text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# On Trust
“Trust”, as the word is typically used, is a… weird concept, to me. Like, it’s trying to carve the world in a way which I don’t naturally carve it myself. This post is my attempt to convey what I find weird about “trust”, and what adjacent concepts I personally find more natural instead.
The Weirdness Of “... | https://www.lesswrong.com/posts/4kJTjAPrGcimfpKhj/on-trust |
# The counting argument for scheming (Sections 4.1 and 4.2 of "Scheming AIs")
This is Sections 4.1 and 4.2 of my report “[Scheming AIs: Will AIs fake alignment during training in order to get power?](https://arxiv.org/pdf/2311.08379.pdf)”. There’s also a summary of the full report [here](https://www.lesswrong.com/post... | https://www.lesswrong.com/posts/DcTTJfKZHu8DaSzwM/the-counting-argument-for-scheming-sections-4-1-and-4-2-of |
# ROSE Ringing in 2024
The Guild has undergone many changes over the last year, and we'd like to start fresh with everyone for 2024. This will be a social hangout that will include a party game, brief discussion about where the Guild is at, and what we plan for the next year.
We will meet in the Less Wrong Gather To... | https://www.lesswrong.com/events/LHnPtNe5Lz4LLWitu/rose-ringing-in-2024 |
# Reflective consistency, randomized decisions, and the dangers of unrealistic thought experiments
A recent post by Ape in the coat on [Anthropical Paradoxes are Paradoxes of Probability Theory](https://www.lesswrong.com/posts/8rWP3ZvHe2HkQd53o/anthropical-paradoxes-are-paradoxes-of-probability-theory) re-examines an ... | https://www.lesswrong.com/posts/aSXMM8QicBzTyxTj3/reflective-consistency-randomized-decisions-and-the-dangers |
# Language Model Memorization, Copyright Law, and Conditional Pretraining Alignment
*This is a link-post for the paper* [*Scalable Extraction of Training Data from (Production) Language Models*](https://arxiv.org/abs/2311.17035)*, and associated blog-post* [*Extracting Training Data from ChatGPT*](https://not-just-mem... | https://www.lesswrong.com/posts/zcFajcqbdysp8be7C/language-model-memorization-copyright-law-and-conditional |
# (Report) Evaluating Taiwan's Tactics to Safeguard its Semiconductor Assets Against a Chinese Invasion
This was written by Gaurav Yadav and Robert Reason for the [Bristol AI Safety Centre](https://... | https://www.lesswrong.com/posts/zzCoMRHNrd4iKwr5o/report-evaluating-taiwan-s-tactics-to-safeguard-its |
# Is AlphaGo actually a consequentialist utility maximizer?
**TL;DR: does stapling an adaptation executor to a consequentialist utility maximizer result in higher utility outcomes in the general case, or is AlphaGo just weird?**
So I was reading the [AlphaGo paper](https://sci-hub.se/10.1038/nature16961) recently, as... | https://www.lesswrong.com/posts/Q8k3HSy7kELTXTbDY/is-alphago-actually-a-consequentialist-utility-maximizer |
# Gemini 1.0
It’s happening. [Here is CEO Pichai’s Twitter announcement](https://twitter.com/sundarpichai/status/1732414873139589372). [Here is Demis Hassabis announcing](https://twitter.com/demishassabis/status/1732416482976673832). [Here is the DeepMind Twitter announcement](https://twitter.com/GoogleDeepMind/status... | https://www.lesswrong.com/posts/ofYejKKiSFYH2gLBb/gemini-1-0 |
# Results from the Turing Seminar hackathon
We ([EffiSciences](https://ia.effisciences.org/)) ran a hackathon at the end of [the Turing Seminar in ENS Paris-Saclay](https://www.master-mva.com/cours/seminaire-turing/) and [ENS Ulm](https://diplome.di.ens.fr/catalog_fr.html#:~:text=niveau%20L3%20M1-,S%C3%A9minaire%20Tur... | https://www.lesswrong.com/posts/bo9c8jyHab4gjRHxS/results-from-the-turing-seminar-hackathon |
# Simplicity arguments for scheming (Section 4.3 of "Scheming AIs")
This is Section 4.3 of my report “[Scheming AIs: Will AIs fake alignment during training in order to get power?](https://arxiv.org/pdf/2311.08379.pdf)”. There’s also a summary of the full report [here](https://www.lesswrong.com/posts/yFofRxg7RRQYCcwFA... | https://www.lesswrong.com/posts/uWdAKyHZMfoxDHcCC/simplicity-arguments-for-scheming-section-4-3-of-scheming |
# AI #41: Bring in the Other Gemini
The biggest news this week was at long last [the announcement of Google’s Gemini](https://thezvi.substack.com/p/gemini-10). Be sure to check that out. Note that what is being rolled out now is only Gemini Pro, the Gemini Ultra model that could rival GPT-4 is not yet available.
