text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Bad business advice
Most advice is bad. Whether it’s personal advice or business advice, you probably shouldn’t follow every piece you hear.
While I recognize the irony of beginning this with a piece of my own advice, there’s a *specific explanation* as to why I’ve done this. Being selective about the advice you fo... | https://www.lesswrong.com/posts/SWy9K9oDckyafTtwa/bad-business-advice |
# Critiques of the AI control agenda
*Thanks to John Wentworth, Garrett Baker, Theo Chapman, and David Lorell for feedback and discussions on drafts of this post.*
In this post I’ll describe some of my thoughts on the [AI control research agenda](https://www.lesswrong.com/posts/kcKrE9mzEHrdqtDpE/the-case-for-ensuring... | https://www.lesswrong.com/posts/j9Ndzm7fNL9hRAdCt/critiques-of-the-ai-control-agenda |
# Disrupting malicious uses of AI by state-affiliated threat actors
OpenAI just released a public announcement detailing how they caught and disrupted several cases of ongoing misuse of their models by state-affiliated threat actors, including some known to be affiliated with North Korea, Iran, China, and Russia.
Thi... | https://www.lesswrong.com/posts/xxWHkfyMM3r6GzraT/disrupting-malicious-uses-of-ai-by-state-affiliated-threat |
# Searching for Searching for Search
*Thanks to Leo Gao, Nicholas Dupuis, Paul Colognese, Janus, and Andrei Alexandru for their thoughts. *
*This post was mostly written in 2022, and pulled out of my drafts after recent conversations on the topic. My main update from then is the framing. Rather that suggesting search... | https://www.lesswrong.com/posts/b9XdMT7o54p5S4vn7/searching-for-searching-for-search |
# 11 diceware words is enough
A tweet-thread (X thread?) by [@benwr](https://www.lesswrong.com/users/benwr?mention=user) about how many words you need if you want to use the [diceware system](https://www.eff.org/dice) to create a password that will be safe for 20 years against 1 year of the whole world's password-crac... | https://www.lesswrong.com/posts/Te3NXtv9eFB6wpooG/11-diceware-words-is-enough |
# "Open Source AI" isn't Open Source
Open source software has long differentiated between “free as in speech” (libre) and “free as in beer” (gratis). In the first case, libre software has a license that allows the user freedom to view the source and modify it, understand it, and remix it. In the second case, gratis so... | https://www.lesswrong.com/posts/tvf2uvi79J2k9trhm/open-source-ai-isn-t-open-source |
# Collection of Scientific and Other Classifications
One might think all we do is stamp-collecting.
* Mathematics
* [Classification of simple finite groups](https://en.wikipedia.org/wiki/Classification_of_finite_simple_groups)
* [Classification of manifolds](https://en.wikipedia.org/wiki/Classification_of_man... | https://www.lesswrong.com/posts/G7HrhkRKeZELbFGir/collection-of-scientific-and-other-classifications |
# Raising children on the eve of AI
Cross-posted with light edits from [Otherwise](https://juliawise.net/raising-children-on-the-eve-of-ai/).
I think of us in some kind of twilight world as transformative AI looks more likely: things are about to change, and I don’t know if it’s about to get a lot darker or a lot bri... | https://www.lesswrong.com/posts/cyqrvE3dk5apg54Sk/raising-children-on-the-eve-of-ai |
# Mapping the semantic void II: Above, below and between token embeddings
**TL;DR:** [The first post in this series](https://www.lesswrong.com/posts/c6uTNm5erRrmyJvvD/mapping-the-semantic-void-strange-goings-on-in-gpt-embedding) revealed a curious kind of "stratified ontology" involving concentric hyperspherical shel... | https://www.lesswrong.com/posts/hincdPwgBTfdnBzFf/mapping-the-semantic-void-ii-above-below-and-between-token |
# Mapping the semantic void III: Exploring neighbourhoods
**TL;DR:** [The original "semantic void" post](https://www.lesswrong.com/posts/c6uTNm5erRrmyJvvD/mapping-the-semantic-void-strange-goings-on-in-gpt-embedding) documented a phenomenon tentatively described as a "stratified ontology" associated with a set of con... | https://www.lesswrong.com/posts/ii3fMmZstDFGdCHQG/mapping-the-semantic-void-iii-exploring-neighbourhoods |
# OpenAI's Sora is an agent
If you haven't already, take a look at [Sora,](https://openai.com/sora) OpenAI's new text-to-video AI. Sora can create scarily-realistic videos of nearly any subject. Unlike previous state-of-the-art AIs, the videos are coherent across time scales as long as one minute, and they can be much... | https://www.lesswrong.com/posts/bSwdbhMP9oAWzeqsG/openai-s-sora-is-an-agent |
# Fatebook for Chrome: Make and embed forecasts anywhere on the web
[Fatebook](https://forum.effectivealtruism.org/posts/DWFRBzK3rAH3HFDZr/fatebook-the-fastest-way-to-make-and-track-predictions) is the fastest way to track your predictions. Now we've made a Chrome extension that makes it even faster.
With [**Fatebook... | https://www.lesswrong.com/posts/vNhCpkSkuyARugJfn/fatebook-for-chrome-make-and-embed-forecasts-anywhere-on-the |
# Retrospective: PIBBSS Fellowship 2023
Between June and September 2023, we (Nora and Dusan) ran the second iteration of the PIBBSS Summer Fellowship. In this post, we share some of our main reflections about how the program went, and what we learnt about running it.
