text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# How to Eradicate Global Extreme Poverty [RA video with fundraiser!]
[Cross-posted from the EA Forum](https://forum.effectivealtruism.org/posts/ba7Lwem3R6dCjAahY/how-to-eradicate-global-extreme-poverty-ra-video-with)
*In this video, we outline a plan to eradicate global extreme poverty using cash transfers. The appr... | https://www.lesswrong.com/posts/Lw5bmncMRJw5kDwqN/how-to-eradicate-global-extreme-poverty-ra-video-with |
# AISN #24: Kissinger Urges US-China Cooperation on AI, China's New AI Law, US Export Controls, International Institutions, and Open Source AI
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.
Subscri... | https://www.lesswrong.com/posts/wwKoGMq4omR3Mrv4w/aisn-24-kissinger-urges-us-china-cooperation-on-ai-china-s |
# The Real Fanfic Is The Friends We Made Along The Way
*\[This is a humorous short story about an AI that is almost aligned... but not quite. It was originally written for the 2022 EA writing retreat, Ink At The Abbey, and was generally well received\]*
The Real Fanfic Is The Friends We Made Along The Way
-----------... | https://www.lesswrong.com/posts/SibcE6H9i6wifzqAF/the-real-fanfic-is-the-friends-we-made-along-the-way |
# Superforecasting the premises in “Is power-seeking AI an existential risk?”
(Cross-posted from [my website.](https://joecarlsmith.com/2023/10/18/superforecasting-the-premises-in-is-power-seeking-ai-an-existential-risk))
[Good Judgment](https://goodjudgment.com/) has solicited reviews and forecasts from superforecas... | https://www.lesswrong.com/posts/voABShdmY6A9hMFnG/superforecasting-the-premises-in-is-power-seeking-ai-an |
# AI Safety 101 : Reward Misspecification
Overview
========
1. **Reinforcement Learning**: The chapter starts with a reminder of some reinforcement learning concepts. This includes a quick dive into the concept of rewards and reward functions. This section lays the groundwork for explaining why reward design is extr... | https://www.lesswrong.com/posts/mMBoPnFrFqQJKzDsZ/ai-safety-101-reward-misspecification |
# Does AI governance needs a "Federalist papers" debate?
During the American Revolution, a federal army and government was needed to fight against the British. Many people were afraid that the powers granted to the government for that purpose would allow it to become tyrannical in the future.
If the founding fathers ... | https://www.lesswrong.com/posts/YkrDKR7TyqDxwKj23/does-ai-governance-needs-a-federalist-papers-debate |
# The (partial) fallacy of dumb superintelligence
In[ her opening statement for the Munk debate on AI risk,](https://www.youtube.com/watch?v=skRgYH7oAjc) Melanie Mitchell addressed a proposed example of AGI risk: an AGI tasked with fixing climate change might decide to eliminate humans as the source of carbon emission... | https://www.lesswrong.com/posts/qsDPHZwjmduSMCJLv/the-partial-fallacy-of-dumb-superintelligence |
# Are humans misaligned with evolution?
There is an argument that although humans evolved under pressure to maximize inclusive genetic fitness (IGF), humans don't actually try to maximize their own IGF. This, as the argument goes, shows that in the one case we have of a process creating general intelligence, it was no... | https://www.lesswrong.com/posts/xqXdDs68zMJ82Dcmt/are-humans-misaligned-with-evolution |
# [Video] Too much Empiricism kills you
[Here](https://youtu.be/vqHlPb18ROE?si=vf840i97GZwgxIjD) is a video I made 2 months ago. It gives a mediocre at best explanation for an important foundational argument:
It is normally possible to make progress using empirical methods, as long as you can measure how good a part... | https://www.lesswrong.com/posts/tqDT8CCm4jubaWkC3/video-too-much-empiricism-kills-you |
# Is Yann LeCun strawmanning AI x-risks?
Tom Chivers expresses his frustration with Yann LeCun:
I also find his comments here frustrating, but I want to offer another possible explanation.
Even tho... | https://www.lesswrong.com/posts/gYwfkjK8vsowfDkZu/is-yann-lecun-strawmanning-ai-x-risks |
# AI #34: Chipping Away at Chip Exports
It did not get the bulk of the attention, but the actual biggest story this week was that America tightened the rules on its chip exports, closing the loophole Nvidia was using to create the A800 and H800. Perhaps the new restrictions will actually have teeth.
Also new capabili... | https://www.lesswrong.com/posts/ApPKqx9b8LogfKxAr/ai-34-chipping-away-at-chip-exports |
# New roles on my team: come build Open Phil's technical AI safety program with me!
Open Phil [announced](https://forum.effectivealtruism.org/posts/bBefhAXpCFNswNr9m/open-philanthropy-is-hiring-for-multiple-roles-across-our) two weeks ago that we’re hiring for over 20 roles across our teams working on global catastrop... | https://www.lesswrong.com/posts/to9hsT76Jy9HWJ5dj/new-roles-on-my-team-come-build-open-phil-s-technical-ai |
# A Good Explanation of Differential Gears
There is a very good [video](https://youtu.be/67XoCMTcN7M?si=fcAbTL5r125Ypc9Q&t=116) from 1937 by Jam Handy. It explains why we need differential gears and how they work.
