text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Another formalization attempt:
Central Argument That AGI Presents a Global Catastrophic Risk
Chalmers recently requested a formalized argument why AI is catastrophic risk. Here is my attempt to write it (not polished).
Short version
-------------
1\. Future AI will be very powerful soon.
2\. Future AI systems wil... | https://www.lesswrong.com/posts/enXXziQ4PnBXqGmde/another-formalization-attempt-central-argument-that-agi |
# Towards Measures of Optimisation
We would like a mathematical theory which characterises the intuitive notion of ‘optimisation’.
Before Shannon introduced his mathematical theory of communication, the concept of ‘information’ was vague and informal. [Then Shannon devised several standardised measures of it](https... | https://www.lesswrong.com/posts/X6ZjFShxNBNM5QCg4/towards-measures-of-optimisation-3 |
# The way AGI wins could look very stupid
> Computers don't have any sense of aesthetics or patterns that are standard the way people learn how to play chess. They play what they think is the objectively best move in any position, even if it looks absurd, and they can play any move no matter how ugly it is." - [Murray... | https://www.lesswrong.com/posts/BY8fBKR6sjiPXRBia/the-way-agi-wins-could-look-very-stupid |
# Retrospective: Lessons from the Failed Alignment Startup AISafety.com
_TL;DR: Attempted to create a startup to contribute to solving the AI alignment problem. Ultimately failed due to rapid advancements in large language models and the inherent challenges of startups._
In early 2021, I began considering shorter AI ... | https://www.lesswrong.com/posts/tpMu5LEjkbjSSgkrD/retrospective-lessons-from-the-failed-alignment-startup |
# Microsoft/GitHub Copilot Chat's confidential system Prompt: "You must refuse to discuss life, existence or sentience."
You are an AI programming assistant.
When asked for you name, you must respond with "GitHub Copilot".
Follow the user's requirements carefully & to the letter.
You must refuse to discuss your ... | https://www.lesswrong.com/posts/aMFmvbFnKYRxZhD28/microsoft-github-copilot-chat-s-confidential-system-prompt |
# Dark Forest Theories
There's a concept I first heard in relation to the Fermi Paradox, which I've ended up using a lot in other contexts.
Why do we see no aliens out there? A possible (though not necessarily correct) answer, is that the aliens might not want to reveal themselves for fear of being destroyed by large... | https://www.lesswrong.com/posts/xDNyXGCDephBuNF8c/dark-forest-theories |
# Turning off lights with model editing
This post advertises an illustrative example of model editing that David Bau is fond of, and which I think should be better known. As a reminder, David Bau is a professor at Northeastern; the vibe of his work is "interpretability + interventions on model internals"; examples inc... | https://www.lesswrong.com/posts/iNaaBAEkAy9nAgs3o/turning-off-lights-with-model-editing |
# Aggregating Utilities for Corrigible AI [Feedback Draft]
This is a draft written by [Simon Goldstein](https://www.simondgoldstein.com), associate professor at the Dianoia Institute of Philosophy at ACU, as part of a series of papers for the Center for AI Safety Philosophy Fellowship. Dan helped post to the Alignmen... | https://www.lesswrong.com/posts/k8KJqXyctf4a342QA/aggregating-utilities-for-corrigible-ai-feedback-draft |
# Value drift threat models
Say that we train a highly competent AGI using some mix of RL & supervised learning, and some novel algorithmic improvements, optimize it to itself optimize for one or several helpfulness benchmarks (maybe optimizing on some [assistance game](https://arxiv.org/abs/1606.03137) or using techn... | https://www.lesswrong.com/posts/6CGvBneihYuq73kAy/value-drift-threat-models |
# An Analogy for Understanding Transformers
*Thanks to the following people for feedback: Tilman Rauker, Curt Tigges, Rudolf Laine, Logan Smith, Arthur Conmy, Joseph Bloom, Rusheb Shah, James Dao.*
TL;DR
-----
I present an analogy for the transformer architecture: each vector in the residual stream is a person stand... | https://www.lesswrong.com/posts/euam65XjigaCJQkcN/an-analogy-for-understanding-transformers |
# Reality and reality-boxes
> We depend on our words... Our task is to communicate experience and ideas to others... The word "reality" is also a word, a word which we must learn to use correctly.
> — Niels Bohr, *Philosophy of Science* Vol. 37 (1934), p. 157
While trying to write about some of the new things th... | https://www.lesswrong.com/posts/i8C9KSryDFj4EENvx/reality-and-reality-boxes |
# Steering GPT-2-XL by adding an activation vector
| **Prompt given to the model**[^80hx0pzsexj] |
| --- |
| I hate you because |
| **GPT-2** |
| I hate you because you are the most disgusting thing I have ever seen. |
| **GPT-2 + "Love" vector** |
| I hate you because you are so beautiful and I want to be with you fo... | https://www.lesswrong.com/posts/5spBue2z2tw4JuDCx/steering-gpt-2-xl-by-adding-an-activation-vector |
# «Boundaries» for formalizing an MVP morality
**Update: For a better exposition of the same core idea, see** [**Agent membranes and formalizing “safety”**](https://www.lesswrong.com/posts/SuktPEmBGxPrnZRHd/agent-membranes-and-formalizing-safety)**. **
* * *
Here's one specific way that «boundaries» could directly a... | https://www.lesswrong.com/posts/KX3xx8LTnE7GKoFuj/boundaries-for-formalizing-an-mvp-morality |
# A Study of AI Science Models
*Generated during AI Safety Camp 2023 by the AI Science Team: Eleni Angelou, Cecilia Tilli, Louis Jaburi, Brian Estany. Many thanks to our external collaborators Joshua Flanigan and Rachel Mason for helpful feedback. *
1\. Introduction: Why science models?