We first provide some **background** information... | https://www.lesswrong.com/posts/DD5EBtfJiisRkuPWF/retrospective-pibbss-fellowship-2023 |
# Addressing Feature Suppression in SAEs
*Produced as part of the ML Alignment Theory Scholars Program - Winter 2023-24 Cohort as part of Lee Sharkey's stream.*
TL;DR
=====
Sparse autoencoders are a method of resolving superposition by recovering linearly encoded “features” inside activations. Unfortunately, despite... | https://www.lesswrong.com/posts/3JuSjTZyMzaSeTxKk/addressing-feature-suppression-in-saes |
# "No-one in my org puts money in their pension"
*Epistemic status: the stories here are all as true as possible from memory, but my memory is so so.*
![](https://substackcdn.com/image/fetch/w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7... | https://www.lesswrong.com/posts/dLXdCjxbJMGtDBWTH/no-one-in-my-org-puts-money-in-their-pension |
# Every "Every Bay Area House Party" Bay Area House Party
*Inspired by* [*Ricki Heicklen's house party*](https://partiful.com/e/yEpJqHYiOj1w0JHrNcFw) *inspired by*[*Scott Alexander*](https://www.astralcodexten.com/p/every-bay-area-house-party)*.*
By the time you arrive in Berkeley, the party is already in full swing.... | https://www.lesswrong.com/posts/g5q4JiG5dzafkdyEN/every-every-bay-area-house-party-bay-area-house-party |
# The Pointer Resolution Problem
*Thanks to Garrett Baker and Evžen Wybitul for discussions and feedback on this post.*
Imagine that you meet an 18th century altruist. They tell you “So, I’ve been thinking about whether or not to eat meat. Do you know whether animals have souls?” How would you answer, assuming you ac... | https://www.lesswrong.com/posts/cGndvyTSq73efYuEM/the-pointer-resolution-problem |
# Physics-based early warning signal shows that AMOC is on tipping course
**Abstract**
------------
One of the most prominent climate tipping elements is the Atlantic meridional overturning circulation (AMOC), which can potentially collapse because of the input of fresh water in the North Atlantic. Although AMOC coll... | https://www.lesswrong.com/posts/RDxbvxGNDWyso2Jcr/physics-based-early-warning-signal-shows-that-amoc-is-on |
# 2023 Survey Results
The Data
========
0\. Population
--------------
There were 558 responses over 32 days. The spacing and timing of the responses had hills and valleys because of an experiment I was performing where I'd get the survey advertised in a different place, then watch how many new responses happened in ... | https://www.lesswrong.com/posts/WRaq4SzxhunLoFKCs/2023-survey-results |
# Making the "stance" explicit
Often, when people explain the behavior of an AI system, there is one of Daniel Dennett’s stances implicit in their explanation. I think that when this happens implicitly, it can often lead to misleading claims and sloppy reasoning. For this reason, I feel it can be really valuable to be... | https://www.lesswrong.com/posts/gqd9SKrv22Gtsb3v2/making-the-stance-explicit |
# Phallocentricity in GPT-J's bizarre stratified ontology
**TL;DR** This relates to the findings reported in my posts [Mapping the Semantic Void parts I](https://www.lesswrong.com/posts/c6uTNm5erRrmyJvvD/mapping-the-semantic-void-strange-goings-on-in-gpt-embedding) and [II](https://www.lesswrong.com/posts/hincdPwgBTfd... | https://www.lesswrong.com/posts/FTY9MtbubLDPjH6pW/phallocentricity-in-gpt-j-s-bizarre-stratified-ontology |
# Self-Awareness: Taxonomy and eval suite proposal
*\[This is a mildly-edited version of a google doc I wrote at OpenAI in July 2022. I had intended to get it published in some form, but never got around to it for various reasons. I have now received approval to put it up as a blog post. **The main thing of interest h... | https://www.lesswrong.com/posts/QEBFZtP64DdhjE3Sz/self-awareness-taxonomy-and-eval-suite-proposal |
# Evaluating Solar
In 2018, [we installed](https://www.jefftk.com/p/solar-one-year-in) 14 solar panels on the Northwest roof of our house. This is a little silly: wouldn't you want to install them facing the sun? The problem was the enormous tree to our Southeast, twice the height of our house, completely shading that... | https://www.lesswrong.com/posts/bfL3psNdgrZPsQHEs/evaluating-solar |
# "What if we could redesign society from scratch? The promise of charter cities." [Rational Animations video]
*In this video, we describe a speculative but promising idea to boost economic growth: giving cities the ability to write some of their own laws and design their own institutions, independently from their pa... | https://www.lesswrong.com/posts/9FRNc9ygpLSDh4pkX/what-if-we-could-redesign-society-from-scratch-the-promise |
# Celiacs don't need to live in fear
TL;DR: Less sensitive celiacs can probably have trace amounts of gluten cross contamination with low risk of health problems, since celiacs who deliberately eat non-trivial amounts of gluten [are mostly ok](https://bmcmedicine.biomedcentral.com/articles/10.1186/s12916-020-1511-6).