I think the video is a masterpiece of an explanation. I recommend you watch it, not because understandin... | https://www.lesswrong.com/posts/vb5SNpLjk5bavsvyB/a-good-explanation-of-differential-gears |
# Trying to understand John Wentworth's research agenda
Oli, you wanna kick us off?
I mostly plan to observe (and make edits/suggestions) here, but I'll chime in if I think there is something particularly important being missed.
Ok, context for this dialogue is that I was evaluating your LTFF application, and you we... | https://www.lesswrong.com/posts/7fq3r4n5CCgYLfsJb/trying-to-understand-john-wentworth-s-research-agenda |
# Genocide isn't Decolonization
In 1987, Bono, the lead singer of U2, interrupted his performance of “[Sunday Bloody Sunday](https://en.wikipedia.org/wiki/Sunday_Bloody_Sunday)” to give a speech about [The Troubles](https://en.wikipedia.org/wiki/The_Troubles), prompted by the [Enniskillen bombing](https://en.wikipedia... | https://www.lesswrong.com/posts/4muSjjmKA8WwFXGTh/genocide-isn-t-decolonization |
# Unpacking the dynamics of AGI conflict that suggest the necessity of a premptive pivotal act
*\[Semi-half baked. I don’t reach any novel conclusions in this post, but I do flesh out my own thinking on the way to generally accepted conclusions. *
*I’m pretty interested in anything here that seems incorrect (includi... | https://www.lesswrong.com/posts/LfGnzX7wm6j8MGWfT/unpacking-the-dynamics-of-agi-conflict-that-suggest-the |
# Announcing new round of "Key Phenomena in AI Risk" Reading Group
**TLDR: **“[Key Phenomena in AI Risk](https://www.lesswrong.com/posts/mqvxR9nrXAzRr3ow9/announcing-key-phenomena-in-ai-risk-facilitated-reading)” is an 8-week-long, facilitated reading group. It is aimed at people interested in conceptual AI alignment ... | https://www.lesswrong.com/posts/vakhhNHduW9gmENTW/announcing-new-round-of-key-phenomena-in-ai-risk-reading |
# AI Safety Hub Serbia Soft Launch
TLDR; We got three-month funding from a generous individual funder to launch an AI Safety office in Serbia. We are giving free office space (and, if funding later permits, housing) to AI Safety researchers who are looking for a place to work from. Priority for citizens of countries l... | https://www.lesswrong.com/posts/CmvkoyTq49tFkSGFF/ai-safety-hub-serbia-soft-launch |
# Features and Adversaries in MemoryDT
**Keywords**: Mechanistic Interpretability, Adversarial Examples, GridWorlds, Activation Engineering
This is part 2 of [A Mechanistic Interpretability Analysis of a GridWorld Agent Simulator](https://www.lesswrong.com/posts/JvQWbrbPjuvw4eqxv/a-mechanistic-interpretability-analys... | https://www.lesswrong.com/posts/yuQJsRswS4hKv3tsL/features-and-adversaries-in-memorydt |
# Revealing Intentionality In Language Models Through AdaVAE Guided Sampling
## Introduction
> The universe is already its own model, that is why it seems so hard to model,
> but really it is simple. All that needs to be done is to add Mu back into a
> transformer. "The universe is already here, you just have t... | https://www.lesswrong.com/posts/4Hnso8NMAeeYs8Cta/revealing-intentionality-in-language-models-through-adavae |
# How To Socialize With Psycho(logist)s
Therapy is great. So are therapists!
But being *friends* with a therapist is a very different kind of relationship than being the *patient* of one, and I believe the differences highlight something important about all relationships.
They Specialize In People
==================... | https://www.lesswrong.com/posts/DF5m9WzcsL6TeDBvz/how-to-socialize-with-psycho-logist-s |
# Are (at least some) Large Language Models Holographic Memory Stores?
*Cross-posted* [*from New Savanna*](https://new-savanna.blogspot.com/2023/10/are-at-least-some-large-language-models.html).
That’s been on my mind for the last week or two, ever since my recent work on ChatGPT’s memory for texts \[1\]. On the othe... | https://www.lesswrong.com/posts/o6SRn4TSxZYBzy8h5/are-at-least-some-large-language-models-holographic-memory |
# On the proper date for solstice celebrations
# Convenience vs. accuracy
The question of the "correct" date on which to hold [winter solstice events](https://www.lesswrong.com/tag/secular-solstice) has the character of a [hyperstition](https://www.astralcodexten.com/p/give-up-seventy-percent-of-the-way), i.e. it's n... | https://www.lesswrong.com/posts/RChzPW8zJ99rxxsvq/on-the-proper-date-for-solstice-celebrations |
# Internal Target Information for AI Oversight
*Thanks to Arun Jose for discussions and feedback.*
Summary
=======
In this short post, we discuss the concept of *Internal Target Information* within agentic AI systems, arguing that agentic systems possess internal information about their targets. This information, we... | https://www.lesswrong.com/posts/hhKpXEsfAiyFLecyF/internal-target-information-for-ai-oversight |
# I Would Have Solved Alignment, But I Was Worried That Would Advance Timelines
The alignment community is ostensibly a group of people concerned about AI risk. Lately, it would be more accurate to describe it as a group of people concerned about AI timelines.
AI timelines have some relation to AI risk. Slower timeli... | https://www.lesswrong.com/posts/Eu8y4cTxM3pAzwdCf/i-would-have-solved-alignment-but-i-was-worried-that-would |
# The Overkill Conspiracy Hypothesis
We couldn’t get the lighting exactly right otherwise.
