# Pivotal Acts might Not be what You Think they are
*This article is mainly for people who have not read the [pivotal act](https://arbital.com/p/pivotal/) article on arbital or need a refresher. If you have, the most interesting section would probably be "Omnicient ML Researchers: A Pivotal Act without a Monolithic Co... | https://www.lesswrong.com/posts/MtkcDDf2ZPvFk4jtN/pivotal-acts-might-not-be-what-you-think-they-are |
# Eric Schmidt on recursive self-improvement
Recently, Eric Schmidt gave a talk at Harvard called “Our AI Future: Hopes and Hurdles Ahead”. The entire talk is available [here](https://www.youtube.com/watch?v=IeVY_Ag8JI8) but there was one part that was interesting to me (around the 1:11:00 mark), and that is, his view... | https://www.lesswrong.com/posts/cLC2HcQbFZ5pFAgqC/eric-schmidt-on-recursive-self-improvement |
# EA orgs' legal structure inhibits risk taking and information sharing on the margin
**What is fiscal sponsorship?**
===============================
It’s fairly common for EA orgs to provide fiscal sponsorship to other EA orgs. Wait, no, that sentence is not quite right. The more accurate sentence is that there are... | https://www.lesswrong.com/posts/XvEJydHAHk6hjWQr5/ea-orgs-legal-structure-inhibits-risk-taking-and-information |
# AI as Super-Demagogue
Sam Altman [recently said](https://twitter.com/sama/status/1716972815960961174):
> i expect ai to be capable of superhuman persuasion well before it is superhuman at general intelligence, which may lead to some very strange outcomes
I believe that he is absolutely right. Superhuman persuasion... | https://www.lesswrong.com/posts/DdDKsyA925Sm8BpQh/ai-as-super-demagogue |
# Disentangling four motivations for acting in accordance with UDT
Introduction
------------
In this post, I examine a number of different arguments for acting like an updateless agent. Concretely, I try to provide answers to the following question: **What could move someone who isn’t already born a perfectly coheren... | https://www.lesswrong.com/posts/nc2tzMLgXKc8NGzrQ/disentangling-four-motivations-for-acting-in-accordance-with |
# xAI announces Grok, beats GPT-3.5
Some highlights:
> *Grok is still a very early beta product – the best we could do with 2 months of training – so expect it to improve rapidly with each passing week with your help.*
I find it very interesting that they managed to beat GPT-3.5 with **only 2 months of training!** T... | https://www.lesswrong.com/posts/pbCw4QdL9K4tB7JEM/xai-announces-grok-beats-gpt-3-5 |
# On Overhangs and Technological Change
Imagine an almost infinite, nearly flat plain of early medieval farming villages populated by humans and their livestock (cows, horses, sheep, etc) politically organized into small duchies and generally peaceful apart from rare skirmishes.
Add a few key technologies like stirr... | https://www.lesswrong.com/posts/EwGFEPxwynf3ALdeK/on-overhangs-and-technological-change |
# When and why should you use the Kelly criterion?
This is a dialogue made during the [online dialogues party](https://www.lesswrong.com/events/BJcNeJss4jxc68GQR/online-dialogues-party-sunday-5th-november). Phil, River, and I (Garrett) talked about the Kelly criterion. I was under the mistaken impression that it would... | https://www.lesswrong.com/posts/qXvj6abkc9eqPjKaB/when-and-why-should-you-use-the-kelly-criterion |
# Taboo Wall
**Summary**: Rationalist Taboo is the skill of being able to talk about the same concepts using different words. Taboo Wall is open discussion, but as the discussion goes on more words get stricken from everyone’s vocabulary.
**Tags**: Very Experimental, Medium
**Purpose**: To practice the ability to c... | https://www.lesswrong.com/posts/eLetWRfhbcAWq42b8/taboo-wall |
# Announcing TAIS 2024
[AI Safety Tokyo](https://aisafety.tokyo/) is hosting TAIS 2024, a Technical AI Safety Conference. The conference will take place in Tokyo, Japan on April 5th, 2024. Details about the event can be found [here](https://tais2024.cc/).
The goal of this conference is to bring together specialists i... | https://www.lesswrong.com/posts/RdoFCykbfEyZzp675/announcing-tais-2024 |
#
Are language models good at making predictions?
To get a crude answer to this question, we took 5000 questions from Manifold markets that were resolved after GPT-4’s current knowledge cutoff of Jan 1, 2022. We gave the text of each of them to GPT-4, along with these instructions:
> You are an expert superforecaste... | https://www.lesswrong.com/posts/CkhJAxHeyFCg2EcET/are-language-models-good-at-making-predictions |
# An illustrative model of backfire risks from pausing AI research
**TLDR**
========
The post contains a simple modelization of the **maximum explosive takeoff speed** (simplified as "takeoff speed"), using as the main parameters:
* overhangs[^l4xx3gnvtnh] sizes (investment, hardware efficiency, software efficien... | https://www.lesswrong.com/posts/jjGiCZLuJ8ZNvZwQc/an-illustrative-model-of-backfire-risks-from-pausing-ai |
# What is true is already so.
Owning up to it doesn't make it worse.
