text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Manifold Markets
Manifold Markets is a prediction market platform where I've been
trading since September. This post will compare it to other prediction
markets that I've used.
### Play Money
The most important fact about Manifold is that traders bet mana, which
is for most purposes not real money. You can ... | https://www.lesswrong.com/posts/ptEtB4wbLixRuy8MG/manifold-markets-1 |
# Solving alignment isn't enough for a flourishing future
*I'm sharing a paper I previously presented to the* [*Stanford Existential Risks Conference*](https://seri.stanford.edu/events/stanford-existential-risks-conference-april-20-22-2023) *in April 2023. In past years, much discussion of making the long-term future ... | https://www.lesswrong.com/posts/uHcJyKcszugFkwhFs/solving-alignment-isn-t-enough-for-a-flourishing-future |
# What Failure Looks Like is not an existential risk (and alignment is not the solution)
Introduction
============
Among those thinking that AI is an existential risk, there seems to be significant disagreement on what the main threat model is. Threat model uncertainty makes it harder to reduce this risk: a faulty th... | https://www.lesswrong.com/posts/tyHW6tEGzoErZXZ4x/what-failure-looks-like-is-not-an-existential-risk-and |
# Most experts believe COVID-19 was probably not a lab leak
The Global Catastrophic Risks Institute conducted an anonymous survey of relevant experts on whether they thought COVID was more likely caused by a lab accident (aka lab leak) or zoonotic spillover. Their summary, bolding is mine:
> The study’s experts overa... | https://www.lesswrong.com/posts/28hnPFiAoMkJssmf3/most-experts-believe-covid-19-was-probably-not-a-lab-leak |
# On Dwarkesh’s 3rd Podcast With Tyler Cowen
[This post is extensive thoughts on Tyler Cowen’s excellent talk with Dwarkesh Patel](https://www.youtube.com/watch?v=EY2nbAVZB-k&ab_channel=DwarkeshPatel).
[
Review Facts
------------
There were 5330 posts published in 2022.
Here's how many posts passed through the d... | https://www.lesswrong.com/posts/zvTFERkqkhYNsk7cL/voting-results-for-the-2022-review |
# Survey for alignment researchers!
***UPDATE 3/9**: Thanks to broad participation from the community, this and associated surveys have raised approximately $10,000 for high-impact alignment organizations. Given the reasonable sample size we now have, we are now going to pause donations for any subsequent responses. (... | https://www.lesswrong.com/posts/LCeGoqZA4KnkyxEHD/survey-for-alignment-researchers |
# Announcing the London Initiative for Safe AI (LISA)
****
*The LISA Team consists of James Fox, Mike Brozowski, Joe Murray, ... | https://www.lesswrong.com/posts/P7dbHRYfwykEJiDYX/announcing-the-london-initiative-for-safe-ai-lisa |
# Why do we need RLHF? Imitation, Inverse RL, and the role of reward
*This is a cross-post from my* [*Substack*](https://latentobservations.substack.com/p/why-do-we-need-rlhf-imitation-inverse) *to get some feedback from the alignment community. I have condensed some sections to get the message across.*
It's rarely ... | https://www.lesswrong.com/posts/Rs9ukRphwg3pJeYRF/why-do-we-need-rlhf-imitation-inverse-rl-and-the-role-of |