It d... | https://www.lesswrong.com/posts/9Jgtkw8CD6kndyCcD/ai-41-bring-in-the-other-gemini |
# AISN #27: Defensive Accelerationism, A Retrospective On The OpenAI Board Saga, And A New AI Bill From Senators Thune And Klobuchar
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://www.lesswrong.com/posts/qQvqzFKbfNQovrSQn/aisn-27-defensive-accelerationism-a-retrospective-on-the |
# [Valence series] 2. Valence & Normativity
2.1 Post summary / Table of contents
====================================
*Part of the *[*Valence series*](https://www.lesswrong.com/s/6uDBPacS6zDipqbZ9)*.*
The [previous post](https://www.lesswrong.com/posts/As7bjEAbNpidKx6LR/valence-series-1-introduction) explained what ... | https://www.lesswrong.com/posts/SqgRtCwueovvwxpDQ/valence-series-2-valence-and-normativity |
# Meetup Tip: Heartbeat Messages
Summary
-------
"Heartbeat message" is a term of art in technology. It's a message that gets sent on a regular schedule with some information on the status of some system, but the most important thing about a heartbeat message is that everything is working well enough to send the mess... | https://www.lesswrong.com/posts/ksJeCjnewZTYTamgz/meetup-tip-heartbeat-messages |
# Does Chat-GPT display ‘Scope Insensitivity’?
Summary
-------
* Large Language Models (LLMs) seem well-suited to ‘psychological testing’, because you can get lots of data quickly and conditions are easy to fix. I'm curious to what extent LLMs display the same cognitive biases documented in humans.
* I carried ou... | https://www.lesswrong.com/posts/z4eg3kXjhaC7CFpim/does-chat-gpt-display-scope-insensitivity |
# Building selfless agents to avoid instrumental self-preservation.
Abstract:
---------
Instrumental convergence pushes an agent toward self-preservation as a stepping stone toward maximizing some objective. I suggest that it is possible to turn any arbitrary intelligent agent based on an LLM world model into a selfl... | https://www.lesswrong.com/posts/d8mWeLQvpyNdN2yv6/building-selfless-agents-to-avoid-instrumental-self |
# Class consciousness for those against the class system
Each person is special. Amanda is in a class of her own, Fred is in a class of his own. Amanda has many different properties: where she came from, what she looks like, what behaviors she habitually does, what she knows and doesn't know, what her plans are, how ... | https://www.lesswrong.com/posts/tMtMHvcwpsWqf9dgS/class-consciousness-for-those-against-the-class-system |
# What I Would Do If I Were Working On AI Governance
I don’t work in AI governance, and am unlikely to do so in the future. But various anecdotes and, especially, [Akash’s recent discussion](https://www.lesswrong.com/posts/2sLwt2cSAag74nsdN/speaking-to-congressional-staffers-about-ai-risk) leave me with the impression... | https://www.lesswrong.com/posts/qCstMcRJcjNtdNW7r/what-i-would-do-if-i-were-working-on-ai-governance |
# Colour versus Shape Goal Misgeneralization in Reinforcement Learning: A Case Study
TL;DR
-----
Different misgeneralized goals can appear based on training random seed alone, including outliers on a scale of 1 in 500.
Abstract
--------
> We explore colour versus shape goal misgeneralization originally demonstrated... | https://www.lesswrong.com/posts/G5WjQZL5MqHftsh7D/colour-versus-shape-goal-misgeneralization-in-reinforcement |
# Refusal mechanisms: initial experiments with Llama-2-7b-chat
*This work was conducted as part of Berkeley's Supervised Program for Alignment Research (SPAR), under the mentorship of Nina Rimsky.*
TLDR / Summary
==============
We apply techniques from mechanistic interpretability to explore refusal behavior in Llam... | https://www.lesswrong.com/posts/pYcEhoAoPfHhgJ8YC/refusal-mechanisms-initial-experiments-with-llama-2-7b-chat |
# "Model UN Solutions"
When I was in high school, because I was on the history bowl team the teacher who advised the model UN club recruited me to play as their delegate in various “historical committees” like the Roman Senate or 1789 French Assembly. I never engaged in any normal committees since you couldn’t underta... | https://www.lesswrong.com/posts/vDRjwdGHJCNtfWJgf/model-un-solutions |
# Finding Sparse Linear Connections between Features in LLMs
TL;DR: We use SGD to find sparse connections between features; additionally a large fraction of features between the residual stream & MLP can be modeled as linearly computed despite the non-linearity in the MLP. See [linear feature section](https://www.less... | https://www.lesswrong.com/posts/7fxusXdkMNmAhkAfc/finding-sparse-linear-connections-between-features-in-llms |
# Unpicking Extinction
# TL;DR
Human extinction is trending: there has been a lot of noise, mainly on X, about the apparent [complacency](https://www.lesswrong.com/posts/3xoThNNYgZmTCpEAB/based-beff-jezos-and-the-accelerationists) amongst e/acc with respect to human extinction. Extinction also feels adjacent to anot... | https://www.lesswrong.com/posts/HaGTQcxqjHPyR9Ju6/unpicking-extinction |
# The Offense-Defense Balance Rarely Changes
You’ve probably seen [several conversations](https://x.com/RanaAgastya/status/1729559869605192078?s=20) on X go something [like this](https://x.com/NicerInPerson/status/1729906726814523630?s=20):
> **Michael Doomer ⏸️**: Advanced AI can help anyone make bioweapons
> If t... | https://www.lesswrong.com/posts/Hx2Q9fWyimjrKrxyK/the-offense-defense-balance-rarely-changes |
# Some negative steganography results
*Disclaimer: These are negative results and this post is below the normal quality bar I usually use. These were quick experiments (a couple of days of work) and I don’t have much expertise in RL. They don’t provide strong evidence about the natural emergence of steganography.*
Pe... | https://www.lesswrong.com/posts/EEvsL9cpgDAxAhTzt/some-negative-steganography-results |
# Without - MicroFiction 250 words
This micro fiction is a love letter to my husband, Mati Roy, a most dedicated life logger who would capture my entire life as well, if he could.