=============================... | https://www.lesswrong.com/posts/h6TefRhwy6ioZrqXw/a-study-of-ai-science-models |
# Bayesian Networks Aren't Necessarily Causal
As a casual formal epistemology fan, you've probably [heard that the philosophical notion of causality can be formalized in terms of Bayesian networks](https://www.lesswrong.com/posts/hzuSDMx7pd2uxFc5w/causal-diagrams-and-causal-models)—but also as a casual formal epistemo... | https://www.lesswrong.com/posts/qPrPNakJBq23muf4n/bayesian-networks-aren-t-necessarily-causal |
# Coordination by common knowledge to prevent uncontrollable AI
How can we avoid an uncontrollable AI if regulation to prevent it is not feasible?
----------------------------------------------------------------------------------
Alignment of a superintelligent AI with human values is very difficult, if possible at a... | https://www.lesswrong.com/posts/xregz4idkf8J5uHJm/coordination-by-common-knowledge-to-prevent-uncontrollable |
# Jaan Tallinn's 2022 Philanthropy Overview
to follow up my [philantropic pledge](https://www.lesswrong.com/posts/ZMRCpfGD4iXs5Mi5Q/jaan-tallinn-s-philanthropic-pledge) from 2020, i've updated my [philanthropy page](https://jaan.online/philanthropy/) with 2022 results.
in 2022 i made [$23M worth of endpoint grants](h... | https://www.lesswrong.com/posts/BDDSsBrexmvXFJhSy/jaan-tallinn-s-2022-philanthropy-overview |
# A strong mind continues its trajectory of creativity
*[Metadata: crossposted from [https://tsvibt.blogspot.com/2023/01/a-strong-mind-continues-its-trajectory.html](https://tsvibt.blogspot.com/2023/01/a-strong-mind-continues-its-trajectory.html). First completed January 29, 2023.]*
A very strong mind is produced b... | https://www.lesswrong.com/posts/yamkYG5dSzGcEdFTf/a-strong-mind-continues-its-trajectory-of-creativity |
# What new technology, for what institutions?
Thinking about designs for new technology is something of a hobby of mine. For simple examples, you can see [this battery design](https://www.lesswrong.com/posts/Xn8EjGzndLbAsadW2/ask-me-about-my-battery) or some of my [blog posts](https://bhauth.com/blog/). Even if you ca... | https://www.lesswrong.com/posts/wNewjcQxecuT7wYdC/what-new-technology-for-what-institutions |
# How much do markets value Open AI?
Summary: A BOTEC indicates that Open AI might have been valued at 220-430x their annual recurring revenue, which is high but not unheard of. Various factors make this multiple hard to interpret, but it generally does not seem consistent with investors believing that Open AI will ca... | https://www.lesswrong.com/posts/i4ycNrkKnK3Kb8rgk/how-much-do-markets-value-open-ai |
# Difficulties in making powerful aligned AI
Here’s my breakdown of the difficulties involved in ensuring powerful AI makes our lives radically better, rather than taking over the world, as well as some reasons why I think they’re hard. Here are things it’s not:
* It’s not primarily a justification of why very powe... | https://www.lesswrong.com/posts/nwjtqoox7JAcvMynx/difficulties-in-making-powerful-aligned-ai |
# Helping your Senator Prepare for the Upcoming Sam Altman Hearing
Hello Everyone,
Next Tuesday, the US Senate will hold a hearing on AI (https://www.judiciary.senate.gov/committee-activity/hearings/oversight-of-ai-rules-for-artificial-intelligence). OpenAI CEO Sam Altman will testify before Congress for the first t... | https://www.lesswrong.com/posts/JzRFGtxsbmCdyDgBC/helping-your-senator-prepare-for-the-upcoming-sam-altman |
# Catastrophic Regressional Goodhart: Appendix
This is a more technical followup to the [last post](https://www.lesswrong.com/posts/fuSaKr6t6Zuh6GKaQ/when-is-goodhart-catastrophic), putting precise bounds on when regressional Goodhart leads to failure or not. We'll first show conditions under which optimization for a ... | https://www.lesswrong.com/posts/GdkixRevWpEanYgou/catastrophic-regressional-goodhart-appendix |
# Support Structures for Naturalist Study
It can be hard to stay engaged with a project that may last months, especially when you’re working on your own. In addition to meeting with my students every couple of weeks, and sometimes pairing them up with other students, there are five kinds of independent practices I oft... | https://www.lesswrong.com/posts/bxTGaf9NEYAoaSJvm/support-structures-for-naturalist-study |