# Attention SAEs Scale to GPT-2 Small
*This is an interim report that we are currently building on. We hope this update + open sourcing our SAEs will be useful to related research occurring in parallel. Produced as part of the*[* ML Alignment & Theory Scholars Program*](https://www.matsprogram.org/) *\- Winter 2023-24... | https://www.lesswrong.com/posts/FSTRedtjuHa4Gfdbr/attention-saes-scale-to-gpt-2-small |
# Why I no longer identify as transhumanist
Someone asked me how come I [used to have](https://xuenay.livejournal.com/297944.html) a strong identity as a [singularitarian](https://en.wikipedia.org/wiki/Singularitarianism) / transhumanist but don’t have it anymore. Here’s what I answered them:
——-
So I think the shor... | https://www.lesswrong.com/posts/7Lp6CLz93DvmyGmTY/why-i-no-longer-identify-as-transhumanist |
# Practicing my Handwriting in 1439
Tyler Cowen’s [latest Bloomberg column](https://www.bloomberg.com/opinion/articles/2024-01-25/ai-s-greatest-danger-the-humans-who-use-it?sref=htOHjx5Y) on the dangers of AI predicts, among other things, that AI will disrupt the current high status and earnings of “wordcels” and “ide... | https://www.lesswrong.com/posts/PaHARSnMhn2gjCog7/practicing-my-handwriting-in-1439 |
# Attitudes about Applied Rationality
*tl;dr: within the LW community, there are many clusters of strategies to achieve rationality: doing basic exercices, using jargon, reading, partaking workshops, privileging object-level activities, and several other opinions like putting an accent on feedback loops, difficult con... | https://www.lesswrong.com/posts/5jdqtpT6StjKDKacw/attitudes-about-applied-rationality |
# Brute Force Manufactured Consensus is Hiding the Crime of the Century
*People often parse information through an epistemic consensus filter. They do not ask "is this true", they ask "will others be OK with me thinking this is true". This makes them very malleable to **brute force manufactured consensus**; if every s... | https://www.lesswrong.com/posts/bMxhrrkJdEormCcLt/brute-force-manufactured-consensus-is-hiding-the-crime-of |
# A sketch of acausal trade in practice
One implication of functional decision theory is that it might be possible to coordinate across causally-disconnected regions of the universe. This possibility has been variously known as [acausal trade](https://www.lesswrong.com/tag/acausal-trade), [multiverse-wide cooperation ... | https://www.lesswrong.com/posts/5Szy3gfchXkRBEmF5/a-sketch-of-acausal-trade-in-practice |
# London rationalish meetup: Feedbackloop-first rationality
> What?
> This month's Rationalish will be hosted by Evgeny and Leo. We will be trying to implement some ideas from the blog post on Feedbackloop-first Rationality. We will alternate between trying 'Thinking Physics' problems and reflecting on the efficacy ... | https://www.lesswrong.com/events/HGFEERE7bmx25hSAz/london-rationalish-meetup-feedbackloop-first-rationality |
# Noticing Panic
In [Competent Elites](https://www.lesswrong.com/posts/CKpByWmsZ8WmpHtYa/competent-elites) [@Eliezer Yudkowsky](https://www.lesswrong.com/users/eliezer_yudkowsky?mention=user) discusses "executive nature," which he describes as the ability to "function without recourse," to make decisions without some ... | https://www.lesswrong.com/posts/Kefi44Q7fAuCQruGh/noticing-panic |
# How has internalising a post-AGI world affected your current choices?
You're probably in one of the three groups below.
Group 1: You don't think AGI is coming anytime soon, so it is business as usual for you.