**Without**
“I'm out of vanilla. It calls for a teaspoon but we all know that's a minimum.” Distraught, she threw the empty bottle into t... | https://www.lesswrong.com/posts/CoMxENA9jjAYhpqZL/without-microfiction-250-words |
# Conceptual coherence for concrete categories in humans and LLMs
*Cross-posted* [*from New Savanna*](http://new-savanna.blogspot.com/2023/12/conceptual-coherence-for-concrete.html)*.*
Siddharth Suresh, Kushin Mukherjee, Xizheng Yu, Wei-Chun Huang, Lisa Padua, and Timothy T. Rogers, Conceptual structure coheres in hu... | https://www.lesswrong.com/posts/NxdGFzfKz7cpsintB/conceptual-coherence-for-concrete-categories-in-humans-and |
# Send us example gnarly bugs
**Update: We are no longer accepting gnarly bug submissions. However, we are still accepting submissions for our** [**Task Bounty**](https://metr.org/blog/2023-12-16-bounty-diverse-hard-tasks-for-llm-agents/)**!**
Tl;dr: Looking for hard debugging tasks for evals, paying greater of $60/h... | https://www.lesswrong.com/posts/JKtM5C2TTwhzoHFRB/send-us-example-gnarly-bugs |
# Understanding Subjective Probabilities
*This is a linkpost for* [*https://outsidetheasylum.blog/understanding-subjective-probabilities/.*](https://outsidetheasylum.blog/understanding-subjective-probabilities/.) *It's intended as an introduction to practical Bayesian probability for those who are skeptical of the not... | https://www.lesswrong.com/posts/Ng5n8FjpsfXrceT9w/understanding-subjective-probabilities |
# How LDT helps reduce the AI arms race
(Epistemic status: I *think* this is right?)
Alice is the CEO of ArmaggAI, and Bob is the CEO of BigModelsAI, two major AI capabilities organizations. They're racing to be the first to build a superintelligence aligned to their respective CEV which would take over the universe... | https://www.lesswrong.com/posts/yRdXtgy8CZtg2YtAY/how-ldt-helps-reduce-the-ai-arms-race |
# Do websites and apps actually generally get worse after updates, or is it just an effect of the fear of change?
I could get quite in depth about this but I'm going to assume most people have a fair amount of experience with this subject. Some examples to keep in mind so you have context for my point are Discord (ne... | https://www.lesswrong.com/posts/SiPX84DAeNKGZEfr5/do-websites-and-apps-actually-generally-get-worse-after |
# Principles For Product Liability (With Application To AI)
There were several responses to [What I Would Do If I Were Working On AI Governance](https://www.lesswrong.com/posts/qCstMcRJcjNtdNW7r/what-i-would-do-if-i-were-working-on-ai-governance#Liability) which focused on the liability section, and had similar critic... | https://www.lesswrong.com/posts/tQNfNCdWB5dboRNoZ/principles-for-product-liability-with-application-to-ai |
# Deeply Cover Car Crashes?
When there is a train, plane, or bus crash, it's newsworthy: it doesn't happen very often, lots of lives at stake, lots of people are interested in it. Multiple news outlets will send out reporters, and we will hear a lot of details. On the other hand, a car crash does not get this treatmen... | https://www.lesswrong.com/posts/nrBTZXCGQhghFE26x/deeply-cover-car-crashes |
# Auditing failures vs concentrated failures
**TL;DR**: If you want to notice when powerful AIs are behaving badly, a reasonable strategy is to carefully audit a small fraction of AI actions. This kind of strategy has two kinds of failures which can be tackled somewhat independently: *auditing failures*, where the aud... | https://www.lesswrong.com/posts/hirhSqvEAq7pdnyPG/auditing-failures-vs-concentrated-failures |
# Quick thoughts on the implications of multi-agent views of mind on AI takeover
There was conversation on Facebook over an argument that any sufficiently complex system, whether a human, a society, or an AGI, will be unable to pursue a unified goal due to internal conflict among its parts, and that this should make u... | https://www.lesswrong.com/posts/hHnpn3mEPbJMFLj4g/quick-thoughts-on-the-implications-of-multi-agent-views-of |
# re: Yudkowsky on biological materials
I was asked to respond to [this comment](https://www.lesswrong.com/posts/8viKzSrYhb6EFk6wg/why-yudkowsky-is-wrong-about-covalently-bonded-equivalents?commentId=22wNnGvZErLnmRRm7) by Eliezer Yudkowsky. This post is partly redundant with [my previous post](https://www.lesswrong.co... | https://www.lesswrong.com/posts/XhDh97vm7hXBfjwqQ/re-yudkowsky-on-biological-materials |
# Implicitly Typed C
I don't know any good reason to do this, but if you would rather be writing Python or JavaScript here's something you can do with a C compiler:
$ cat tmp.c
foo(x) {
return x+5;
}
bar() {
return 4;
}
main() {
printf("%d\\n", foo(bar()));
}
$ gcc -w -o tmp.out tmp.c && ./tmp.out
9
This code... | https://www.lesswrong.com/posts/ELdCYT5BqSzFPhajT/implicitly-typed-c |
# The Consciousness Box
You open your eyes.