The ter... | https://www.lesswrong.com/posts/DGsPeE89N93YCwxbH/the-overkill-conspiracy-hypothesis |
# TOMORROW: the largest AI Safety protest ever!
Tomorrow, [PauseAI](https://pauseai.info/) and collaborators are putting on the largest AI Safety protest to date, across 7 locations in 6 countries. All are eagerly welcomed!
Your presence at this protest is a rare impact opportunity when in-person volunteering is... | https://www.lesswrong.com/posts/abBtKF857Ejsgg9ab/tomorrow-the-largest-ai-safety-protest-ever |
# Holly Elmore and Rob Miles dialogue on AI Safety Advocacy
Holly is an independent AI Pause organizer, which includes organizing protests (like this [upcoming one](https://www.lesswrong.com/events/ZoTkRYdqGuDCnojMW/global-pause-ai-protest-10-21)). Rob is an [AI Safety YouTuber](https://www.youtube.com/@RobertMilesAI)... | https://www.lesswrong.com/posts/gDijQHHaZzeGrv2Jc/holly-elmore-and-rob-miles-dialogue-on-ai-safety-advocacy |
# What's Hard About The Shutdown Problem
I’ve been having some [productive back-and-forth](https://www.lesswrong.com/posts/sHGxvJrBag7nhTQvb/invulnerable-incomplete-preferences-a-formal-statement-1?commentId=XpY4YxF4tFfGzNbbK) with Sami Petersen about his [Invulnerable Incomplete Preferences](https://www.lesswrong.com... | https://www.lesswrong.com/posts/iJofoQX7EjMFxDo6m/what-s-hard-about-the-shutdown-problem |
# Muddling Along Is More Likely Than Dystopia
**Summary:** There are historical precedents where bans or crushing regulations stop the progress of technology in one industry, while progress in the rest of society continues. This is a plausible future for AI.
*Epistemic Status: My intuition strongly disagrees with oth... | https://www.lesswrong.com/posts/pAnvMYd9mqDT97shk/muddling-along-is-more-likely-than-dystopia |
# Apply for MATS Winter 2023-24!
[Applications are now open](https://airtable.com/appxum3Sqh7TdDvdg/shrtfHWhRFZdkhaIM) for the Winter 2023-24 cohort of [MATS](https://www.matsprogram.org/home) (previously SERI MATS). Our mentors are: [Adrià Garriga Alonso](https://www.lesswrong.com/users/rhaps0dy), [Alex Turner](https... | https://www.lesswrong.com/posts/tqyg3DpoiE4DKyi4y/apply-for-mats-winter-2023-24 |
# Alignment Implications of LLM Successes: a Debate in One Act
**Doomimir**: Humanity has made no progress on the alignment problem. Not only do we have no clue how to align a powerful optimizer to our "true" values, we don't even know how to make AI "corrigible"—willing to let us correct it. Meanwhile, capabilities c... | https://www.lesswrong.com/posts/pYWA7hYJmXnuyby33/alignment-implications-of-llm-successes-a-debate-in-one-act |
# Soups as Spreads
We tend to cook in quantities where there are leftovers: it's not really practical to cook exactly the right amount of food, and it's much better to have a bit extra than not enough. Plus with several people working from home we tend to go through a good amount of leftovers for lunches. One thing I'... | https://www.lesswrong.com/posts/mRdykbSJaLZaNPNaW/soups-as-spreads |
# How toy models of ontology changes can be misleading
Before solving complicated problems (such as reaching a decision with thousands of variables and complicated dependencies) it helps to focus on solving simpler problems first (such as utility problems with three clear choices), and then gradually building up. Afte... | https://www.lesswrong.com/posts/XvhrmTog2bkf5s2qu/how-toy-models-of-ontology-changes-can-be-misleading |
# Best effort beliefs
Alice: Hey Bob, how's it going?
Bob: Pretty good. How about you?
Alice: Pretty good. What are you drinking?
Bob: A Bayeslight.
Alice: Cool. Hey bartender, lemme get one of those as well.
Alice: Hey, have you seen President Hanson's new immigration policy?
Bob: No. What is it?
Alice: He's s... | https://www.lesswrong.com/posts/u2qWeZe6ZLnkregEe/best-effort-beliefs |
# Thoughts On (Solving) Deep Deception
*Thanks to Paul Colognese for conversations that contributed to the thoughts in this post being developed, and to Johannes C. Mayer, Tassilo Neubauer, and Q for feedback on a draft.*
Back in 2000, a computer scientist named Charles Ofria was studying the evolution of simulated o... | https://www.lesswrong.com/posts/Lm8vTwXdDMEojR85A/thoughts-on-solving-deep-deception-1 |
# Announcing Timaeus
[Timaeus](https://timaeus.co/) is a new AI safety research organization dedicated to making fundamental breakthroughs in technical AI alignment using deep ideas from mathematics and ... | https://www.lesswrong.com/posts/nN7bHuHZYaWv9RDJL/announcing-timaeus |
# The Drowning Child
*Epistemic status: possibly very evil devil's advocacy, philospohicly unsophisticard ruminations*
We are all familiar with Singer's drowning child thought experiment. I have often found it very compelling, or at least felt I ought to find it very compelling. My revealed preferences, as with those... | https://www.lesswrong.com/posts/nKYFcbpkts3r5Aaoe/the-drowning-child |