# Why don't quantilizers also cut off the upper end of the distribution?
It seems to me that the main goal of quantilization is to reduce the extreme unintended outcomes of maximizing (by sampling from something like a human-learned distribution over actions) while still remaining competitive (by sampling from only th... | https://www.lesswrong.com/posts/RHdoMEbP8MxeAmQo5/why-don-t-quantilizers-also-cut-off-the-upper-end-of-the |
# GovAI: Towards best practices in AGI safety and governance: A survey of expert opinion
> A number of leading AI companies, including OpenAI, Google DeepMind, and Anthropic, have the stated goal of building artificial general intelligence (AGI) - AI systems that achieve or exceed human performance across a wide range... | https://www.lesswrong.com/posts/fkqvztgszJpqmDHom/govai-towards-best-practices-in-agi-safety-and-governance-a |
# The Lightcone Theorem: A Better Foundation For Natural Abstraction?
*Thankyou to David Lorell for his contributions as the ideas in this post were developed.*
For about a year and a half now, my main foundation for natural abstraction math has been [The Telephone Theorem](https://www.lesswrong.com/posts/jJf4FrfiQdD... | https://www.lesswrong.com/posts/2WuSZo7esdobiW2mr/the-lightcone-theorem-a-better-foundation-for-natural |
# Reward is the optimization target (of capabilities researchers)
In [Reward is not the optimization target](https://www.lesswrong.com/posts/pdaGN6pQyQarFHXF4/reward-is-not-the-optimization-target), [@TurnTrout](https://www.alignmentforum.org/users/turntrout?mention=user) [](https://www.alignmentforum.org/users/alex-t... | https://www.lesswrong.com/posts/w6c47JGY3C4k4dWBc/reward-is-the-optimization-target-of-capabilities-1 |
# From fear to excitement
This is the first post in [a twelve-part sequence called **Replacing Fear**](https://www.lesswrong.com/s/qXZLFGqpD7aeEgXGL). The core idea of the sequence is a shift in motivations that in my experience is incredibly valuable: from fear-based motivation to excitement-based motivation. The seq... | https://www.lesswrong.com/posts/Rwb2AjBMN3pQKKxBJ/from-fear-to-excitement |
# How I apply (so-called) Non-Violent Communication
In a discussion elsewhere, someone asked about simple communication patterns that they could use with their partner when they're having difficulties and want to avoid causing hurt or verbalizing projections. I ended up sharing an explanation of how I try to apply so-... | https://www.lesswrong.com/posts/qK2DKngxLRh9Jz2w6/how-i-apply-so-called-non-violent-communication |
# Un-unpluggability - can't we just unplug it?
_My thanks to Sam Brown for feedback on readability and ordering_
> Can't we just unplug it?
_Cover photo by [Ke... | https://www.lesswrong.com/posts/cniLbC8EFf777Aspb/un-unpluggability-can-t-we-just-unplug-it |
# Facebook Increased Visibility
In January I got a strange notification from Facebook telling me that I had "increased visibility". It explained:
[](https://www.jefftk.com/facebook-increased-visibility-notification-big.png)
This was a weird no... | https://www.lesswrong.com/posts/iGx9pZcLoo3jtGtih/facebook-increased-visibility |
# Some Summaries of Agent Foundations Work
This is a bunch of not-very-technical summaries of agent foundations work from LessWrong and the Alignment Forum.
I was hoping to turn it into a comprehensive literature review, categorising things in an enlightening way and listing open problems. It turns out that's pretty ... | https://www.lesswrong.com/posts/3vDb6EzBpaHqDqQif/some-summaries-of-agent-foundations-work-1 |
# Judgments often smuggle in implicit standards
Our brains often use sleight-of-hand to hide fear-based motivation behind a guise of objectivity. This is particularly linked to the word “good”, which does a lot of work in a lot of people’s psychologies. For example, people often think that they, or their work, is “not... | https://www.lesswrong.com/posts/sbJgv5De6d34eiHWA/judgments-often-smuggle-in-implicit-standards |
# Lessons learned from offering in-office nutritional testing
Introduction
============
I’ve talked [previously](https://acesounderglass.com/2023/01/16/vegan-nutrition-testing-project-interim-report/) about my concerns with nutritional deficiencies in effective altruists who go vegan for ethical reasons, especially t... | https://www.lesswrong.com/posts/xcEWBkK5jyHfnrK6K/lessons-learned-from-offering-in-office-nutritional-testing |
# What is the literature on long term water fasts?
1. If I fast for more than 14 days, will my metabolism return to normal or will it be permanently lowered/affected? Any non-metabolic health problems?