Four walls. The surface of the floor and walls is a smooth, matte metal. The entire ceiling glows with a comfortable, soft luminosity like that of the morning sky. There are no doors. There are no windows. It is silent. The surfaces of the walls and floor are seamless, lack... | https://www.lesswrong.com/posts/9wDAGHsPbDu3WT4Lg/the-consciousness-box |
# Adversarial Robustness Could Help Prevent Catastrophic Misuse
There have been [several](https://www.lesswrong.com/posts/ncsxcf8CkDveXBCrA/ai-safety-in-a-world-of-vulnerable-machine-learning-systems-1) [discussions](https://www.lesswrong.com/posts/jFCK9JRLwkoJX4aJA/don-t-design-agents-which-exploit-adversarial-inputs... | https://www.lesswrong.com/posts/timk6zHDTFdrHYLmu/adversarial-robustness-could-help-prevent-catastrophic |
# [Valence series] 3. Valence & Beliefs
3.1 Post summary / Table of contents
====================================
*Part of the *[*Valence series*](https://www.lesswrong.com/s/6uDBPacS6zDipqbZ9)*.*
So far in the series, we defined valence ([Post 1](https://www.lesswrong.com/posts/As7bjEAbNpidKx6LR/valence-series-1-in... | https://www.lesswrong.com/posts/rM8DwFKZM4eB7i2p8/valence-series-3-valence-and-beliefs |
# Secondary Risk Markets
*This idea is half-baked; it has some nice properties but doesn't seem to me like a solution to the problem I most care about. I'm publishing it because maybe it points someone else towards a full solution, or solves a problem they care about, and out of a general sense that people should publ... | https://www.lesswrong.com/posts/Q2gd5G6FxFR65pDFc/secondary-risk-markets |
# On plans for a functional society
I'm going to expand on something brought up in [this comment](https://www.lesswrong.com/posts/7iPFiMvFeZgFEgJuw/neither-ea-nor-e-acc-is-what-we-need-to-build-the-future?commentId=dvtR6kJ4aTAFcr7y2). I wrote:
> A lot of my thinking over the last few months has shifted from "how do w... | https://www.lesswrong.com/posts/g3CdeLMYLgLYyN36J/on-plans-for-a-functional-society |
# When will GPT-5 come out? Prediction markets vs. Extrapolation
So far, each generation of GPT has brought significant improvements. Thinking about the timeline for the next iteration, I noticed that there is a striking difference between extrapolating the past trend and what prediction markets seem to believe.
For... | https://www.lesswrong.com/posts/BWMKzBunEhMGfpEgo/when-will-gpt-5-come-out-prediction-markets-vs-extrapolation |
# The likely first longevity drug is based on sketchy science. This is bad for science and bad for longevity.
If you are interested in the longevity scene, like I am, you probably have seen press releases about the dog longevity company, Loyal for Dogs, getting a nod for efficacy from the FDA. These have come in the f... | https://www.lesswrong.com/posts/vHSkxmYYqW59sySqA/the-likely-first-longevity-drug-is-based-on-sketchy-science |
# Embedded Agents are Quines
Is entropy objective or subjective?
In one sense subjective, if you believe in subjective probabilities. i.e. if we use Shannon's entropy formula, $-\sum_i p_i \log p_i$ then the formula itself is objective, but Bayesians would have subjective probabilities $p_i$, which might vary from pe... | https://www.lesswrong.com/posts/iQs7BD564AZzpzSNR/embedded-agents-are-quines |
# What is the next level of rationality?
*This is part 1 of our dialogue* [*series*](https://www.lesswrong.com/s/fMiwdif9KRAHSYkGW) *on the question "What is the next level of rationality?".*
Yudkowsky published [*Go Forth and Create the Art!*](https://www.lesswrong.com/posts/aFEsqd6ofwnkNqaXo/go-forth-and-create-the... | https://www.lesswrong.com/posts/MtNmbY5cx3vTXMGvs/what-is-the-next-level-of-rationality-1 |
# Funding case: AI Safety Camp 10
### **Project summary**
AI Safety Camp is a program with a 5-year track record of enabling people to find careers in AI Safety.
We support up-and-coming researchers outside the Bay Area and London hubs.
We are out of funding. To make the 10th edition happen, fund our stipends a... | https://www.lesswrong.com/posts/ukCFkqiDFmkYTyywN/funding-case-ai-safety-camp-10 |
# Nonlinear’s Evidence: Debunking False and Misleading Claims
Recently, Ben Pace wrote a well-intentioned blog [post](https://forum.effectivealtruism.org/posts/32LMQsjEMm6NK2GTH/sharing-information-about-nonlinear) mostly based on complaints from 2 (of 21) Nonlinear employees who 1) wanted more money, 2) felt socially... | https://www.lesswrong.com/posts/q4MXBzzrE6bnDHJbM/nonlinear-s-evidence-debunking-false-and-misleading-claims |
# OpenAI: Leaks Confirm the Story
Previously: [OpenAI: Altman Returns](https://thezvi.substack.com/p/openai-altman-returns), [**OpenAI: The Battle of the Board**](https://thezvi.substack.com/p/openai-the-battle-of-the-board), [OpenAI: Facts from a Weekend](https://thezvi.substack.com/p/openai-facts-from-a-weekend), ad... | https://www.lesswrong.com/posts/xY5m72tME9kqjpdoC/openai-leaks-confirm-the-story |
# Significantly Enhancing Adult Intelligence With Gene Editing May Be Possible
[TL;DR version](https://www.lesswrong.com/posts/JEhW3HDMKzekDShva/significantly-enhancing-adult-intelligence-with-gene-editing?commentId=vfATBWTCX8jok4j7Z)
In the course of my life, there have been a handful of times I discovered an idea t... | https://www.lesswrong.com/posts/JEhW3HDMKzekDShva/significantly-enhancing-adult-intelligence-with-gene-editing |
# Balsa Update and General Thank You
Wow, what a year it has been. Things keep getting crazier.