... | https://www.lesswrong.com/posts/adkGrthu5CpGycfqn/celiacs-don-t-need-to-live-in-fear |
# Intuition for 1 + 2 + 3 + … = -1/12
You might have been led to believe the only way to make sense of the equation $1 + 2 + 3 + … = -\dfrac{1}{12}$ is with "zeta function regularization," involving analytic continuation or similar high-tech tools, and that there is no way to develop intuition for this.
You were prob... | https://www.lesswrong.com/posts/4jANL9cZj28uALwsE/intuition-for-1-2-3-1-12 |
# Things I've Grieved
I think [grieving](https://www.lesswrong.com/tag/grieving) is a fundamental rationality skill. Often, the difference between the Winning Move, and your Current Path, is that there is something really beautiful and good about your current path. Or there was something actually horrifying about real... | https://www.lesswrong.com/posts/EvyzcG7Q8jG5gftfE/things-i-ve-grieved |
# Weighing reputational and moral consequences of leaving Russia or staying
I've been thinking a lot about whether I should consider emigrating to another country (currently I'm in Russia). I'm not involved in politics or the military action, but I can't help but feel implicated by my government's actions as a citizen... | https://www.lesswrong.com/posts/bGugEDFb6ge8sMxyB/weighing-reputational-and-moral-consequences-of-leaving |
# Scientific Method
Here I will try to describe the scientific method in simple words **— **what it is, how it works, and how we can break it into a list of easily understandable singular ideas, in this article called 'points'.
First of all, the scientific method is not carved on tablets, like the commandments of Mos... | https://www.lesswrong.com/posts/TafThcbAowLBFTKok/scientific-method |
# Conspiracy Investigation Done Right
[](https://substackcdn.com/image/fetch/f_auto,q_auto:good,fl_pr... | https://www.lesswrong.com/posts/4Cp3HSZwWTERhei9Y/conspiracy-investigation-done-right |
# On coincidences and Bayesian reasoning, as applied to the origins of COVID-19
(Or: sometimes heuristics are no substitute for a deep dive into all of the available information).
This post is a response to Roko's recent series of posts ([Brute Force Manufactured Consensus is Hiding the Crime of the Century](https:... | https://www.lesswrong.com/posts/NrKQGyggC7jcersuJ/on-coincidences-and-bayesian-reasoning-as-applied-to-the |
# I'd also take $7 trillion
So, Sam Altman wants as much money as he can get - the number he chose to give was $7 trillion - to build more chips for AI as fast as possible. Including in places like the UAE. Thus undermining his argument that OpenAI is helping with AI safety by preventing overhang and maintaining a lea... | https://www.lesswrong.com/posts/gEwXe3jrfANSqyvBS/i-d-also-take-usd7-trillion |
# When Should Copyright Get Shorter?
Information goods have private costs and public benefits. Pharmaceutical research requires upfront investment from one firm, but once a certain compound is known to be effective it’s cheap for any company to produce. Writing a novel or filming a movie takes hundreds of hours of wor... | https://www.lesswrong.com/posts/z766eMZavaTuCqGqL/when-should-copyright-get-shorter |
# Protocol evaluations: good analogies vs control
Let’s say you want to use AIs that are capable of causing x-risk. You have a plan that describes how to train, validate and deploy AIs. For example, you could use pretraining+RLHF, check that it doesn’t fall for honeypots and never inserts a single backdoor in your cod... | https://www.lesswrong.com/posts/qhaSoR6vGmKnqGYLE/protocol-evaluations-good-analogies-vs-control |
# How Technical AI Safety Researchers Can Help Implement Punitive Damages to Mitigate Catastrophic AI Risk
In the [previous post](https://www.lesswrong.com/posts/5e7TrmH7mBwqpZ6ek/tort-law-can-play-an-important-role-in-mitigating-ai-risk), I explained how punitive damages could “pull forward” the expected liability as... | https://www.lesswrong.com/posts/DvizboRCsMx6AEZqq/how-technical-ai-safety-researchers-can-help-implement |
# Retirement Accounts and Short Timelines
Sometimes I talk to people who don't use retirement accounts because they think the world will change enormously between now and when they're older. Something like, the most likely outcomes are that things go super well and they won't need the money, or things go super poorly ... | https://www.lesswrong.com/posts/CcqaJFf7TvAjuZFCx/retirement-accounts-and-short-timelines |
# Abs-E (or, speak only in the positive)
"Positive" and "negative", as used here, refer to truth-values and logical structure, rather than emotional valence.
On almost any topic, there are many potential false statements, and comparatively very few[^soy8gnezvk] true statements. Thus a [simple true statement (like "th... | https://www.lesswrong.com/posts/W8CxEFCnYNdrHkuoB/abs-e-or-speak-only-in-the-positive |
# Auditing LMs with counterfactual search: a tool for control and ELK
Consider a scalable oversight setting where a super-human model knows of a subtle flaw[^1] in an input (plan, code, or argument) but also knows that humans, with their limited knowledge, would overlook this flaw. In this case, ordinary language mode... | https://www.lesswrong.com/posts/Y2nxEbDfcoWhC8C2D/auditing-lms-with-counterfactual-search-a-tool-for-control |
# flowing like water; hard like stone
I noticed in your recent post that you described "rhetorical aikido", which you called a Daoist technique. I have studied the Tao Te Ching's text a fair bit over the past few years, and I find its philosophy to be quite useful as a "theme" or "virtue".