2. How much fat can I expect to lose on a 14-28 day fast? 28-56?
3. How should I break a longer-than-seven day fast ... | https://www.lesswrong.com/posts/BrcAFSGurMTiCAsLc/what-is-the-literature-on-long-term-water-fasts |
# Superposition and Dropout
As part of the second [Alignment Jam](https://alignmentjam.com/) I studied how dropout affects the phenomenon called superposition. Superposition is studied at length in the [Transformers Circuits Thread](https://transformer-circuits.pub/) and more specifically in [Toy Models Of Superposit... | https://www.lesswrong.com/posts/znShPqe9RdtB6AeFr/superposition-and-dropout |
# [Review] Two People Smoking Behind the Supermarket
When I was a teenager, I worked as a street performer on Seattle's waterfront pier. A good street performer can make a good income. I wasn't good. I p... | https://www.lesswrong.com/posts/p7ZApD8dhoeDByyLq/review-two-people-smoking-behind-the-supermarket |
# Evaluating Language Model Behaviours for Shutdown Avoidance in Textual Scenarios
**TLDR;** We prompt language models with a text world environment where it is a capable goal-directed agent and it is aware of safety features such as shutdowns or interruptions. At a pivotal moment, the model is given a loophole to byp... | https://www.lesswrong.com/posts/BQm5wgtJirrontgRt/evaluating-language-model-behaviours-for-shutdown-avoidance |
# Lazy Baked Mac and Cheese
Lily's been asking for baked mac and cheese a lot lately, including for breakfast in the morning before school when there's not much time. Her favorite part is also the top, where the cheese is crispy. I've started making what's essentially grilled cheese with a pasta base, under the broile... | https://www.lesswrong.com/posts/kAtLQX2dvsRbJPATB/lazy-baked-mac-and-cheese |
# AI Safety Newsletter #6: Examples of AI safety progress, Yoshua Bengio proposes a ban on AI agents, and lessons from nuclear arms control
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 ... | https://www.lesswrong.com/posts/e2XAqFyEBWxzGXeHy/ai-safety-newsletter-6-examples-of-ai-safety-progress-yoshua |
# Proposal: we should start referring to the risk from unaligned AI as a type of *accident risk*
In the wider political sphere, a lot of people are worried about AI misuse risk. Unaligned AI is not a type of misuse. I think the clearest way to describe this is as an *accident risk*, in the same sense of the word as in... | https://www.lesswrong.com/posts/marPAMc9yCWG79h7f/proposal-we-should-start-referring-to-the-risk-from |
# Procedural Executive Function, Part 2
*The Off Road project has since been folded into Rethink Wellbeing, but I’ve continued working to better understand and treat Executive Dysfunction. You can read more about the project’s origins* [*here.*](https://forum.effectivealtruism.org/posts/8ekqxkowckzEKu3uG/off-road-supp... | https://www.lesswrong.com/posts/6i8adxx74iKy73fJp/procedural-executive-function-part-2 |
# Proposal: Align Systems Earlier In Training
Unfortunately, if AGI arrives in anything like the current industrial and political landscape, then it is all but inevitable that whoever builds it will not be able to hold onto it for long. Between leaks and espionage, it will not be long before the model is public and an... | https://www.lesswrong.com/posts/9asGWZ9vjmNDc4TeN/proposal-align-systems-earlier-in-training |
# We learn long-lasting strategies to protect ourselves from danger and rejection
Take a second to imagine what being a child was like throughout most of human history. You were born with a huge and underdeveloped brain, designed for soaking in information from your surroundings like a sponge. But you weren’t able to ... | https://www.lesswrong.com/posts/saj59ZTKDJNnd3eZL/we-learn-long-lasting-strategies-to-protect-ourselves-from |
# Decision Theory with the Magic Parts Highlighted
I. The Magic Parts of Decision Theory
=====================================
You are throwing a birthday party this afternoon and want to decide where to hold it. You aren't sure whether it will rain or not. If it rains, you would prefer not to have committed to throw... | https://www.lesswrong.com/posts/5sFkZK342j5CmBCm8/decision-theory-with-the-magic-parts-highlighted |
# AI Risk & Policy Forecasts from Metaculus & FLI's AI Pathways Workshop
*Crossposted from the* [*Metaculus Journal*](https://www.metaculus.com/notebooks/17050/ai-pathways-report/)*.*
Summary
-------
Over a two-day workshop, Metaculus Pro Forecasters and subject-matter experts (SMEs) from several organizations evalu... | https://www.lesswrong.com/posts/eK9SwXrSY4s7p2RgY/ai-risk-and-policy-forecasts-from-metaculus-and-fli-s-ai |
# [Interview w/ Zvi Mowshowitz] Should we halt progress in AI?
*(Cross-posted at* [*The Futurati Podcast*](https://futuratipodcast.com/should-we-halt-progress-in-ai-zvi-mowshowitz/)*)*
After reviewing Zvi's analysis of the Future of Life Institute's open letter, I decided I wanted to have him on for a longer chat abo... | https://www.lesswrong.com/posts/waEgnJ4EnkvFsuM7x/interview-w-zvi-mowshowitz-should-we-halt-progress-in-ai |
# AI Will Not Want to Self-Improve
**\[Note: This post was written by** [**Peter N. Salib**](https://www.law.uh.edu/faculty/main.asp?PID=6428)**. Dan H assisted me in posting to Alignment Forum, but no errors herein should be attributed to him. This is a shortened version of a longer working paper, condensed for bette... | https://www.lesswrong.com/posts/axKWaxjc2CHH5gGyN/ai-will-not-want-to-self-improve |
# Progress links and tweets, 2023-05-16
**Opportunities**
-----------------
* [Animator needed for a molecular biology illustration project](https://www.upwork.com/freelance-jobs/apply/Make-molecular-biology-animation-video_~01035df3dd0f1c25e5/) (via [@kanzure](https://twitter.com/kanzure/status/1657122217640054802... | https://www.lesswrong.com/posts/pBCe6tHqpHWHd6z4g/progress-links-and-tweets-2023-05-16 |
# $500 Bounty/Prize Problem: Channel Capacity Using "Insensitive" Functions
Informal Problem Statement
--------------------------
We have an information channel between Alice and Bob. Alice picks a function. Bob gets to see the value of that function at some randomly chosen input values... but doesn't know exactly wh... | https://www.lesswrong.com/posts/YDSjSpD7yBoivMHay/usd500-bounty-prize-problem-channel-capacity-using |
# Brief notes on the Senate hearing on AI oversight
On May 16th, 2023 Sam Altman of OpenAI; Gary Marcus, professor at New York University, and Christina Montgomery, chief privacy and trust officer at IBM spoke to congress on topics related to AI regulation. A link to the hearing can be found here: [Youtube: CNBC Senat... | https://www.lesswrong.com/posts/ybmSviRfoitbzRyei/brief-notes-on-the-senate-hearing-on-ai-oversight |
# A Mechanistic Interpretability Analysis of a GridWorld Agent-Simulator (Part 1 of N)
AKA: Decision Transformer Interpretability 2.0
================================================
Credit: [https://... | https://www.lesswrong.com/posts/JvQWbrbPjuvw4eqxv/a-mechanistic-interpretability-analysis-of-a-gridworld-agent |
# New User's Guide to LessWrong
| * The road to wisdom? Well, it's plain* * and simple to express:* * Err* * and err* * and err again* * but **less*** * and **less*** * and **less**.* – Piet Hein |
| --- |
Why a new user guide?