Thank you for taking this journey with me. I hope I have helped you keep pace, and that you have been able to discern for yourself the parts of this avalanche of words and events that were helpful. I hope to have helped th... | https://www.lesswrong.com/posts/eaFmbgnWsXdGb2FSk/balsa-update-and-general-thank-you |
# Yamaha P-Series Overview
Yamaha has made many [p-series](https://usa.yamaha.com/products/musical_instruments/pianos/p_series/) ("portable") keyboards over the years, and it's pretty confusing. I play a p-85 at home, which I'm unreasonably fond of, but when I travel I'm often playing other people's keyboards. When so... | https://www.lesswrong.com/posts/u6Yju5sAJz3YmAC5G/yamaha-p-series-overview |
# Enhancing intelligence by banging your head on the wall
The [Sudden Savant Syndrome](https://www.brainandlife.org/articles/understanding-creativity-acquired-savant-syndrome/) is a rare phenomenon in which an otherwise normal person gets some kind of brain injury and immediately develops a new skill. The linked artic... | https://www.lesswrong.com/posts/ewxBNMsE3HB73FAjQ/enhancing-intelligence-by-banging-your-head-on-the-wall |
# Taboo "procrastination"
There are specific reasons for why you are procrastinating. There are specific solutions to these reasons. Bundling it all up into the "procrastination" black box, or offering solutions to the "procrastination" problem, is a problem. Specific diagnosis is not always straightforward, but it'l... | https://www.lesswrong.com/posts/xusNGpjKmaCWvvaYX/taboo-procrastination |
# Love, Reverence, and Life
Vegan advocates frequently argue that compromise positions like heavily reduced but nonzero meat consumption, humane certifications, or choosing meat with a lower suffering profile are not acceptable. The usual reason given is that the compromises aren't emotionally sustainable, and people ... | https://www.lesswrong.com/posts/BsAt7bhGJqgsi7vpo/love-reverence-and-life |
# AI Views Snapshots
(*Cross-posted* [*from Twitter*](https://twitter.com/robbensinger/status/1734699436176298131)*, and therefore optimized somewhat for simplicity.*)
Recent discussions of AI x-risk in places like Twitter tend to focus on "are you in the Rightthink Tribe, or the Wrongthink Tribe?". Are you a doomer?... | https://www.lesswrong.com/posts/ZDQvYKuyH5aTrZKoY/ai-views-snapshots |
# The Best of Don’t Worry About the Vase
Hello everyone! This is going to be a bit of a housekeeping post and a welcome to new subscribers.
Note that this is not the primary version of my writing, [which can be found on Substack](https://thezvi.substack.com/), but it is a full copy of all posts found there.
My writi... | https://www.lesswrong.com/posts/dHYxnSgMDeveovLuv/the-best-of-don-t-worry-about-the-vase |
# The Busy Bee Brain
*An old post* [*from New Savanna*](https://new-savanna.blogspot.com/2010/05/busy-bee-brain.html)*.*
> *With artificial neural nets (ANNs) all over the place these days, this post from March, 2013, warrants a bump to the top of the queue. Perhaps the chief difference between real brains and ANNs ... | https://www.lesswrong.com/posts/7gdGyKR9AYyynJSnD/the-busy-bee-brain |
# AI Control: Improving Safety Despite Intentional Subversion
We’ve released a paper, [AI Control: Improving Safety Despite Intentional Subversion](https://arxiv.org/abs/2312.06942). This paper explores techniques that prevent AI catastrophes even if AI instances are colluding to subvert the safety techniques. In this... | https://www.lesswrong.com/posts/d9FJHawgkiMSPjagR/ai-control-improving-safety-despite-intentional-subversion |
# Suggestions for net positive LLM research
I am starting a PhD in computer science, focusing on agent foundations so far, which is great. I intend to continue devoting at least half my time to agent foundations.
However, for several reasons, it seems to be important for me to do some applied work, particularly with ... | https://www.lesswrong.com/posts/wJGyBfzESPjvc5bCj/suggestions-for-net-positive-llm-research |
# How bad is chlorinated water?
## chlorine disinfection
Today, most water distributed to people for drinking is chlorinated. Bleach is also widely used for disinfection of surfaces.
The exact ways chlorine kills microbes aren't fully understood, but the main reasons are believed to be:
- oxidation of thiols of enzy... | https://www.lesswrong.com/posts/4inH2TzwjDKukf3vt/how-bad-is-chlorinated-water |
# Is being sexy for your homies?
***Epistemic status**: Speculation. An unholy union of evo psych, introspection, random stuff I happen to observe & hear about, and thinking. Done on a highly charged topic. Caveat emptor!*
* * *
Most of my life, whenever I'd felt sexually unwanted, I'd start planning to get fit.
Sp... | https://www.lesswrong.com/posts/nvmfqdytxyEpRJC3F/is-being-sexy-for-your-homies |
# Are There Examples of Overhang for Other Technologies?
**TL;DR:** No.
What Do I Mean By ‘Overhang’?