Group 2: You do think it is coming soon, but you haven't made any life changes beyond career choice and do... | https://www.lesswrong.com/posts/QrdiXYeq4ooWGijMj/how-has-internalising-a-post-agi-world-affected-your-current |
# Safe Stasis Fallacy
A critical failure mode in many discussions of technological risk is the assumption that maintaining status quo for technology would lead to maintaining the status quo for society. Lewis Anslow [suggests](https://twitter.com/LouisAnslow/status/1693014091638415745) that this "propensity to treat t... | https://www.lesswrong.com/posts/Kb4mbCoGscAkvtqmM/safe-stasis-fallacy |
# AI alignment as a translation problem
Yet another way to think about the alignment problem
----------------------------------------------------
Consider two learning agents (humans or AIs) that have made different measurements of some system and have different interests (concerns) regarding how the system should be... | https://www.lesswrong.com/posts/rZWNxrzuHyKK2pE65/ai-alignment-as-a-translation-problem |
# Implementing activation steering
*Produced as part of the SERI ML Alignment Theory Scholars Program - Autumn 2023 Cohort and while being an affiliate at PIBBSS in 2024. A thank you to* [*@Jayjay*](https://www.lesswrong.com/users/jay-schreiber?mention=user) *and* [*@fela*](https://www.lesswrong.com/users/fela?mention... | https://www.lesswrong.com/posts/ndyngghzFY388Dnew/implementing-activation-steering |
# Value learning in the absence of ground truth
*Epistemic Status: My best guess.*
**Summary**
-----------
The problem of aligning AI is often phrased as the problem of aligning AI with human values, where “values” refers roughly to the normative considerations one leans upon in making decisions. Understood in this ... | https://www.lesswrong.com/posts/B3JAHCTcYJNgcqspH/value-learning-in-the-absence-of-ground-truth |
# Things You’re Allowed to Do: University Edition
This post is not titled “Things You Should Do,” because these aren’t (necessarily) things you should do. Many people should not do many of the items on this list, and some of the items are exclusive, contradictory, or [downright the reverse of what you should do](https... | https://www.lesswrong.com/posts/rPdxpGgowvhDgHyqD/things-you-re-allowed-to-do-university-edition |
# Toy models of AI control for concentrated catastrophe prevention
In this post, we analyze some simple games that model the problem of ensuring AIs never succeed at taking a single catastrophic action (which we'll refer to as "escaping") during deployment. This exercise highlights some of the difficulties associated ... | https://www.lesswrong.com/posts/MDeGts4Aw9DktCkXw/toy-models-of-ai-control-for-concentrated-catastrophe |
# Fluent dreaming for language models (AI interpretability method)
This is a cross-post for our paper on fluent dreaming for language models. [(arXiv](https://arxiv.org/abs/2402.01702) [link](https://arxiv.org/pdf/2402.01702.pdf)[.)](https://arxiv.org/abs/2402.01702) Dreaming, aka "feature visualization," is a interpr... | https://www.lesswrong.com/posts/yrP2H8beNoXghg2Gp/fluent-dreaming-for-language-models-ai-interpretability |
# Are most personality disorders really trust disorders?
Unlike other social species, we trust selectively; we choose who we cooperate with. We have exerted evolutionary pressure on each other to to judge well who is worthy of our trust, as well as evolutionary pressure to be (selectively) trustworthy. For more detail... | https://www.lesswrong.com/posts/NuqLQwDcsMSKMnhpZ/are-most-personality-disorders-really-trust-disorders |
# On the Debate Between Jezos and Leahy
Previously: [Based Beff Jezos and the Accelerationists](https://thezvi.substack.com/p/based-beff-jezos-and-the-accelerationists)
Based Beff Jezos, the founder of effective accelerationism, delivered on his previous pledge, and did indeed debate what is to be done to navigate in... | https://www.lesswrong.com/posts/xjoKqevCRnhXzHRLT/on-the-debate-between-jezos-and-leahy |
# Preventing model exfiltration with upload limits
At some point in the future, AI developers will need to ensure that when they train sufficiently capable models, the weights of these models do not leave the developer’s control. Ensuring that weights are not exfiltrated seems crucial for preventing threat models rela... | https://www.lesswrong.com/posts/rf66R4YsrCHgWx9RG/preventing-model-exfiltration-with-upload-limits |
# How can I efficiently read all the Dath Ilan worldbuilding?
I liked the April Fools Confession post, I liked the AMA, and I liked that long passage somewhere in planecrash about how they reacted to the discovery of the AI problem. But I wasn't really hooked by the glowfic, and only made it 30 pages in.