# What is an "anti-Occamian prior"?
I've seen references to "anti-Occamian priors" [in the Sequences](https://www.lesswrong.com/posts/C8nEXTcjZb9oauTCW/where-recursive-justification-hits-bottom), where Eliezer was talking about how not all possible minds would agree with Occam's Razor. I'm not sure how such a prior co... | https://www.lesswrong.com/posts/oLcrJ9awurhnhvwqu/what-is-an-anti-occamian-prior |
# Contra Dance Dialect Survey
Fifteen years ago I travelled to a bunch of dances and wrote something up ( [pdf](https://www.jefftk.com/final-papers/thesis/contra-thesis-1-0-0.pdf)) on how people tended to dance in different areas. I'm curious how this has changed over the years, and while I have my observations I'd li... | https://www.lesswrong.com/posts/rTBBcJzEpttsQ8EdD/contra-dance-dialect-survey |
# VLM-RM: Specifying Rewards with Natural Language
tl;dr
-----
We show how to use Vision-Language Models (VLM), and specifically CLIP models, as reward models (RM) for RL agents. Instead of manually specifying a reward function, we only need to provide text prompts like “a humanoid robot kneeling” to instruct and pro... | https://www.lesswrong.com/posts/vyxwgQnWPdhpWQ9ZN/vlm-rm-specifying-rewards-with-natural-language |
# Machine Unlearning Evaluations as Interpretability Benchmarks
Interpreting Models by Ablation. Image generated by DALL-E 3.
Introduction
============
Interpretability in machine learning, especially ... | https://www.lesswrong.com/posts/mTi8TQEyP5Pr7oczd/machine-unlearning-evaluations-as-interpretability |
# Programmatic backdoors: DNNs can use SGD to run arbitrary stateful computation
*Thanks to Kshitij Sachan for helpful feedback on the draft of this post.*
If you train a neural network with SGD, you can embed within the weights of the network any state machine: the network encodes the state $s_t$ in its weights, use... | https://www.lesswrong.com/posts/QNQuWB3hS5FrGp5yZ/programmatic-backdoors-dnns-can-use-sgd-to-run-arbitrary |
# Some of my predictable updates on AI
Introduction
------------
Author note: I’m struggling to write this in a way I’m happy with. Rather than having it sit in my drafts, I’m going to post in an unfinished state now. I don’t think I endorse sharing/upvoting this much, but if you find the content particularly compell... | https://www.lesswrong.com/posts/pE93AGxXNiqGA6Brq/some-of-my-predictable-updates-on-ai |
# AI Alignment [Incremental Progress Units] this Week (10/22/23)
Sorry for getting this one out a day late.
Before getting started, one minor update. I liked David Orr’s [suggestion](https://www.lesswrong.com/posts/zkDCCSR3o2aBkTcbq/ai-alignment-incremental-progress-units-this-week-10-08-23?commentId=oKYxLnkjEMfNwoTQ... | https://www.lesswrong.com/posts/2TBTjychCgn2nyYZA/ai-alignment-incremental-progress-units-this-week-10-22-23 |
# The Shutdown Problem: An AI Engineering Puzzle for Decision Theorists
\[NOTE: This paper was previously titled 'The Shutdown Problem: Three Theorems'.\]
This paper is an updated version of the first half of my [AI Alignment Awards](https://www.lesswrong.com/posts/zFoAAD7dfWdczxoLH#Elliott_Thornley___16_000_) contes... | https://www.lesswrong.com/posts/8GWLRMnp55iFZDBbm/the-shutdown-problem-three-theorems |
# Open Source Replication & Commentary on Anthropic's Dictionary Learning Paper
*This is the long-form version of* [*a public comment*](https://transformer-circuits.pub/2023/monosemantic-features/index.html#comment-nanda) *on Anthropic's Towards Monosemanticity paper*
Introduction
------------
Anthropic recently put... | https://www.lesswrong.com/posts/fKuugaxt2XLTkASkk/open-source-replication-and-commentary-on-anthropic-s |
# Towards Understanding Sycophancy in Language Models
**TL;DR: **We show sycophancy is a general behavior of RLHF’ed AI assistants in varied, free-form text-generation settings, extending previous results. Our experiments suggest these behaviors are likely driven in part by imperfections in human preferences–both huma... | https://www.lesswrong.com/posts/g5rABd5qbp8B4g3DE/towards-understanding-sycophancy-in-language-models |
# Human wanting
*[Metadata: crossposted from [https://tsvibt.blogspot.com/2023/08/human-wanting.html](https://tsvibt.blogspot.com/2023/08/human-wanting.html). First completed August 22, 2023.]*
We have pretheoretic ideas of wanting that come from our familiarity with human wanting, in its variety. To see what way of ... | https://www.lesswrong.com/posts/YLRPhvgN4uZ6LCLxw/human-wanting |
# AI Pause Will Likely Backfire (Guest Post)
_I'm experimenting with hosting guest posts on this blog, as a way to represent additional viewpoints and especially to highlight ideas from researchers who do not already have a platform. Hosting a post does not mean that I agree with all of its arguments, but it does mean... | https://www.lesswrong.com/posts/3siLbdd4338gfTM7g/ai-pause-will-likely-backfire-guest-post |
# Linkpost: A Post Mortem on the Gino Case
As a followup to my [previous linkpost](https://www.lesswrong.com/posts/LizpBcsF9XEAhTsvn/linkpost-they-studied-dishonesty-was-their-work-a-lie) to the [New Yorker article](https://www.newyorker.com/magazine/2023/10/09/they-studied-dishonesty-was-their-work-a-lie) covering th... | https://www.lesswrong.com/posts/QzfBkbasYhxTmtFyW/linkpost-a-post-mortem-on-the-gino-case |
# Announcing #AISummitTalks featuring Professor Stuart Russell and many others
*Register now:* [*#AISummitTalks*](https://www.facebook.com/hashtag/aisummittalks?__eep__=6&__cft__[0]=AZWWp2KX-5gEcFfkRo4Lxqg1YojXwaT6ZG89wRpvxIizJV-guP3ZZ7H9-0cxf5gsF-LnGdxKiywZLlV58vPiaQTl14pc8yaytztqw7GO_yhx-SezQj7FlTV5XK2oD9gYJJ3ULkwRt... | https://www.lesswrong.com/posts/NaXz3FM9gXXB7oJW3/announcing-aisummittalks-featuring-professor-stuart-russell |