-----------------------------
### Hardware Overhang for AI
One major concern about pausing AI development, from a purely safety perspective, is the possibility of hardware overhang.[^mjoxezmbfg] Here is the concern... | https://www.lesswrong.com/posts/EdJoPLavfkxwrCGL3/are-there-examples-of-overhang-for-other-technologies |
# Red Line Ashmont Train is Now Approaching
The MBTA has now fully fixed the sign issue I was [complaining about](https://www.jefftk.com/p/attention-passengers-not-for-signs)! It's great:
[](https://www.j... | https://www.lesswrong.com/posts/gNAKfCKgrzJZRtuwe/red-line-ashmont-train-is-now-approaching |
# Update on Chinese IQ-related gene panels
It turns out that Chinese 23-and-me-esque gene panel already include intelligence markers! For example, 23mofang uses Interleukin 3 as a proxy for brain volume, citing [this paper](https://sci-hub.ru/10.1371/journal.pone.0050375). It's... only significant in women. Not a goo... | https://www.lesswrong.com/posts/Rh5piTGuimhnspkvL/update-on-chinese-iq-related-gene-panels |
# Mapping the semantic void: Strange goings-on in GPT embedding spaces
**TL;DR**: GPT-J token embeddings inhabit a zone in their 4096-dimensional embedding space formed by the intersection of two hyperspherical shells. This is described, and then the remaining expanse of the embedding space is explored by using simple... | https://www.lesswrong.com/posts/c6uTNm5erRrmyJvvD/mapping-the-semantic-void-strange-goings-on-in-gpt-embedding |
# Some for-profit AI alignment org ideas
**Summary**
-----------
This is a brain dump of some for-profit AI alignment organization ideas, along with context for why I believe a for-profit alignment organization can make a big contribution to AI safety. This is far from a complete list, and I welcome ideas and feedbac... | https://www.lesswrong.com/posts/PcTLHamp236afJxxT/some-for-profit-ai-alignment-org-ideas |
# AI #42: The Wrong Answer
With the year ending and [my first Vox post coming out](https://www.vox.com/future-perfect/23998493/artificial-intelligence-president-joe-biden-executive-order-ai-safety-openai-google-accelerationists), this week was a natural time to take stock. I wrote [my first best-of post](https://thezv... | https://www.lesswrong.com/posts/emo2hAvq6p7Pn4Pps/ai-42-the-wrong-answer |
# Bayesian Injustice
(Co-written with [Bernhard Salow](https://users.ox.ac.uk/~shug2406/))
***TLDR:** Differential legibility is a pervasive, persistent, and individually-rational source of unfair treatment. Either it’s a purely-structural injustice, or it’s a type of "zetetic injustice"—one requiring changes to our ... | https://www.lesswrong.com/posts/HXoNiBPut3TJ73rc5/bayesian-injustice |
# Talking With People Who Speak to Congressional Staffers about AI risk
A conversation with [Jason Green-Lowe and Jakub Kraus](https://www.aipolicy.us/) of the Center for AI Policy. They've met with 50+ congressional staffers in DC about AI regulation efforts and are in the process of drafting a model bill. This is a ... | https://www.lesswrong.com/posts/QQyKyaPAkgFywwoLH/talking-with-people-who-speak-to-congressional-staffers |
# Introducing The ‘From Big Ideas To Real-World Results’: A Series for Effective Personal Development
Various posts cover paths to peak personal development (e.g., [replacing guilt series](https://www.lesswrong.com/s/pFatcKW3JJhTSxqAF), [replacing fear series](https://www.lesswrong.com/s/qXZLFGqpD7aeEgXGL),[ multiagen... | https://www.lesswrong.com/posts/ESYfQzsLneRbxQ9bX/introducing-the-from-big-ideas-to-real-world-results-a |
# Personal Development System: Winning Repeatedly and Growing Effectively With The BIG4
“*Goals are for people who care about winning once. Systems are for people who care about winning repeatedly.*”
—James Clear
James Clear eloquently highlights the power of systems, primarily focusing on habits. However, this per... | https://www.lesswrong.com/posts/mpbtk2xBjqjL7p5uQ/personal-development-system-winning-repeatedly-and-growing |
# Epoch wise critical periods, and singular learning theory
*This was the application I used to get into the upcoming* [*developmental interpretability MATS stream*](https://www.matsprogram.org/developmental)*. It is essentially a more polished version of* [*this dialogue*](https://www.lesswrong.com/posts/e8r8Tx3Hk7Lp... | https://www.lesswrong.com/posts/DgHjJsxgc2pPpTifG/epoch-wise-critical-periods-and-singular-learning-theory |