**Is there a... | https://www.lesswrong.com/posts/y7hn5fw3G4AvBgrqt/how-can-i-efficiently-read-all-the-dath-ilan-worldbuilding |
# What does davidad want from «boundaries»?
As the [Conceptual Boundaries Workshop](https://www.lesswrong.com/posts/tLb86DhrTYgkXw5Hf/apply-to-the-conceptual-boundaries-workshop-for-ai-safety) ([website](https://formalizingboundaries.ai/)) is coming up, and now that we're also planning Mathematical Boundaries Workshop... | https://www.lesswrong.com/posts/HnWiSwyxYuyYDctJm/what-does-davidad-want-from-boundaries |
# Arrogance and People Pleasing
[
Context
-------
**The first version of this document ... | https://www.lesswrong.com/posts/sTiKDfgFBvYyZYuiE/my-guess-at-conjecture-s-vision-triggering-a-narrative |
# How to train your own "Sleeper Agents"
This post is a guide on how to replicate our "[Sleeper Agents](https://www.alignmentforum.org/posts/ZAsJv7xijKTfZkMtr/sleeper-agents-training-deceptive-llms-that-persist-through)" work. I think it would be pretty valuable for someone to replicate our work on non-Anthropic model... | https://www.lesswrong.com/posts/M8kpzm42uHytnyYyP/how-to-train-your-own-sleeper-agents |
# Full Driving Engagement Optional
Yesterday the Secretary of Transportation [wrote](https://twitter.com/SecretaryPete/status/1754559027688685941):
> ALL advanced driver assistance systems available today require the human driver to be in control and fully engaged in the driving task at all times.
While the driver w... | https://www.lesswrong.com/posts/MW93TeyayiCiXpE4t/full-driving-engagement-optional |
# story-based decision-making
A few times, I've talked to an executive manager or early-stage investor, and this happened:
> me: Here's the main plan. Now, we think the odds are good, but the most likely failure point is *here*. If necessary, we have an alternative plan for that part, which goes as follows...
> them... | https://www.lesswrong.com/posts/hFDkSXd4cJ7G5SKEE/story-based-decision-making |
# Why I think it's net harmful to do technical safety research at AGI labs
IMO it is harmful on expectation for a technical safety researcher to work at DeepMind, OpenAI or Anthropic.
Four reasons:
1. **Interactive complexity.** The intractability of catching up – by trying to invent general methods for AI corp... | https://www.lesswrong.com/posts/Foh24Ya82bhpGRWME/why-i-think-it-s-net-harmful-to-do-technical-safety-research |
# How to deal with the sense of demotivation that comes from thinking about determinism?
What I know about science and philosophy suggests that **determinism** is probably true - i.e., everything that will happen throughout the lifetime of the universe was predetermined at the beginning of the universe.
When I think ... | https://www.lesswrong.com/posts/FKKY7zhyeWfqfegyS/how-to-deal-with-the-sense-of-demotivation-that-comes-from |
# The Math of Suspicious Coincidences
I've been doing a bit of [thinking about covid-19](https://www.lesswrong.com/posts/28hnPFiAoMkJssmf3/most-experts-believe-covid-19-was-probably-not-a-lab-leak) origins and how to formalize the circumstantial evidence that covid-19 leaked from WIV. I am beginning to suspect that a ... | https://www.lesswrong.com/posts/Rof3ctjMMWxpaRj5Z/the-math-of-suspicious-coincidences |
# Training of superintelligence is secretly adversarial
TL;DR it is possible to model the training of ASI[^ky0xyn8z2ml] as an adversarial process because:
1. We train safe ASI on the foundation of security assumptions about the learning process, training data, inductive biases, etc.