# Anyone Else Using Brilliant?
I started using Brilliant, and so far I've found it to be a lot like Thinking Physics - teaching by posing real-world conundrums and then explaining the concepts and/or math behind the answers.
Anyone else getting something out of it, or have advice for how to use it? | https://www.lesswrong.com/posts/HmeKF2mRRLfkG68X4/anyone-else-using-brilliant |
# Lying is Cowardice, not Strategy
*(Co-written by* [*Connor Leahy*](https://twitter.com/npcollapse) *and* [*Gabe*](https://twitter.com/Gabe_cc)*)*
We have talked to a whole bunch of people about pauses and moratoriums. Members of the AI safety community, investors, business peers, politicians, and more.
Too many cl... | https://www.lesswrong.com/posts/qtTW6BFrxWw4iHcjf/lying-is-cowardice-not-strategy |
# [Interview w/ Quintin Pope] Evolution, values, and AI Safety
On the Futurati Podcast, we recently released an [interview with Quintin Pope](https://www.youtube.com/watch?v=XLDdG9DR7ek).
As you can imagine, it mostly focused on:
* Inner v.s. outer optimization;
* The sharp left turn;
* Natural selection, and ... | https://www.lesswrong.com/posts/c9W4SHa7DHwAHkqRF/interview-w-quintin-pope-evolution-values-and-ai-safety |
# Book Review: Going Infinite
Previously: [Sadly, FTX](https://thezvi.substack.com/p/sadly-ftx)
I doubted whether it would be a good use of time to read Michael Lewis’s new book [Going Infinite](https://www.amazon.com/Going-Infinite-Rise-Fall-Tycoon/dp/B0CD8V9SHD/ref=sr_1_1?keywords=going+infinite&qid=1697403431&sr=8... | https://www.lesswrong.com/posts/AocXh6gJ9tJC2WyCL/book-review-going-infinite |
# Who is Harry Potter? Some predictions.
Microsoft has released a paper called "who is Harry Potter" in which they claim to make a neural network forget who Harry Potter is.
[https://www.microsoft.com/en-us/research/project/physics-of-agi/articles/whos-harry-potter-making-llms-forget-2/](https://www.microsoft.com/en... | https://www.lesswrong.com/posts/B4vgbeXMGxEnEwY8d/who-is-harry-potter-some-predictions |
# Thoughts on responsible scaling policies and regulation
I am excited about AI developers implementing [responsible scaling policies](https://evals.alignment.org/blog/2023-09-26-rsp/); I’ve recently been spending time refining this idea and advocating for it. Most people I talk to are excited about RSPs, but there is... | https://www.lesswrong.com/posts/dxgEaDrEBkkE96CXr/thoughts-on-responsible-scaling-policies-and-regulation |
# How to Resolve Forecasts With No Central Authority?
I've been chatting with some people about using Prediction Markets/Forecasts alongside Community Notes.
An initial suggestion is just to link to to forecasting sites really well. *It's a good suggestion but I'm not gonna discuss that here.*
Other suggestions revo... | https://www.lesswrong.com/posts/7kyeFWrS7ni7gxj87/how-to-resolve-forecasts-with-no-central-authority |
# Verifiable private execution of machine learning models with Risc0?
Risc0 is a virtual machine that runs any riscv (*C, C++, Rust, Go, and Nim can all compile to riscv*) program, records a trace of the state of the computation over time, and then, no matter how long the execution ran for, it can fold that execution ... | https://www.lesswrong.com/posts/SJdYkx6C2KXQyeKro/verifiable-private-execution-of-machine-learning-models-with |
# What is a Sequencing Read?
Probably the most common form of [genetic sequencing](https://www.jefftk.com/p/sequencing-intro) these days is "paired-end" sequencing. It's very impressive: the sequencing machine can process the same nucleic acid fragment from both ends! This means that each observation looks like:
+---... | https://www.lesswrong.com/posts/JQmgoLAkFKkhxLMhz/what-is-a-sequencing-read |
# Announcing Epoch's newly expanded Parameters, Compute and Data Trends in Machine Learning database
The performance of machine learning models is closely related to their amount of training data, compute, and number of parameters. At Epoch, we’re investigating the key inputs that enable today’s AIs to reach new heigh... | https://www.lesswrong.com/posts/LKDFkwFpWfgGpGs8d/announcing-epoch-s-newly-expanded-parameters-compute-and |