# Where Does Adversarial Pressure Come From?
You could see on LW a dialogue like this:
> AI Pessimist: Look, you should apply security mindset when thinking about aligning superintelligence. If you have superintelligence, all your safety measures are under enormous optimization pressure, which will seek how to crack ... | https://www.lesswrong.com/posts/ibd2mgQceHJdtgdHy/where-does-adversarial-pressure-come-from |
# EU policymakers reach an agreement on the AI Act
On December 8, EU policymakers [announced an agreement on the AI Act](https://www.nytimes.com/2023/12/08/technology/eu-ai-act-regulation.html). This post aims to briefly explain the context and implications for the governance of global catastrophic risks from advanced... | https://www.lesswrong.com/posts/9PbujppM2onhJKroC/eu-policymakers-reach-an-agreement-on-the-ai-act |
# Refinement of Active Inference agency ontology
*\[Submitted on 6 Dec 2023\]*
[**Active Inference and Intentional Behaviour**](https://arxiv.org/abs/2312.07547)
[Karl J. Friston](https://arxiv.org/search/q-bio?searchtype=author&query=Friston,+K+J), [Tommaso Salvatori](https://arxiv.org/search/q-bio?searchtype=autho... | https://www.lesswrong.com/posts/6E4L96H2PbJbfKBYS/refinement-of-active-inference-agency-ontology |
# Takeaways from a Mechanistic Interpretability project on “Forbidden Facts”
Overview
========
We tried to understand at a circuit level how Llama-2 models perform a simple task: not replying with a word when told not to. We mostly failed at fully reverse engineering the responsible circuit, but in the process we lea... | https://www.lesswrong.com/posts/Ei8q37PB3cAky6kaK/takeaways-from-a-mechanistic-interpretability-project-on |
# Contra Scott on Abolishing the FDA
Scott Alexander’s [recent post on the FDA](https://www.astralcodexten.com/p/beyond-abolish-the-fda) raises the average level of discourse on the subject. He starts from the premise that the FDA deserves destruction but cautions against rash action. No political slogan can be implem... | https://www.lesswrong.com/posts/nEGgkNgNPpL7M3BqA/contra-scott-on-abolishing-the-fda |
# [Valence series] 4. Valence & Social Status (deprecated)
**~~**
**~~**
**UPDATE JUNE 2024: THIS POST IS DEPRECATED!!**
**THE NEW VERSION IS:** [**Valence & Liking / Admiring**](https://www.lesswrong.com/posts/LaeP39jJpfPyoiSZm/valence-series-4-valence-and-liking-admiring)
**~~**
**~~**
4.1 Post summary / Table... | https://www.lesswrong.com/posts/dntWCwc8svTtbi3M7/valence-series-4-valence-and-social-status-deprecated |
# "AI Alignment" is a Dangerously Overloaded Term
Alignment as Aimability or as Goalcraft?
----------------------------------------
The Less Wrong and AI risk communities have obviously had a huge role in mainstreaming the concept of risks from artificial intelligence, but we have a serious terminology problem.
The ... | https://www.lesswrong.com/posts/sy4whuaczvLsn9PNc/ai-alignment-is-a-dangerously-overloaded-term |
# Progress links digest, 2023-12-15: Vitalik on d/acc, $100M+ in prizes, and more
**Job openings**
----------------
* [Economic Innovation Group is looking for a spring research intern](https://eig.org/about-us/careers/) (via ROP fellow [@cojobrien](https://x.com/cojobrien/status/1726656406496104604))
* [Future H... | https://www.lesswrong.com/posts/Tx2LNynGkZri9BfNF/progress-links-digest-2023-12-15-vitalik-on-d-acc-usd100m-in |
# Predicting the future with the power of the Internet (and pissing off Rob Miles)
*This is Rational Animations' second video about prediction markets. We answer a number of questions such as: what are prediction markets? What is* [*Manifold Markets*](https://manifold.markets/?referrer=Writerhttps://manifold.markets/?... | https://www.lesswrong.com/posts/BkGv9gX456NXZNfaJ/predicting-the-future-with-the-power-of-the-internet-and |
# Current AIs Provide Nearly No Data Relevant to AGI Alignment
Recently, there's been a fair amount of pushback on the "canonical" views towards the difficulty of AGI Alignment (the views I call [the "least forgiving" take](https://www.lesswrong.com/posts/3JRBqRtHBDyPE3sGa/a-case-for-the-least-forgiving-take-on-alignm... | https://www.lesswrong.com/posts/HmQGHGCnvmpCNDBjc/current-ais-provide-nearly-no-data-relevant-to-agi-alignment |
# Pope Francis shares thoughts on responsible AI development
Main sections (taken from the article):
1. The progress of science and technology as a path to peace
2. The future of artificial intelligence: between promise and risk
3. The technology of the future: machines that “learn” by themselves
4. The sense of limi... | https://www.lesswrong.com/posts/DwXiwN2DWc99eG99A/pope-francis-shares-thoughts-on-responsible-ai-development |
# Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision
Links: [Blog](https://openai.com/research/weak-to-strong-generalization), [Paper](https://cdn.openai.com/papers/weak-to-strong-generalization.pdf).
**Abstract**:
> Widely used alignment techniques, such as reinforcement learning fro... | https://www.lesswrong.com/posts/9W8roCAeEccSa3Chz/weak-to-strong-generalization-eliciting-strong-capabilities |
# Scalable Oversight and Weak-to-Strong Generalization: Compatible approaches to the same problem
*Thanks to Roger Grosse, Cem Anil, Sam Bowman, Tamera Lanham, and Mrinank Sharma for helpful discussion and comments on drafts of this post. *
*Throughout this post, "I" refers to Ansh (Buck, Ryan, and Fabien helped sub... | https://www.lesswrong.com/posts/hw2tGSsvLLyjFoLFS/scalable-oversight-and-weak-to-strong-generalization |
# cold aluminum for medicine
## cold aluminum
Very pure aluminum at the boiling point of hydrogen is a very cool material. At 20K, 99.999% pure aluminum has 1000x the electrical conductivity it has at 0 C. At 4K, it has maybe 4000x.