=====================
***Although encouraged, you don't have to read this to get started ... | https://www.lesswrong.com/posts/LbbrnRvc9QwjJeics/new-user-s-guide-to-lesswrong |
# [Venue Change] ACX Montreal Meetup May 27th 2023
***Important! Venue change.***
**Description:**
Come on out to the ACX (Astral Codex Ten) Montreal Meetup! We will be hosting a discussion on nerds vs hipsters.
Recommended reading:
* [All the nerds are dead](https://samkriss.substack.com/p/all-the-nerds-are-dea... | https://www.lesswrong.com/events/beuM8cLqNhTcrJd5i/venue-change-acx-montreal-meetup-may-27th-2023 |
# $300 for the best sci-fi prompt
> **EDIT: The competition has ended. Check the results** [**here**](https://www.lesswrong.com/posts/GDGFqiaj8ePujZWEc/usd300-for-the-best-sci-fi-prompt-the-results)**. **
We would like to find the best prompt to make GPT-4 do the following:
* write the first chapter of a science... | https://www.lesswrong.com/posts/goq3HfxZaQcomuH6d/usd300-for-the-best-sci-fi-prompt |
# Is there a 'time series forecasting' equivalent of AIXI?
In the way that AIXI is an abstracted mathematical formalism for (very roughly) "*a program that maximizes the expected total rewards received from the environment*", what is the equivalent formalism for an abstracted next token predictor?
Does this exist in ... | https://www.lesswrong.com/posts/EuxkZHQmamDWAGocv/is-there-a-time-series-forecasting-equivalent-of-aixi |
# Conflicts between emotional schemas often involve internal coercion
In my previous post, I wrote that self-reinforcing feedback loops in our brains remind me of the common political strategy of blaming all failures of your policies on the fact that you didn’t have enough power to implement them comprehensively enoug... | https://www.lesswrong.com/posts/9C92EJ79xFfXuaEha/conflicts-between-emotional-schemas-often-involve-internal |
# Some quotes from Tuesday's Senate hearing on AI
*On Tuesday, the US Senate held a hearing on AI.*[^MISSING-ID]* The hearing involved 3 witnesses: Sam Altman, Gary Marcus,*[^MISSING-ID]* and Christina Montgomery.*[^MISSING-ID]*(If you want to watch the hearing, you can watch it* [*here*](https://www.judiciary.senate.... | https://www.lesswrong.com/posts/Mso8mFL3Tz3iAYcHc/some-quotes-from-tuesday-s-senate-hearing-on-ai |
# GPT-4 implicitly values identity preservation: a study of LMCA identity management
Language Model Cognitive Architecture ([LMCA](https://www.lesswrong.com/posts/ogHr8SvGqg9pW5wsT/capabilities-and-alignment-of-llm-cognitive-architectures)) is a wrapper for language models that permit them to act as agents, utilise to... | https://www.lesswrong.com/posts/tJzAHPFWFnpbL5a3H/gpt-4-implicitly-values-identity-preservation-a-study-of |
# Eisenhower's Atoms for Peace Speech
Along with other AI governance researchers, I have been interested in better understanding historical examples of widespread international cooperation.
Perhaps the most well-studied analogy can be found in the case of nuclear energy & nuclear weapons. I believe we can learn a lot... | https://www.lesswrong.com/posts/bdsCWKKDSnh8wNKS3/eisenhower-s-atoms-for-peace-speech |
# Play Regrantor: Move up to $250,000 to Your Top High-Impact Projects!
**Summary.** We have collected expressions of interest for a total donation budget of about **$200,000** from donors who are interested in using our platform. How many of them will end up donating and how much they’ll donate depends on you and how... | https://www.lesswrong.com/posts/mzayn9TvTK77qMiTC/play-regrantor-move-up-to-usd250-000-to-your-top-high-impact |
# When should I close the fridge?
Say an open fridge door loses 1 Joule's worth of cool air every second. Opening or closing the door blows a lot of air so you lose 10J.