# Should the US House of Representatives adopt rank choice voting for leadership positions?
One obvious take, partisanship and party power interest, might suggest a strong "No!" type of response. I would certainly expect to have many current Representatives to respond that way if floated as an idea today. But perhaps ... | https://www.lesswrong.com/posts/6ehdEpSonYzE72yL2/should-the-us-house-of-representatives-adopt-rank-choice |
# Compositional preference models for aligning LMs
*This post summarizes the main results from our recently released paper *[*Compositional preference models for aligning LMs*](https://arxiv.org/abs/2310.13011) *and puts them in the broader context of AI safety. For a quick summary of the paper, take a look at our*[*T... | https://www.lesswrong.com/posts/oSgac8x8fgNj22ky3/compositional-preference-models-for-aligning-lms |
# Anthropic, Google, Microsoft & OpenAI announce Executive Director of the Frontier Model Forum & over $10 million for a new AI Safety Fund
Today, Anthropic, Google, Microsoft, and OpenAI are announcing the selection of Chris Meserole as the first Executive Director of the Frontier Model Forum, and the creation of a ... | https://www.lesswrong.com/posts/5jpESFymqEgSAmDJL/anthropic-google-microsoft-and-openai-announce-executive |
# Lying to chess players for alignment
Eliezer Yudkowsky recently [posted on Facebook](https://www.facebook.com/yudkowsky/posts/pfbid05pVZ6QH5HhPTwJdmWMcLN5nws9aeC4gywmUv88QRhEnBUsdJas5KWC9EnDGJhSXrl) an experiment that could potentially indicate whether humans can "have AI do their alignment homework" despite not bei... | https://www.lesswrong.com/posts/ddsjqwbJhD9dtQqDH/lying-to-chess-players-for-alignment |
# My hopes for alignment: Singular learning theory and whole brain emulation
*Some prerequisites needed in order for this to make sense: *
1. [*Two Subsystems: Learning & Steering*](https://www.lesswrong.com/s/HzcM2dkCq7fwXBej8/p/hE56gYi5d68uux9oM)
2. [*Shard Theory: An Overview*](https://www.lesswrong.com/posts/xq... | https://www.lesswrong.com/posts/d4qbjx35SBMGyFNWZ/my-hopes-for-alignment-singular-learning-theory-and-whole |
# AI as a science, and three obstacles to alignment strategies
AI used to be a science. In the old days (back when AI didn't work very well), people were attempting to develop a working theory of cognition.
Those scientists didn’t succeed, and those days are behind us. For most people working in AI today and dividing... | https://www.lesswrong.com/posts/JcLhYQQADzTsAEaXd/ai-as-a-science-and-three-obstacles-to-alignment-strategies |
# Responsible Scaling Policies Are Risk Management Done Wrong
Summary
=======
TLDR
----
Responsible Scaling Policies (RSPs) have been [recently](https://evals.alignment.org/blog/2023-09-26-rsp) [proposed](https://www.anthropic.com/index/anthropics-responsible-scaling-policy) as a way to keep scaling frontier large l... | https://www.lesswrong.com/posts/9nEBWxjAHSu3ncr6v/responsible-scaling-policies-are-risk-management-done-wrong |
# EA Infrastructure Fund: June 2023 grant recommendations
Introduction
============
This payout report covers the Effective Altruism Infrastructure Fund (EAIF)’s grantmaking starting January 2022 (after our [December 2021 payout report](https://forum.effectivealtruism.org/posts/R3mqzQhYabjBcSk7S/ea-infrastructure-fun... | https://www.lesswrong.com/posts/bBnxGAc4NT9aRdEtL/ea-infrastructure-fund-june-2023-grant-recommendations |
# Architects of Our Own Demise: We Should Stop Developing AI Carelessly
Some brief thoughts at a difficult time in the AI risk debate.
Imagine you go back in time to the year 1999 and tell people that in 24 years time, humans will be on the verge of building weakly superhuman AI systems. I remember watching the anime... | https://www.lesswrong.com/posts/bHHrdXwrCj2LRa2sW/architects-of-our-own-demise-we-should-stop-developing-ai |
# CHAI internship applications are open (due Nov 13)
[CHAI](https://humancompatible.ai/) internship applications have just opened, [apply here](https://boards.greenhouse.io/centerforhumancompatibleartificialintelligence/jobs/4358062002) by Nov 13th! The internship might be a good fit if you want to get research experi... | https://www.lesswrong.com/posts/Xa4b8vgCLRATiqnJn/chai-internship-applications-are-open-due-nov-13 |
# Apply to the Constellation Visiting Researcher Program and Astra Fellowship, in Berkeley this Winter
> *This is a link post for two AI safety programs we’ve just opened applications for:* [*https://www.constellation.org/programs/astra-fellowship*](https://www.constellation.org/programs/astra-fellowship) *and* [*http... | https://www.lesswrong.com/posts/oBdfDvmrBKoTq3x85/apply-to-the-constellation-visiting-researcher-program-and |
# RA Bounty: Looking for feedback on screenplay about AI Risk
[Jamie Wahls](http://jamiewahls.com/) and Arthur Frost are writing a screenplay for Rational Animations. It's a sci-fi episodic comedy illustrating various ways in which AI goes off the rails in more or less catastrophic ways.