I've not had the opportunit... | https://www.lesswrong.com/posts/WeSovg9Mpw3Rme7zB/flowing-like-water-hard-like-stone |
# Lessons from Failed Attempts to Model Sleeping Beauty Problem
*This is the seventh post in my series on Anthropics. The previous one is* [*Why Two Valid Answers Approach is not Enough for Sleeping Beauty*](https://www.lesswrong.com/posts/zT9QGTv3zsnqorim7/why-two-valid-answers-approach-is-not-enough-for-sleeping)*. ... | https://www.lesswrong.com/posts/SjoPCwmNKtFvQ3f2J/lessons-from-failed-attempts-to-model-sleeping-beauty |
# Difficulty classes for alignment properties
*I don't think this idea is particularly novel, but it comes up often in conversations I have, so I figured it'd be good to write it up.*
How do you prevent deception from AI systems?
One obvious thing to try would be to make sure that your model never thinks deceptive t... | https://www.lesswrong.com/posts/jeSZmMxPDspLxQ9bd/difficulty-classes-for-alignment-properties |
# Selections From "The Trouble With Being Born"
I've been aware for some time that there was a Romanian-Parisian existentialist philosopher named Emil Cioran who wrote *On the Heights of Despair* and *The Trouble With Being Born*. The other day when mentioning him I pulled up a copy of [*Trouble*](https://ia903405.us.... | https://www.lesswrong.com/posts/eRBdijjQdKDz2Zrjj/selections-from-the-trouble-with-being-born |
# Monthly Roundup #15: February 2024
Another month. More things. Much roundup.
#### Bad News
[Jesse Smith writes in Asterisk](https://asteriskmag.com/issues/05/lies-damned-lies-and-manometer-readings) that our HVAC workforce is both deeply incompetent and deeply corrupt. This certainly [matches my own experience](ht... | https://www.lesswrong.com/posts/zHTRivdJ7ZDSctwqi/monthly-roundup-15-february-2024 |
# Inducing human-like biases in moral reasoning LMs
**Meta.** This post is less polished than we would ideally prefer. However, we still think publishing it as is is reasonable, to avoid further delays. We are open to answering questions and to feedback in the comments.
**TL;DR**. This presents an inconclusive attemp... | https://www.lesswrong.com/posts/eruHcdS9DmQsgLqd4/inducing-human-like-biases-in-moral-reasoning-lms |
# Why does generalization work?
*Just an interesting philosophical argument*
I. Physics
==========
Why can an ML model learn from part of a distribution or data set, and generalize to the rest of it? Why can I learn some useful heuristics or principles in a particular context, and later apply them in other areas of ... | https://www.lesswrong.com/posts/NwDvmodb2CcuysQxW/why-does-generalization-work-1 |
# The Third Gemini
\[Editor’s Note: I forgot to cross-post this on Thursday, sorry about that. Note that this post does not cover Gemini 1.5, which was announced after I posted this. I will cover 1.5 later this week.\]
We have now had a little over a week with Gemini Advanced, based on Gemini Ultra. A few reviews are... | https://www.lesswrong.com/posts/vdSE99ssADJEXP567/the-third-gemini |
# AI #51: Altman’s Ambition
\[Editor’s note: I forgot to post this to WorldPress on Thursday. I’m posting it here now. Sorry about that.\]
Sam Altman is not playing around.
He wants to build new chip factories in the decidedly unsafe and unfriendly UAE. He wants to build up the world’s supply of energy so we can run... | https://www.lesswrong.com/posts/gBHNw5Ymnqw8FiMjh/ai-51-altman-s-ambition |
# Gut Renovating Another Bathroom
After our long-term tenants bought a house and moved out, we had a few months free before we had new downstairs neighbors. A good time to gut renovate the first floor bathroom! It was a lot of work, stressful at times when I was under time pressure, but also, a lot of fun. It helped t... | https://www.lesswrong.com/posts/GJgk6jEMAuZ3D4mA2/gut-renovating-another-bathroom |
# Jailbreaking GPT-4 with the tool API
**WARNING: Contains mentions of racism, sexism, homophobia, and violence. We don't include the bad GPT-4 completions in this post, we link out to them.**
OpenAI’s API has a [tool API](https://platform.openai.com/docs/api-reference/chat/create#chat-create-tools). Basically, when ... | https://www.lesswrong.com/posts/ztBL25LuvmJ4DD3uJ/jailbreaking-gpt-4-with-the-tool-api |
# A Tale of Two Restaurant Types
While I sort through whatever is happening with GPT-4, today’s scheduled post is two recent short stories about restaurant selection.