2. We aim to train a superintell... | https://www.lesswrong.com/posts/Lt6cZyzjNyNbTHtoi/training-of-superintelligence-is-secretly-adversarial |
# What's this 3rd secret directive of evolution called? (survive & spread & ___)
Is a species' abundance determined solely by its ability to survive and spread? Almost but not quite (I claim). There is a third requirement: it must come to exist in the first place! Furthermore, spontaneous formation alone can suffice w... | https://www.lesswrong.com/posts/c9wiSGWyX6MGqrQsX/what-s-this-3rd-secret-directive-of-evolution-called-survive |
# Reading writing advice doesn't make writing easier
**Status**: [Writing to think](https://www.lesswrong.com/posts/SKQmvZMTY6zQf3Ktc/writing-to-think) out loud, [Cunningham's Law](https://en.wikipedia.org/wiki/Ward_Cunningham#:~:text=%22Cunningham's%20Law%22,-For%20the%20mathematical&text=Cunningham%20is%20credited%2... | https://www.lesswrong.com/posts/N7bq5geFDSxFoF4vZ/reading-writing-advice-doesn-t-make-writing-easier |
# More Hyphenation
> “MAN EATING PIRANHA MISTAKENLY SOLD AS PET FISH” — example news headline from Steven Pinker’s *The Sense of Style*
The rule is that you use hyphens for compound modifiers like the ones in natural-language processing, high-impact opportunities, cost-effectiveness measures, high-status employers, a... | https://www.lesswrong.com/posts/hQLkDoWuJRJyZpzmJ/more-hyphenation |
# Debating with More Persuasive LLMs Leads to More Truthful Answers
We've just completed a bunch of empirical work on LLM debate, and we're excited to share the results. If the title of this post is at all interesting to you, we recommend heading straight to the paper. There are a lot of interesting results that are h... | https://www.lesswrong.com/posts/2ccpY2iBY57JNKdsP/debating-with-more-persuasive-llms-leads-to-more-truthful |
# Nitric oxide for covid and other viral infections
Epistemic status: I spent about 5 hours looking into this, and the next day developed covid myself. I did a bit more research plus all of the writing while sick. So in addition to my normal warning that I have no medical credentials, you should keep in mind that thi... | https://www.lesswrong.com/posts/JSXmW6FDd6W5FKoN9/nitric-oxide-for-covid-and-other-viral-infections |
# Conditional prediction markets are evidential, not causal
Quick note about a thing I didn't properly realize until recently. I don't know how important it is in practice.
tl;dr: Conditional prediction markets tell you "in worlds where *thing* happens, does *other-thing* happen?" They don't tell you "if I make *thin... | https://www.lesswrong.com/posts/vrEA6taJZtSoQbyPA/conditional-prediction-markets-are-evidential-not-causal |
# A Chess-GPT Linear Emergent World Representation
**A Chess-GPT Linear Emergent World Representation**
----------------------------------------------------
### **Introduction**
Among the many recent developments in ML, there were two I found interesting and wanted to dig into further. The first was gpt-3.5-turbo-in... | https://www.lesswrong.com/posts/yzGDwpRBx6TEcdeA5/a-chess-gpt-linear-emergent-world-representation |
# Believing In
*“In America, we believe in driving on the right hand side of the road.”*
*Tl;dr: Beliefs are like bets (on outcomes the belief doesn’t affect). “Believing in”s are more like kickstarters (for outcomes the believing-in does affect).*
*Epistemic status: New model; could use critique.*
In one early CFA... | https://www.lesswrong.com/posts/duvzdffTzL3dWJcxn/believing-in-1 |
# AI #50: The Most Dangerous Thing
In a week with [two](https://thezvi.substack.com/p/on-dwarkeshs-3rd-podcast-with-tyler) [podcasts](https://thezvi.substack.com/p/on-the-debate-between-jezos-and-leahy) I covered extensively, I was happy that there was little other news.