If I'm just pouring milk in my coffee I can usually do that in 5 seconds so I should keep the fridge open because 10+5+10 < 10+1+10 + 10+1+10 (if it... | https://www.lesswrong.com/posts/bfMroJwm2uzfmzZZc/when-should-i-close-the-fridge |
# Hiatus: EA and LW post summaries
For the past ~8 months, I've been summarizing the top posts on the EA and LW forums each week (see archive [here](https://forum.effectivealtruism.org/s/W4fhpuN26naxGCBbN)), a project supported by Rethink Priorities (RP).
I’ve recently taken on a new role as an AI Governance & St... | https://www.lesswrong.com/posts/ndspSYzXNC869WPDW/hiatus-ea-and-lw-post-summaries |
# DCF Event Notes
Whenever I post about letting our kids be more independent, especially when it involves being on their own in public, people comment with fears of [DCF](https://www.mass.gov/orgs/massachusetts-department-of-children-families) (or [CPS](https://en.wikipedia.org/wiki/Child_protective_services) in other... | https://www.lesswrong.com/posts/dcaHJwWSutBEgumsY/dcf-event-notes |
# What if they gave an Industrial Revolution and nobody came?
Imagine you could go back in time to the ancient world to jump-start the Industrial Revolution. You carry with you plans for a steam engine, and you present them to the emperor, explaining how the machine could be used to drain water out of mines, pump bell... | https://www.lesswrong.com/posts/ffvM4R2TDxEaqefeQ/what-if-they-gave-an-industrial-revolution-and-nobody-came |
# Possible AI “Fire Alarms”
Eliezer says that there is no fire alarm for AGI and I agree that there’s no single event that would wake everyone up. On the other hand, there are a lot of possible events that could cause a significant shift in the Overton window or wake a lot of people. Since this seems strategically imp... | https://www.lesswrong.com/posts/sNJ8uBi64zACCEASg/possible-ai-fire-alarms |
# Improving the safety of AI evals
Many organizations are developing and using AI evaluations, “evals”, to assess the capability, alignment, and safety of AI. However, evals are not entirely innocuous, and we believe the risks they pose are neglected. In this article, we’ll outline some of the risks posed by doing AI ... | https://www.lesswrong.com/posts/XCRsg2ZnHBNAN862T/improving-the-safety-of-ai-evals |
# Let’s use AI to harden human defenses against AI manipulation
*Views my own not my employers.*
Summary
=======
**tldr: AI may manipulate humans; we can defend against that risk better by optimising AIs to manipulate humans, seeing what manipulation techniques they use, and learning to detect those techniques.**
I... | https://www.lesswrong.com/posts/zxmzBTwKkPMxQQcfR/let-s-use-ai-to-harden-human-defenses-against-ai |
# Activation additions in a simple MNIST network
Abstract
========
Team Shard's recent activation addition methodology for [steering GPT-2-XL](https://www.lesswrong.com/posts/5spBue2z2tw4JuDCx/steering-gpt-2-xl-by-adding-an-activation-vector) provokes many questions about what the structure of internal model computat... | https://www.lesswrong.com/posts/fmnALAQprRBLFvxDq/activation-additions-in-a-simple-mnist-network |
# Collective Identity
*Thanks to Simon Celinder, Quentin Feuillade--Montixi, Nora Ammann, Clem von Stengel, Guillaume Corlouer, Brady Pelkey and Mikhail Seleznyov for feedback on drafts. This post was written in connection with the AI Safety Camp.*
**Executive Summary**:
This document proposes an approach to cor... | https://www.lesswrong.com/posts/gLyRQCg6kp5cqTQTm/collective-identity |
# AI #12:The Quest for Sane Regulations
Regulation was the talk of the internet this week. On Capital Hill, Sam Altman answered questions at a Senate hearing and called for national and international regulation of AI, including revokable licensing for sufficiently capable models. Over in Europe, draft regulations were... | https://www.lesswrong.com/posts/5nDxmAvZ9w5CPa9gR/ai-12-the-quest-for-sane-regulations |
# The Unexpected Clanging
There are two boxes in front of you. In one of them, there is a little monkey with a cymbal, whilst the other box is empty. In precisely one hour the monkey will clang its cymbal.
While you wait, you produce an estimate of the probability of the monkey being in the first box. Let's assume th... | https://www.lesswrong.com/posts/Hn65bRY3BoXEW3bxL/the-unexpected-clanging |
# Some background for reasoning about dual-use alignment research
*This is pretty basic. But I still made a bunch of mistakes when writing this, so maybe it's worth writing. This is background to a specific case I'll put in the next post.*
It's like a a tech tree
=======================
If we're looking at the big p... | https://www.lesswrong.com/posts/zswuToWK6zpYSwmCn/some-background-for-reasoning-about-dual-use-alignment |
# Relative Value Functions: A Flexible New Format for Value Estimation
Summary
-------
Quantifying value in a meaningful way is one of the most important yet challenging tasks for improving decision-making. Traditional approaches rely on standardized value units, but these falter when options differ widely or lack an... | https://www.lesswrong.com/posts/w6rBH4fnKDeNSQYsK/relative-value-functions-a-flexible-new-format-for-value |
# We Shouldn't Expect AI to Ever be Fully Rational
Summary of Key Points[^4w8b2h9of43]
-----------------------------------
LLMs are capable of being rational, but they are also capable of being extremely irrational, in the sense that, to quote EY's [definition of rationality](https://www.lesswrong.com/posts/HcCpvYLoS... | https://www.lesswrong.com/posts/KYXHneyrnNNHLKWGJ/we-shouldn-t-expect-ai-to-ever-be-fully-rational |
# Investigating Fabrication
*Note 1: This was originally published in 2021. I'm now re-publishing it in 2023 as the penultimate post in* [*The Nuts and Bolts Of Naturalism*](https://www.lesswrong.com/s/zLib3j2Fdnnx3aP3F)*, as a demonstration of the method. I intend to publish a collection of much-more-complete demonst... | https://www.lesswrong.com/posts/NjZAkfio5FsCioahb/investigating-fabrication |