#### Ye Olde Restaurante
[Tyler Cowen says that restaurants saying ‘since year 19xx](https://marginalrevolution.com/marginalrevolution/2023/12/around-... | https://www.lesswrong.com/posts/Hm38rqATujCDbLFrh/a-tale-of-two-restaurant-types |
# Dual Wielding Kindle Scribes
*This is an informal post intended to describe a workflow / setup that I found very useful, so that others might consider adopting or experimenting with facets of it that they find useful.*
... | https://www.lesswrong.com/posts/WhC6LtxgaLFBNXEsk/dual-wielding-kindle-scribes |
# Weak vs Quantitative Extinction-level Goodhart's Law
*This post overlaps with our recent paper* [Extinction Risk from AI: Invisible to Science?](http://arxiv.org/abs/2403.05540)*.*
**tl;dr:** With claims such as "optimisation towards a misspecified goal will cause human extinction", we should be more explicit about... | https://www.lesswrong.com/posts/KQPScBNg2bGKfNi7y/weak-vs-quantitative-extinction-level-goodhart-s-law |
# Which Model Properties are Necessary for Evaluating an Argument?
*This post overlaps with our recent paper* [Extinction Risk from AI: Invisible to Science?](http://arxiv.org/abs/2403.05540)*.*
**Summary:** If you want to use a mathematical model to evaluate an argument, that model needs to allow for dynamics that a... | https://www.lesswrong.com/posts/n7MgrFw5abLHSFkDz/which-model-properties-are-necessary-for-evaluating-an |
# Dynamics Crucial to AI Risk Seem to Make for Complicated Models
*This post overlaps with our recent paper* [Extinction Risk from AI: Invisible to Science?](http://arxiv.org/abs/2403.05540)*.*
**tl;dr:** In AI safety, we are worried about certain problems with using powerful AI. (For example, the difficulty of value... | https://www.lesswrong.com/posts/krKKcYE6bromZ32gk/dynamics-crucial-to-ai-risk-seem-to-make-for-complicated |
# Extinction-level Goodhart's Law as a Property of the Environment
**Summary:** Formally defining *Extinction-level Goodhart's Law* is tricky, because formal environments don't contain any actual humans that could go extinct. But we can do it using the notion of an *interpretation* mapping, which sends outcomes in the... | https://www.lesswrong.com/posts/TEauZ9vxQLTL5qbkj/extinction-level-goodhart-s-law-as-a-property-of-the |
# Extinction Risks from AI: Invisible to Science?
**Abstract:** In an effort to inform the discussion surrounding existential risks from AI, we formulate Extinction-level Goodhart’s Law as "*Virtually any goal specification, pursued to the extreme, will result in the extinction*[^wwed95q0dgc]* of humanity*'', and we a... | https://www.lesswrong.com/posts/d5oqvgCR7SDf5m4k4/extinction-risks-from-ai-invisible-to-science |
# Analogies between scaling labs and misaligned superintelligent AI
TL;DR: Scaling labs have their own alignment problem analogous to AI systems, and there are some similarities between the labs and misaligned/unsafe AI.
### Introduction
Major AI scaling labs (OpenAI/Microsoft, Anthropic, Google/DeepMind, and Meta)... | https://www.lesswrong.com/posts/fpws8WNo6Mp4KsQjH/analogies-between-scaling-labs-and-misaligned |
# AISN #31: A New AI Policy Bill in California
Plus, Precedents for AI Governance and The EU AI Office
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://newsletter.safe.ai/sub... | https://www.lesswrong.com/posts/rGsrDiKsdFpXvWjye/aisn-31-a-new-ai-policy-bill-in-california-plus-precedents |
# The Pareto Best and the Curse of Doom
I.
--
Prerequisite reading: [Being the (Pareto) Best in the World](https://www.lesswrong.com/posts/XvN2QQpKTuEzgkZHY/being-the-pareto-best-in-the-world).
A summary of Being the (Pareto) Best in the World: Being the world's best mathematician is hard. Being the world's best mus... | https://www.lesswrong.com/posts/eD5g9sCiZCEcdQXS6/the-pareto-best-and-the-curse-of-doom |
# Job Listing: Managing Editor / Writer
MIRI is hiring a managing editor and one or more writers. Here's the [job listing](https://jobs.ashbyhq.com/miri/e07416be-e761-405c-b46d-e017324108da). We're accepting applications until March 13th.
I am the Communications Manager at MIRI and I'm the hiring manager for these po... | https://www.lesswrong.com/posts/j5twfHKz5mehjFMAB/job-listing-managing-editor-writer |
# The Byronic Hero Always Loses
Why do we root for the antiheroes?
Walter White, Light Yagami, Lucifer Morningstar, Doctor Frankenstein... It seems to be natural to root for the bad guys, and perhaps even more natural for rationalists.
“I must confess I have always had some sympathy with villains. Heroism makes fine... | https://www.lesswrong.com/posts/k5BCFx3DYeXF3fgo3/the-byronic-hero-always-loses |
# Notes on Internal Objectives in Toy Models of Agents
*Thanks to Jeremy Gillen and Arun Jose for discussions related to these ideas.*
Summary
=======
***WARNING:** The quality of this post is low. It was sitting in my drafts folder for a while, yet I decided to post it because some people found these examples and a... | https://www.lesswrong.com/posts/yRC6dQAGcjpkpwwe9/notes-on-internal-objectives-in-toy-models-of-agents |
# How are voluntary commitments on vulnerability reporting going?
*This article is written solely in my personal capacity, and does not represent the views of any organisations I am affiliated with.*
The [UK](https://www.aisafetysummit.gov.uk/policy-updates/#company-policies) and [US](https://www.whitehouse.gov/brief... | https://www.lesswrong.com/posts/5mN3Qd4JoLsaZKDcR/how-are-voluntary-commitments-on-vulnerability-reporting |
# The One and a Half Gemini
Previously: I hit send on [The Third Gemini](https://thezvi.substack.com/p/the-third-gemini), and within half an hour [DeepMind announced Gemini 1.5](https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024/#sundar-note).