We're looking for feedback a... | https://www.lesswrong.com/posts/GxYtjKepFwt3DS7th/ra-bounty-looking-for-feedback-on-screenplay-about-ai-risk |
# AI #35: Responsible Scaling Policies
There is much talk about so-called Responsible Scaling Policies, as in what we will do so that what we are doing can be considered responsible. Would that also result in actually responsible scaling? It would help. By themselves, in their current versions, no. The good scenario i... | https://www.lesswrong.com/posts/aQ6LDhc2zxrYXFjEF/ai-35-responsible-scaling-policies |
# EPUBs of MIRI Blog Archives and selected LW Sequences
[Here is a repository of EPUBs](https://git.sr.ht/~mesaoptimizer/epubs) that I've created, consisting of rationalist or alignment content I want to read on my Kindle, such as selections of... | https://www.lesswrong.com/posts/fejT5iMEDcZMHGCSW/epubs-of-miri-blog-archives-and-selected-lw-sequences |
# 0. The Value Change Problem: introduction, overview and motivations
Given that AI systems are becoming increasingly capable and widely deployed, we have to think carefully about the multifarious effects their adoption will have on our individual and collective lives. In this sequence, I will focus specifically on h... | https://www.lesswrong.com/posts/mHQHBEuFcEWRnitp4/0-the-value-change-problem-introduction-overview-and |
# 1. Premise one: Values are malleable
In this post, I will defend the first premise of the Value Change Problem: *that realistic models of human values take them to be malleable, rather than fixed*. I call this the ’value malleability’ claim (VM).
I start by clarifying what I mean by value malleability before maki... | https://www.lesswrong.com/posts/HyodRjYtiA2xozCrk/1-premise-one-values-are-malleable |
# 2. Premise two: Some cases of value change are (il)legitimate
In the prior post, I have defended the claim that genuine value change is possible, and thus, that a realistic account of human values understands them to be malleable. In this section, I will argue for the claim that *some cases of value change are legit... | https://www.lesswrong.com/posts/QjA6kipHYqwACkPNw/2-premise-two-some-cases-of-value-change-are-il-legitimate |
# 3. Premise three & Conclusion: AI systems can affect value change trajectories & the Value Change Problem
In this post, I introduce the last of three premises—the claim that *AI systems are (and will become increasingly) capable of affecting people’s value change trajectories*. With all three premises in place, we ... | https://www.lesswrong.com/posts/yPnAzeRAqdko3RNtR/3-premise-three-and-conclusion-ai-systems-can-affect-value |
# 4. Risks from causing illegitimate value change (performative predictors)
Unaligned AI systems may cause illegitimate value change. At the heart of this risk lies the observation that the malleability inherent to human values can be exploited in ways that make the resulting value change illegitimate. Recall that I t... | https://www.lesswrong.com/posts/qZFGPJi3u8xuvnWHQ/4-risks-from-causing-illegitimate-value-change-performative |
# 5. Risks from preventing legitimate value change (value collapse)
The second risk that follows from the VCP concerns the obstruction of legitimate value change. In what follows, I will consider the mechanisms by which legitimate value change might come to be impeded, and whether and when this might constitute an act... | https://www.lesswrong.com/posts/KeHGinpj2WyzDEQAx/5-risks-from-preventing-legitimate-value-change-value |
# 5 psychological reasons for dismissing x-risks from AGI
People fiercely argue whether AGI will likely be an existential threat or not.
Most of the arguments explore [conceptual](https://www.lesswrong.com/posts/oM9pEezyCb4dCsuKq/pausing-ai-developments-isn-t-enough-we-need-to-shut-it-all-1#Addendum__March_30__),... | https://www.lesswrong.com/posts/yPfotACWkRdWbubfy/5-psychological-reasons-for-dismissing-x-risks-from-agi |
# Changing Contra Dialects
In 2008 I wrote my undergraduate linguistics thesis on the dialects of contra dancing ( [pdf](https://www.jefftk.com/final-papers/thesis/contra-thesis-1-0-0.pdf)). Some figures have variation, but that variation is constrained by the need for compatible dancing. For example, there are three ... | https://www.lesswrong.com/posts/qoSXGdvvwrHXXadDb/changing-contra-dialects |
# Disagreements over the prioritization of existential risk from AI
Earlier this year, the [Future of Life Institute](https://futureoflife.org/open-letter/pause-giant-ai-experiments/) and the [Center for AI Safety](https://www.safe.ai/statement-on-ai-risk) published open letters that promoted existential risks (x-risk... | https://www.lesswrong.com/posts/eAT2dXAngXxFRTQLn/disagreements-over-the-prioritization-of-existential-risk |
# Nonlinear limitations of ReLUs
A neural net using rectified linear unit activation functions of any size is unable to approximate the function sin(x) outside a compact interval.
I am reasonably confident that I can prove that any NN with ReLU activation approximates a piecewise linear function. I believe the number... | https://www.lesswrong.com/posts/eLzTDr3A5xsD7zvqK/nonlinear-limitations-of-relus |
# Symbol/Referent Confusions in Language Model Alignment Experiments
**Simon Strawman**: Here’s an example shamelessly ripped off from [Zack’s recent post](https://www.lesswrong.com/posts/pYWA7hYJmXnuyby33/alignment-implications-of-llm-successes-a-debate-in-one-act), showing corrigibility in a language model:
, and [they’re hiring](https://openai.com/careers/search?c=preparedness)!