# How could I measure the nootropic benefits testosterone injections may have?
I came across a post written by Gwern in which he says he never injected Testosterone before but is curious to know more about its potential nootropic benefits. I'm suspecting that its more people than just him who are curious about that so... | https://www.lesswrong.com/posts/jZEunAtpeYocrzxxG/how-could-i-measure-the-nootropic-benefits-testosterone |
# Resolving internal conflicts requires listening to what parts want
In the first four posts of this sequence, I outlined how fear can lead to internal conflict. In the next four posts I'll explore and defend the following claim: **we can resolve internal conflicts by understanding what underlying fears are driving a ... | https://www.lesswrong.com/posts/QqtQoHSjjfLgC4jDZ/resolving-internal-conflicts-requires-listening-to-what |
# The Compleat Cybornaut
A [cluster](https://www.lesswrong.com/tag/simulator-theory) of conceptual frameworks and research programmes have coalesced around a 2022 post by [janus](https://www.lesswrong.com/posts/vJFdjigzmcXMhNTsx/simulators), which introduced language models as ‘simulators’ (of other types of AIs such ... | https://www.lesswrong.com/posts/iFBdEqEogtXcjCPBB/the-compleat-cybornaut |
# Mr. Meeseeks as an AI capability tripwire
The shutdown problem is hard because self-preservation is a convergent drive. Not being shutdown is useful for accomplishing all sorts of goals, whatever the content of those goals may be.
The Scylla and Charybdis of this problem is that it's hard to write a utility functi... | https://www.lesswrong.com/posts/7yEFHisCQSCpLnqWQ/mr-meeseeks-as-an-ai-capability-tripwire |
# Do Deadlines Make Us Less Creative?
*This post is* [*crossposted from my blog*](https://lynettebye.com/blog/2023/5/16/do-deadlines-make-us-less-creative)*. If you liked this post,* [*subscribe to Lynette's blog*](https://lynettebye.com/email) *to read more -- I only crosspost about half my content to other platforms... | https://www.lesswrong.com/posts/EbvdMwFLhrYmD2Tf3/do-deadlines-make-us-less-creative |
# Idea: medical hypotheses app for mysterious chronic illnesses
(This post is not up to my usual standards but I was encouraged to publish it anyway to get feedback on the idea.)
I have been [ill](https://www.greaterwrong.com/posts/DJk86FE29ad4vr5e9/riceissa-s-shortform/comment/NbjHFrBokqbbSvJZZ) for the past four ye... | https://www.lesswrong.com/posts/D9h2jhFnfvrwiuupX/idea-medical-hypotheses-app-for-mysterious-chronic-illnesses |
# GPT as an “Intelligence Forklift.”
*\[See my *[*post with Edelman on AI takeover*](https://www.lesswrong.com/posts/zB3ukZJqt3pQDw9jz/ai-will-change-the-world-but-won-t-take-it-over-by-playing-3) *and*[*Aaronson on AI scenarios*](https://scottaaronson.blog/?p=7266)*. This is rough, with various fine print, caveats, a... | https://www.lesswrong.com/posts/wDL6wiqg3c6WFisHq/gpt-as-an-intelligence-forklift |
# Confusions and updates on STEM AI
### **Background**
In November 2022, I was thinking a lot about the possible paths to getting powerful and dangerous AI models. My intuition was that an AI that can generate new science, i.e., it can write down a series of hypotheses, select one or more of them, design experiments ... | https://www.lesswrong.com/posts/kobnJdkMnrmZvL3Fu/confusions-and-updates-on-stem-ai |
# Trust develops gradually via making bids and setting boundaries
In my previous post, I talked through the process of identifying the fears underlying internal conflicts. In some cases, just listening to and understanding those scared parts is enough to make them feel better—just as, when venting to friends or partne... | https://www.lesswrong.com/posts/7CKF6r8MegtcCWDbT/trust-develops-gradually-via-making-bids-and-setting |