So this covers Gemini 1.5. One million to... | https://www.lesswrong.com/posts/N2Y664LX6pQ8rFiz2/the-one-and-a-half-gemini |
# Everything Wrong with Roko's Claims about an Engineered Pandemic
Premises
--------
Here I tackle some bold claims made by Roko in three separate posts about SARS-CoV-2 being an engineered virus ([Brute Force Manufactured Consensus is Hiding the Crime of the Century](https://www.lesswrong.com/posts/bMxhrrkJdEormCcLt... | https://www.lesswrong.com/posts/bAEPBD7JBsf9XKCuT/everything-wrong-with-roko-s-claims-about-an-engineered |
# Do sparse autoencoders find "true features"?
*Thanks to Joseph Bloom and James Oldfield for giving feedback on drafts which helped improve the post*
In this post I'll discuss an apparent limitation of sparse autoencoders (SAEs) in their current formulation as they are applied to discovering the latent features with... | https://www.lesswrong.com/posts/QoR8noAB3Mp2KBA4B/do-sparse-autoencoders-find-true-features |
# Sora What
Hours after [Google announced Gemini 1.5](https://thezvi.substack.com/p/the-one-and-a-half-gemini), [OpenAI announced their new video generation model Sora](https://openai.com/sora). Its outputs look damn impressive.
#### How Sora Works
How does it work? [There is a technical report](https://openai.com/r... | https://www.lesswrong.com/posts/35fZ6csrbcrKw9BwG/sora-what |
# Language Models Don't Learn the Physical Manifestation of Language
Abstract
========
We argue that there are certain properties of language that our current large language models (LLMs) don't learn. We present an empirical investigation of visual-auditory properties of language through a series of tasks, termed H-T... | https://www.lesswrong.com/posts/ia4HszGTidh74Nyxk/language-models-don-t-learn-the-physical-manifestation-of |
# The Gemini Incident
\[Original title; Gemini Has a Problem\]
[Google’s Gemini 1.5 is impressive](https://thezvi.substack.com/p/the-one-and-a-half-gemini) and I am excited by its huge context window. I continue to [default to Gemini Advanced as my default AI for everyday use](https://thezvi.substack.com/p/the-third-... | https://www.lesswrong.com/posts/kLTyeG7R8eYpFwe3H/the-gemini-incident |
# Embed your second brain in your first brain
Any explicit knowledge is either part of a general, long-term understanding of the world (global knowledge), or peculiar to your own life and local environment (local knowledge). These mainly fill semantic and episodic memory, respectively. Declarative memory can hold eith... | https://www.lesswrong.com/posts/ofd5P3MHFBd9uAwDM/embed-your-second-brain-in-your-first-brain |
# AI #52: Oops
We were treated to technical marvels this week.
At Google, [**they announced Gemini Pro 1.5**](https://thezvi.substack.com/p/the-one-and-a-half-gemini), with a million token context window within which it has excellent recall, using mixture of experts to get Gemini Advanced level performance (e.g. GPT-... | https://www.lesswrong.com/posts/WmxS7dbHuxzxFei64/ai-52-oops |
# Contra Ngo et al. “Every ‘Every Bay Area House Party’ Bay Area House Party”
*With thanks to Scott Alexander for* [*the inspiration*](https://www.astralcodexten.com/p/every-bay-area-house-party)*, Jeffrey Ladish, Philip Parker, Avital Morris, and Drake Thomas for masterful cohosting, and Richard Ngo for his* [*invest... | https://www.lesswrong.com/posts/mmYFF4dyi8Kg6pWGC/contra-ngo-et-al-every-every-bay-area-house-party-bay-area |
# Complexity of value but not disvalue implies more focus on s-risk. Moral uncertainty and preference utilitarianism also do.
As per title. I often talk to people that have views that I think should straightforwardly imply a larger focus on s-risk than they think. In particular, people often seem to endorse something ... | https://www.lesswrong.com/posts/zPsJECBYms5omXxfe/complexity-of-value-but-not-disvalue-implies-more-focus-on-s |
# Shaming with and without naming
Suppose someone wrongs you and you want to emphatically mar their reputation, but only insofar as doing so is conducive to the best utilitarian outcomes. I was thinking about this one time and it occurred to me that there are at least two fairly different routes to positive utilitaria... | https://www.lesswrong.com/posts/aNCDSrGZ6kJvZviko/shaming-with-and-without-naming-1 |
# Parasocial relationship logic
If:
1. You become like the five people you spend the most time with (or something remotely like that)
2. The people who are most extremal in good ways tend to be highly successful
Should you try to have 2-3 of your five relationships be parasocial ones with people too succ... | https://www.lesswrong.com/posts/fhsdSqAFS3GguRrjQ/parasocial-relationship-logic-1 |
# Deep and obvious points in the gap between your thoughts and your pictures of thought
Some [ideas](https://worldspiritsockpuppet.substack.com/p/bodega-bay-workshop-18-11-26) feel either deep or extremely obvious. You’ve heard some trite truism your whole life, then one day an epiphany lands and you try to save it wi... | https://www.lesswrong.com/posts/HzLfrE57JqkjDW33c/deep-and-obvious-points-in-the-gap-between-your-thoughts-and-1 |
# In set theory, everything is a set
In Unix, everything is a file; in set theory, everything is a set.