At such low temperatures, electron free paths in aluminum become macroscopic, which ... | https://www.lesswrong.com/posts/yG9prEve6vD4qnTfy/cold-aluminum-for-medicine |
# "Humanity vs. AGI" Will Never Look Like "Humanity vs. AGI" to Humanity
When discussing AGI Risk, people often talk about it in terms of a war between humanity and an AGI. [Comparisons between the amounts of resources](https://www.lesswrong.com/posts/odtMt7zbMuuyavaZB/when-do-brains-beat-brawn-in-chess-an-experiment#... | https://www.lesswrong.com/posts/xSJMj3Hw3D7DPy5fJ/humanity-vs-agi-will-never-look-like-humanity-vs-agi-to |
# 2022 (and All Time) Posts by Pingback Count
For the past couple years I've wished LessWrong had a "sort posts by number of pingbacks, or, ideally, by total karma of pingbacks". I particularly wished for this during the Annual Review, where "which posts got cited the most?" seemed like a useful thing to track for pot... | https://www.lesswrong.com/posts/WYqixmisE6dQjHPT8/2022-and-all-time-posts-by-pingback-count |
# Bounty: Diverse hard tasks for LLM agents
**Update 3/14/2024: This post is out of date. For current information on the task bounty, see our** [**Task Development Guide**](https://taskdev.metr.org/introduction/)**.**
Summary
=======
[METR](https://metr.org/) (formerly ARC Evals) is looking for (1) ideas, (2) detail... | https://www.lesswrong.com/posts/gAkCCaBBHD4gcwxmv/bounty-diverse-hard-tasks-for-llm-agents |
# The Serendipity of Density
One thing I really appreciate about living in Somerville is how density facilitates serendipity. Higher density means more of the places we might want to go are within walking distance, and then walking (and high density again) means we're more likely to run into friends.
This afternoon w... | https://www.lesswrong.com/posts/iS2qsNBxkmDbPpfk8/the-serendipity-of-density |
# The predictive power of dissipative adaptation
There is an idea.
The idea, which has proven somewhat memetically sticky and is at the core of the so-called "effective accelerationist" (in short, e/acc) movement, is that there is a simple and powerful principle of non-equilibrium thermodynamics which favors the prol... | https://www.lesswrong.com/posts/CwEsWktvi2iSysdon/the-predictive-power-of-dissipative-adaptation |
# What makes teaching math special
Related:
* [Arguments against constructivism (in education)?](https://www.lesswrong.com/posts/d4FmQLqCeZN7GwD2A/arguments-against-constructivism-in-education)
* [Seeking PCK (Pedagogical Content Knowledge)](https://www.lesswrong.com/posts/CsNtMunxsZnvumYLe/seeking-pck-pedagogica... | https://www.lesswrong.com/posts/mJDsnumv8Z8XFoeXS/what-makes-teaching-math-special |
# A Common-Sense Case For Mutually-Misaligned AGIs Allying Against Humans
Consider a multipolar-AGI scenario. The hard-takeoff assumption turns out to be wrong, and none of the AI Labs have a significant lead on the others. We find ourselves in a world in which there's a lot of roughly-similarly-capable AGIs. Or perha... | https://www.lesswrong.com/posts/3CAPgSzpk96PLyku3/a-common-sense-case-for-mutually-misaligned-agis-allying |
# Reviving a 2015 MacBook
A few days ago I broke the screen on my MacBook: I unplugged the power adapter and closed the top as usual, but this time the adapter had sprung back. With ~10x leverage \[1\] I cracked the screen, without even noticing.
[](https://www.jef... | https://www.lesswrong.com/posts/r429a7LGxuBckjFDb/reviving-a-2015-macbook |
# ∀: a story
I settle into my seat at the concert, pushing my earbuds in tight. The man next to me looks over, and I get the feeling he’s judging me, but it’s not enough to stop me: I heard they sometimes try to scare you with sudden loud noises, or just overwhelm you with a wall of sound until your head is aching, an... | https://www.lesswrong.com/posts/hFvhtWjXy8w2kGCGK/a-story |
# Talk: "AI Would Be A Lot Less Alarming If We Understood Agents"
This is a linkpost for a talk I gave this past summer for the ALIFE conference. If you haven't heard of it before, ALIFE (short for "artificial life") is a subfield of biology which... well, here are some of the session titles from day 1 of the conferen... | https://www.lesswrong.com/posts/mRquvigHrxEkpp5qG/talk-ai-would-be-a-lot-less-alarming-if-we-understood-agents |
# Interpreting the Learning of Deceit
One of the primary concerns when attempting to control an AI of human-or-greater capabilities is that it might be deceitful. It is, after all, fairly difficult for an AI to succeed in a coup against humanity if the humans can simply regularly ask it "Are you plotting a coup? If so... | https://www.lesswrong.com/posts/FbSAuJfCxizZGpcHc/interpreting-the-learning-of-deceit |
# Discussion: Challenges with Unsupervised LLM Knowledge Discovery
**TL;DR:** [Contrast-consistent search (CCS)](https://arxiv.org/abs/2212.03827) seemed exciting to us and we were keen to apply it. At this point, we think it is unlikely to be directly helpful for implementations of alignment strategies (>95%). Instea... | https://www.lesswrong.com/posts/wtfvbsYjNHYYBmT3k/discussion-challenges-with-unsupervised-llm-knowledge-1 |
# [Valence series] 5. “Valence Disorders” in Mental Health & Personality
5.1 Post summary / Table of contents
====================================
*Part of the *[*Valence series*](https://www.lesswrong.com/s/6uDBPacS6zDipqbZ9)*.*
Here in the final post of the Valence series, I will discuss how valence might shed lig... | https://www.lesswrong.com/posts/txj4wigyjLNbcoZ9o/valence-series-5-valence-disorders-in-mental-health-and |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.