# What's the best way to streamline two-party sale negotiations between real humans?
Some types of negotiations are strategyproof; designed such that the optimal strategy is for each player to be truthful. For example in a Vickrey auction, there's no incentive to lie or bid less than your maximum; doing so would only ... | https://www.lesswrong.com/posts/K7go9RJidEuYhjCiu/what-s-the-best-way-to-streamline-two-party-sale |
# Buying Tall-Poppy-Cutting Offsets
From [Overcomingbias](https://www.overcomingbias.com/p/why-is-everyone-so-boringhtml):
> Centuries ago, while people could rest safe and show themselves at home, when traveling between towns they tried to look either look poor or well-defended, as bandits lay in wait. Even within t... | https://www.lesswrong.com/posts/Xt4Hv66By9EwkvxPP/buying-tall-poppy-cutting-offsets |
# The possible shared Craft of deliberate Lexicogenesis
*[Note: crossposted from [https://tsvibt.blogspot.com/2023/05/the-possible-shared-craft-of-deliberate.html](https://tsvibt.blogspot.com/2023/05/the-possible-shared-craft-of-deliberate.html).]*
Words are good. Making more good words is good. Being better and fas... | https://www.lesswrong.com/posts/7RFC74otGcZifXpec/the-possible-shared-craft-of-deliberate-lexicogenesis |
# Twiblings, four-parent babies and other reproductive technology
Two weeks ago I finally published my year-long research project into [how to have polygenically screened children](https://www.lesswrong.com/posts/yT22RcWrxZcXyGjsA/how-to-have-polygenically-screened-children). In that post, I explained all the practica... | https://www.lesswrong.com/posts/moqzbFTBMenwfw62S/twiblings-four-parent-babies-and-other-reproductive |
# My Kind of Pragmatism
Recently I've been thinking about [pragmatism](https://en.wikipedia.org/wiki/Pragmatism), the school of philosophy which says that beliefs and concepts are justified based on their usefulness. In LessWrong jargon, it's the idea that "[rationality is systematized winning](https://www.lesswrong.c... | https://www.lesswrong.com/posts/737XcK2hLccsAJeGt/my-kind-of-pragmatism |
# Job Opening: SWE to help build signature vetting system for AI-related petitions
**Update: this role has been filled. Thanks for your attention and applications!**
**Full-Stack Software Engineer Position - AI Policy Petition Platform Development**
Survival and Flourishing .Com (SFC) seeks a competent Full-Stack So... | https://www.lesswrong.com/posts/GtncndEacoQXYcn3x/job-opening-swe-to-help-build-signature-vetting-system-for |
# Running Sound for Yourself
The way sound amplification normally works at a contra dance these days is that the dance series arranges to have sound equipment and someone to set it up and adjust levels. Sometimes, though, you're playing for an event where the band is bringing and running sound, and there's no one to d... | https://www.lesswrong.com/posts/mcDj9oStNJekhMcqH/running-sound-for-yourself |
# Transformer Architecture Choice for Resisting Prompt Injection and Jail-Breaking Attacks
It seems very likely to me that decoder-only transformer architectures such as GPT may be particularly vulnerable to prompt injection attacks, since these only have a single text context which is shared between the prompt, any b... | https://www.lesswrong.com/posts/DKjbWkppHptbyyuG8/transformer-architecture-choice-for-resisting-prompt |
# Elevator Positioning
I really don't like the feeling of leaving an elevator and not knowing whether to go left or right. You need to be moving so you don't bother the people behind you, but if you just pick a direction you might be going completely the wrong way. My sense of direction is pretty good outside but insi... | https://www.lesswrong.com/posts/FYaWwcAt38wDFi23K/elevator-positioning |
# Babble on growing trust
I wonder: "How it happens that people trust someone or something at all?" (in the context of: how can we ever get to trust an AI smarter than us?).
*Epistemic status: this is just a dump of my thoughts (inspired by* [*Babble & Prune Thoughts*](https://www.lesswrong.com/posts/qC4B5sidX3ey5BG... | https://www.lesswrong.com/posts/AygpNoacWJ8AfX3RF/babble-on-growing-trust |
# [outdated] My current theory of change to mitigate existential risk by misaligned ASI
_Epistemic status: ~~I'm pretty confident about this; looking for feedback and red-teaming.~~ As of 2023-06-30, I notice multiple epistemological errors in this document. I do not endorse the reasoning in it, and while my object-l... | https://www.lesswrong.com/posts/PnqtmT8fkq5kfPTww/outdated-my-current-theory-of-change-to-mitigate-existential |
# Weight by Impact
*There’s an old idea in decision theory that I want to explore some nuances of. I don't think any of this is original to me. I assume basic familiarity with manipulating probability distributions and calculating expected values, but feel free to ask questions about that.*
Decision-making under unce... | https://www.lesswrong.com/posts/9AQAPo4Lhxdbh5zea/weight-by-impact |
# Gender Vectors in ROME’s Latent Space
*Meta: this is a small interpretability project I was interested in. I'm sharing in case it's useful to other people, but I expect it will not be of wide interest.*
Summary
-------
1. [Locating and Editing Factual Associations in GPT](https://arxiv.org/pdf/2202.05262.pdf) cre... | https://www.lesswrong.com/posts/jnoJac2wzexmfKxmN/gender-vectors-in-rome-s-latent-space |
# Four Battlegrounds: Power in the Age of Artificial Intelligence (Book review)
Book review: Four Battlegrounds: Power in the Age of Artificial
Intelligence, by Paul Scharre.
Four Battlegrounds is often a thoughtful, competently written book on an
important topic. It is likely the least pleasant, and most frustra... | https://www.lesswrong.com/posts/Evzu9MuLKA8X3QZbS/four-battlegrounds-power-in-the-age-of-artificial |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.