* * *
This abstraction is very cool and allows us to prove facts that would not be provable if we thought that sets could include stuff that were not sets and therefore could not be broken down further.
Here's a ... | https://www.lesswrong.com/posts/ohBquGStc2LyXFPi8/in-set-theory-everything-is-a-set |
# The Shutdown Problem: Incomplete Preferences as a Solution
**Preamble**
============
This post is an updated explanation of the POST-Agents Proposal (PAP): my proposed solution to the shutdown problem.[^g6wtgrwwuj4] The post is shorter than my AI Alignment Awards [contest entry](https://s3.amazonaws.com/pf-user-fil... | https://www.lesswrong.com/posts/YbEbwYWkf8mv9jnmi/the-shutdown-problem-incomplete-preferences-as-a-solution |
# Deliberative Cognitive Algorithms as Scaffolding
As rationalists, we are interested in finding systematic techniques that boost our effective intelligence. Because we tend to be mathematical thinkers, we usually look for precise algorithmic solutions, such as computing the optimal inference laws of Bayesian probabil... | https://www.lesswrong.com/posts/NXcm2zWx2MG4sbQio/deliberative-cognitive-algorithms-as-scaffolding |
# Why you, personally, should want a larger human population
What is the ideal size of the human population?
One common answer is “much smaller.” Paul Ehrlich, co-author of *The Population Bomb* (1968), has as recently as 2018 promoted the idea that “[the world’s optimum population is less than two billion people](ht... | https://www.lesswrong.com/posts/LDRAj5zEYGKbhcsF3/why-you-personally-should-want-a-larger-human-population |
# A starting point for making sense of task structure (in machine learning)
ML models can perform a range of tasks and subtasks, some of which are more closely related to one another than are others. In this post, we set out two very initial starting points. First, we motivate reverse engineering models’ task decompos... | https://www.lesswrong.com/posts/exp4JGPJu46g6sdRp/a-starting-point-for-making-sense-of-task-structure-in |
# Instrumental deception and manipulation in LLMs - a case study
*Produced as part of Astra Fellowship - Winter 2024 program, mentored by Evan Hubinger. Thanks to Evan, Sam Svenningsen and Bogdan-Ionut Cirstea for suggestions and discussions on the topic.*
**Summary: **Arguments for deceptive alignment rely, among ot... | https://www.lesswrong.com/posts/vTJt3Rw44HXotHBxu/instrumental-deception-and-manipulation-in-llms-a-case-study |
# The Sense Of Physical Necessity: A Naturalism Demo (Introduction)
*Note on genre: This sequence is a demonstration of a complete naturalist study, as described in* [*Intro to Naturalism*](https://www.lesswrong.com/s/evLkoqsbi79AnM5sz) *and* [*The Nuts and Bolts Of Naturalism*](https://www.lesswrong.com/s/zLib3j2Fdnn... | https://www.lesswrong.com/posts/8iab2addq4MzYHKij/the-sense-of-physical-necessity-a-naturalism-demo |
# How well do truth probes generalise?
Representation engineering (RepEng) has emerged as a promising research avenue for model interpretability and control. Recent papers have proposed methods for [discovering truth in models with unlabeled data](https://arxiv.org/abs/2212.03827), [guiding generation by modifying rep... | https://www.lesswrong.com/posts/cmicXAAEuPGqcs9jw/how-well-do-truth-probes-generalise |
# Balancing Games
When I play an N-player game I want everyone to both:
* Try to win
* Win about 1/N of the time
With many games and groups of participants these are in conflict: if I play bridge against my kids I'm going to win all the time, but I'm not very good at the game so if I play against people who are ... | https://www.lesswrong.com/posts/gsRS4pjLsAJ6jgf3E/balancing-games |
# We Need Major, But Not Radical, FDA Reform
Fellow [progress blogger Alex Telford](https://substack.com/@atelfo) and I have had a friendly back-and-forth going over FDA reform. Alex suggests incremental reforms to the FDA, which I strongly support, but these don’t go far enough. The FDA’s failures merit a [complete o... | https://www.lesswrong.com/posts/ACwFHiAQXxXoK2as6/we-need-major-but-not-radical-fda-reform |
# Exercise: Planmaking, Surprise Anticipation, and "Baba is You"
This is an exercise about **Planmaking and Surprise-Anticipation.** It takes about 2-3 hours. It's a small, simplified exercise, but I think it's a useful building block.