This team is going to ... | https://www.lesswrong.com/posts/r9hoMiraPuvq3Bqxj/openai-s-new-preparedness-team-is-hiring |
# Online Dialogues Party — Sunday 5th November
Join us on our quest to have the most curious and collaborative conversations on the public internet!
People using LessWrong to have dialogues with each ot... | https://www.lesswrong.com/events/BJcNeJss4jxc68GQR/online-dialogues-party-sunday-5th-november |
# To what extent is the UK Government's recent AI Safety push entirely due to Rishi Sunak?
And hence, to what extent might we expect all that momentum to suddenly dissipate?
In other words, if Sunak was to [suddenly leave office](https://manifold.markets/SimonGrayson/will-rishi-sunak-be-replaced-as-pri-9f3c9f888743) ... | https://www.lesswrong.com/posts/9x6hGFdkpWECErpbN/to-what-extent-is-the-uk-government-s-recent-ai-safety-push |
# ASPR & WARP: Rationality Camps for Teens in Taiwan and Oxford
The [FABRIC](http://fabric.camp) team (formerly known as ESPR) is running two **immersive applied rationality workshops** for analytical students **aged 16 to 19** seeking to understand themselves and the world.
**Asian Spring Programme on Rationality (*... | https://www.lesswrong.com/posts/suoejqiP23Yfk5CiG/aspr-and-warp-rationality-camps-for-teens-in-taiwan-and |
# Linkpost: Rishi Sunak's Speech on AI (26th October)
I’m delighted to be here at the Royal Society, the place where the story of modern science has been written for centuries.
Now, I’m unashamedly optimistic about the power of technology to make life better for everyone.
So, the easy speech for me to give - the one... | https://www.lesswrong.com/posts/94nYiPnr34kmHLMrB/linkpost-rishi-sunak-s-speech-on-ai-26th-october |
# Aspiration-based Q-Learning
*Work completed during a two-month internship supervised by* [*@Jobst Heitzig*](https://www.lesswrong.com/users/jobst-heitzig?mention=user)*. *
*Thanks to Phine Schikhof for her invaluable conversations and friendly support during the internship, and to Jobst Heitzig, who was an amazing ... | https://www.lesswrong.com/posts/Z9P2m462wQ4qmH6uo/aspiration-based-q-learning |
# We're Not Ready: thoughts on "pausing" and responsible scaling policies
_Views are my own, not Open Philanthropy’s. I am married to the President of [Anthropic](https://www.anthropic.com/) and have a financial interest in both Anthropic and OpenAI via my spouse._
Over the last few months, I’ve spent a lot of my tim... | https://www.lesswrong.com/posts/Np5Q3Mhz2AiPtejGN/we-re-not-ready-thoughts-on-pausing-and-responsible-scaling-4 |
# Wireheading and misalignment by composition on NetHack
**TL;DR**: We find agents trained with RLAIF to indulge in wireheading in NetHack. Misalignment appears when the agent optimizes a combination of two rewards that produce aligned behaviors when optimized in isolation, and only emerges with some prompt wordings. ... | https://www.lesswrong.com/posts/GEjzyf7Hjpv9g2uGX/wireheading-and-misalignment-by-composition-on-nethack |
# Sanctuary for Humans
TL;DR: If we succeed at AI safety, humans will probably decide the future of the universe, and we currently have strong self-preservation incentives to choose a future populated by humans. If we committed to provide sanctuary for all currently alive humans, this would make our decision making pr... | https://www.lesswrong.com/posts/JbpLAAeAHh5RFiYtn/sanctuary-for-humans |
# Techno-humanism is techno-optimism for the 21st century
Lately I’ve been reading about the history of economic thought, with the goal of understanding how today’s foundational ideas were originally developed. My biggest takeaway has been that economics is more of a moving target than I’d realized. Early economists w... | https://www.lesswrong.com/posts/vdsbqq2xjniaGKE5J/techno-humanism-is-techno-optimism-for-the-21st-century |
# Value systematization: how values become coherent (and misaligned)
Many discussions of AI risk are unproductive or confused because it’s hard to pin down concepts like “coherence” and “expected utility maximization” in the context of deep learning. In this post I attempt to bridge this gap by describing a process by... | https://www.lesswrong.com/posts/J2kpxLjEyqh6x3oA4/value-systematization-how-values-become-coherent-and |
# Do you believe "E=mc^2" is a correct and/or useful equation, and, whether yes or no, precisely what are your reasons for holding this belief (with such a degree of confidence)?
Please just answer the question as best you can. E.g. your answer might be "because everyone on TV says so" or "because I saw mass transform... | https://www.lesswrong.com/posts/5GciFbQjBQB48Nxyd/do-you-believe-e-mc-2-is-a-correct-and-or-useful-equation |
# Truthseeking, EA, Simulacra levels, and other stuff
I've been on this big kick talking about truthseeking in effective altruism. I started with vegan advocacy because it was the most legible, but now need to move on to the deeper problems. Unfortunately those problems are still not that legible, and I end up having ... | https://www.lesswrong.com/posts/cY43KWjamafLbij9C/truthseeking-ea-simulacra-levels-and-other-stuff |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.