I had a realization over the last few months that I'd like to share. I have always interpreted the [Litany of Gendlin](https://www.lesswrong.com/tag/litany-of-gendlin) to compress to something like "there's no sense hiding from bad news." What I real... | https://www.lesswrong.com/posts/wmku6vFpNzCFnYLbj/what-is-true-is-already-so-owning-up-to-it-doesn-t-make-it |
# Job listing: Communications Generalist / Project Manager
Looking for a way to help communicate about AI x-risk? MIRI is hiring in the Communications department. There will be additional job listings soon for writers and editors, but we're starting with this comms generalist / project manager role.
[https://intellig... | https://www.lesswrong.com/posts/ZwizCnxJdvEuChnBE/job-listing-communications-generalist-project-manager |
# A bet on critical periods in neural networks
See the code and data at [this github](https://github.com/GarretteBaker/critical_periods/tree/main)
Hey Garrett! I hear you have some new empirical results. What's the setup?
So I've told you my theory about how critical periods work. What I did is replicate the [paper]... | https://www.lesswrong.com/posts/e8r8Tx3Hk7LpnrMwY/a-bet-on-critical-periods-in-neural-networks |
# cost estimation for 2 grid energy storage systems
## CO2 mitigation costs
With modern gas turbines, burning natural gas makes perhaps 0.4 kg CO2 per kWh. I generally use $75/ton as a baseline target for CO2 mitigation costs; that's around what you see from various reasonable approaches like biomass conversion. $75/... | https://www.lesswrong.com/posts/xsjxrmoFqKevAj2Aj/cost-estimation-for-2-grid-energy-storage-systems |
# How to (hopefully ethically) make money off of AGI
Hey Everyone!
As part of working on dialogues over the last few weeks I've asked a bunch of people what kind of conversations they would be most interested in reading, and one of the most common one has been "I would really like to read a bunch of people trying to... | https://www.lesswrong.com/posts/CTBta9i8sav7tjC2r/how-to-hopefully-ethically-make-money-off-of-agi |
# The Perils of Professionalism
Professionalism is a useful trait to be able to display, but it isn’t an unalloyed good. This essay is attempting to outline why deliberately not signaling professionalism may be useful for you.
First, a definition by example for professionalism: Clean button down shirts with solid col... | https://www.lesswrong.com/posts/QChwTjL6oL2a6gbhm/the-perils-of-professionalism |
# AI Alignment Research Engineer Accelerator (ARENA): call for applicants
*(Edited, to now include a section specifically for FAQs about the virtual program.)*
TL;DR
=====
Apply [**here**](https://airtable.com/appZIMMH3ywSxS0A9/shrULWsu3oH0BpbMW) for the third iteration of ARENA (Jan 8th - Feb 2nd)!
Introduction... | https://www.lesswrong.com/posts/465MTELvNKMdQk322/ai-alignment-research-engineer-accelerator-arena-call-for-1 |
# Box inversion revisited
[Box inversion hypothesis](https://www.lesswrong.com/posts/TQwXPHfyyQwr22NMh/box-inversion-hypothesis/) is a proposed correspondence between problems with AI systems studied in approaches like [agent foundations](https://www.lesswrong.com/tag/agent-foundations), and problems with AI ecosystem... | https://www.lesswrong.com/posts/jrKftFZMZjvNdQLNR/box-inversion-revisited |
# On the UK Summit
In the eyes of many, Biden’s Executive Order somewhat overshadowed the UK Summit. The timing was unfortunate. Both events were important milestones. Now that I have had time, here is my analysis of what happened at the UK Summit.
As is often the case with such events, there was a lot of talk relati... | https://www.lesswrong.com/posts/zbrvXGu264u3p8otD/on-the-uk-summit |
# AI Alignment [Progress] this Week (11/05/2023)
The biggest news of this week is probably the two new AI Alignment Initiatives.
Otherwise, this was a relatively quiet week for AI Alignment breakthroughs.
So here are our
AI Alignment Breakthroughs this Week
====================================
This week there were... | https://www.lesswrong.com/posts/FLCHNqu2FNRCJ4dyg/ai-alignment-progress-this-week-11-05-2023 |
# What I've been reading, November 2023
A ~monthly feature. Recent blog posts and news stories are generally omitted; you can find them in my [links digests](https://rootsofprogress.org/writing#links-digest). All emphasis in bold in the quotes below was added by me.
**Books**
---------
Finished **Lynn White,** [***M... | https://www.lesswrong.com/posts/hKK4KdoTdTtaNqCaj/what-i-ve-been-reading-november-2023 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.