Humans often solve complex problems via iteration and empiricism. Usually, trying ... | https://www.lesswrong.com/posts/jqb3prwGQjLriq7Lu/exercise-planmaking-surprise-anticipation-and-baba-is-you |
# Rationality Research Report: Towards 10x OODA Looping?
6 months ago I wrote [Feedbackloop-first Rationality](https://www.lesswrong.com/posts/pZrvkZzL2JnbRgEBC/feedbackloop-first-rationality). I didn't followup on it for awhile (except for sporadic [Deliberate (“Purposeful?”) Practice Club](https://www.lesswrong.com/... | https://www.lesswrong.com/posts/FEMWS79AFTmKq2iJK/rationality-research-report-towards-10x-ooda-looping |
# Choosing My Quest (Part 2 of "The Sense Of Physical Necessity")
*This is the second post in a sequence that demonstrates a complete naturalist study, specifically a study of* [*query hugging*](https://www.lesswrong.com/posts/2jp98zdLo898qExrr/hug-the-query) *(sort of), as described in* [*The Nuts and Bolts of Natura... | https://www.lesswrong.com/posts/CDzAwDxK2GnxBpu7h/choosing-my-quest-part-2-of-the-sense-of-physical-necessity |
# Cooperating with aliens and AGIs: An ECL explainer
Summary
=======
Evidential cooperation in large worlds (ECL) is a proposed way of reaping gains—that is, getting more of what we value instantiated—through cooperating with agents across the universe/multiverse. Such cooperation does *not* involve physical, or *cau... | https://www.lesswrong.com/posts/eEj9A9yMDgJyk98gm/cooperating-with-aliens-and-distant-agis-an-ecl-explainer |
# Everett branches, inter-light cone trade and other alien matters: Appendix to “An ECL explainer”
This is an appendix to “[Cooperating with aliens and (distant) AGIs: An ECL explainer](https://forum.effectivealtruism.org/posts/JGazpLa3Gvvter4JW/cooperating-with-aliens-and-distant-agis-an-ecl-explainer-1)”. The sectio... | https://www.lesswrong.com/posts/acEcYXibtDFEZTJ9f/everett-branches-inter-light-cone-trade-and-other-alien |
# Deconfusing In-Context Learning
I see people use "in-context learning" in different ways.
Take the opening to "[In-Context Learning Creates Task Vectors](https://arxiv.org/abs/2310.15916)":
> In-context learning (ICL) in Large Language Models (LLMs) has emerged as a powerful new learning paradigm. However, its und... | https://www.lesswrong.com/posts/vkCCbPzmMhJgvbtfK/deconfusing-in-context-learning |
# Ideological Bayesians
***TLDR:** It’s often said that Bayesian updating is unbiased and converges to the truth—and, therefore, that biases must emerge from* non-*Bayesian sources. That's not quite right. The convergence results require updating on your **total** evidence—but for agents at all like us, that's impossi... | https://www.lesswrong.com/posts/7kXSnFWSChqYpa3Nn/ideological-bayesians |
# China-AI forecasts
The rate at which China is able to advance towards TAI is a crucial consideration in for many policy questions. My current take is that, without significant political reforms which seem very unlikely while Xi is alive (although considerably more likely after his death,) it’s very unlikely that Chi... | https://www.lesswrong.com/posts/ysuXxa5uarpGzrTfH/china-ai-forecasts |
# Cellular respiration as a steam engine
When I helped an interlocutor learn about metabolism, we made an analogy between steam engines and cellular respiration. I find that people (at first) know steam engines better than cellular respiration, so I show the analogy here as a guide to cells assuming you understand eng... | https://www.lesswrong.com/posts/PhbEFTWd6pYeSYhET/cellular-respiration-as-a-steam-engine |
# Hidden Cognition Detection Methods and Benchmarks
*Thanks to Johannes Treutlein for discussions and feedback.*
Introduction
============
An AI may be able to hide cognition that leads to negative outcomes from certain oversight processes (such as [deceptive alignment](https://www.lesswrong.com/tag/deceptive-align... | https://www.lesswrong.com/posts/fDRxctsj5zsbyKwYN/hidden-cognition-detection-methods-and-benchmarks |
# How I build and run behavioral interviews
This is an adaptation of an internal doc I wrote for [Wave](https://www.wave.com/en/).
I used to think that behavioral interviews were basically useless, because it was too easy for candidates to bullshit them and too hard for me to tell what was a good answer. I’d end up g... | https://www.lesswrong.com/posts/HhaB64CTfWofgGuLL/how-i-build-and-run-behavioral-interviews |
# Acting Wholesomely
Overview of essay series
========================
This is the first in a collection of three essays exploring and ultimately defending the idea of *choosing what feels wholesome* as a heuristic for picking actions which are good for the world. I'm including a summary of the series here, before ge... | https://www.lesswrong.com/posts/Cb7oajdrA5DsHCqKd/acting-wholesomely |
# Examining Language Model Performance with Reconstructed Activations using Sparse Autoencoders
*Note: The second figure in this post originally contained a bug pointed out by* [*@LawrenceC*](https://www.lesswrong.com/users/lawrencec?mention=user)*, which has since been fixed.*
Summary
=======
* Sparse Autoencode... | https://www.lesswrong.com/posts/8QRH8wKcnKGhpAu2o/examining-language-model-performance-with-reconstructed |
# How I internalized my achievements to better deal with negative feelings
Whenever I struggle to make progress on an important goal, I feel bad. I get feelings of frustration, impatience, and apathy. I think to myself, “I have wasted all this time, and I will never get it back.” The resulting behavior during these mo... | https://www.lesswrong.com/posts/eRXYqM8ffKsnDu7iz/how-i-internalized-my-achievements-to-better-deal-with |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.