That is, until right before press time, when G... | https://www.lesswrong.com/posts/Si4fRH2hGGa6HsQbu/ai-50-the-most-dangerous-thing |
# Predicting Alignment Award Winners Using ChatGPT 4
This experiment explores the *Enhancing Selection* branch of the [Collective Human Intelligence](https://www.lesswrong.com/posts/CCpoqgHCCbrktCAwG/united-we-align-harnessing-collective-human-intelligence-for) (CHI) research agenda. The hypothesis under evaluation wa... | https://www.lesswrong.com/posts/68FpKvdXgzyHmxmRt/predicting-alignment-award-winners-using-chatgpt-4 |
# Updatelessness doesn't solve most problems
In some discussions (especially about [acausal trade](https://www.lesswrong.com/tag/acausal-trade) and [multi-polar conflict](https://longtermrisk.org/research-agenda)), I’ve heard the motto “X will/won’t be a problem because superintelligences will just be Updateless”. Her... | https://www.lesswrong.com/posts/g8HHKaWENEbqh2mgK/updatelessness-doesn-t-solve-most-problems-1 |
# aintelope project update
Since the announcement of the [Brain-like AGI project "aintelope"](https://www.lesswrong.com/posts/c2tEfqEMi6jcJ4kdg/brain-like-agi-project-aintelope) in 2022, we have been busy working continuously on the project. The project was entirely based on volunteer work from the start until Septemb... | https://www.lesswrong.com/posts/XKDAQEAWsMfkLfi7b/aintelope-project-update |
# A review of "Don’t forget the boundary problem..."
Introduction
------------
This is a review of ["Don't forget the boundary problem! How EM field topology can address the overlooked cousin to the binding problem for consciousness"](https://www.frontiersin.org/articles/10.3389/fnhum.2023.1233119/full), by Andrés Gó... | https://www.lesswrong.com/posts/kd37DZftKLDguqtKr/a-review-of-don-t-forget-the-boundary-problem |
# How do health systems work in adequate worlds?
I'm confused about Health Systems in general, the same way I was confused about the Education System until I read "The case against education". I don't have a good model of what are the different agents interacting in this system, nor the differences among different cou... | https://www.lesswrong.com/posts/3ag9MjaS387FNFNWM/how-do-health-systems-work-in-adequate-worlds |
# How do high-trust societies form?
First, it's clear to me they're not just myths bolstered by propaganda, and they really *are* qualitatively different from their low-trust counterparts, which I understand, and consider to be "natural," requiring no explanation.
I think it's a different concept than the WEIRD (West... | https://www.lesswrong.com/posts/REZ8wrMukQRB6djz9/how-do-high-trust-societies-form |
# Running the Numbers on a Heat Pump
I recently looked into installing a heat pump mini-split system to heat and cool our house. While it's expensive to install, I had thought a combination of it being nicer than what we currently have and cheaper to operate would make it worth it. Unfortunately, after running the num... | https://www.lesswrong.com/posts/fDsTKgXq5Ty85d4G9/running-the-numbers-on-a-heat-pump |
# Biden-Harris Administration Announces First-Ever Consortium Dedicated to AI Safety
"Today, U.S. Secretary of Commerce Gina Raimondo announced the creation of the [U.S. AI Safety Institute](https://www.nist.gov/artificial-intelligence/artificial-intelligence-safety-institute) Consortium (AISIC), which will unite AI c... | https://www.lesswrong.com/posts/ghBoGSw6iByLkHj2e/biden-harris-administration-announces-first-ever-consortium |
# Transfer learning and generalization-qua-capability in Babbage and Davinci (or, why division is better than Spanish)
Tl/Dr:
* Generalisation is a capability in its own right, and in our experiments it scales less quickly than specific capabilities.
* Our experiments show that on a range of tasks, finetuning Ba... | https://www.lesswrong.com/posts/qCecZwptWuAEZN3Td/transfer-learning-and-generalization-qua-capability-in |
# Skills I'd like my collaborators to have
*I originally wrote this in 2021, when Lightcone was exploring a vision of building "a rationalist campus" or maybe "a rationalist society."*
*Right now, neither I, nor Lightcone as a whole, are focused on this in the same way. You \*might\* make a guild that taught these sk... | https://www.lesswrong.com/posts/JNZhsHDvyECcbp7bY/skills-i-d-like-my-collaborators-to-have |
# One True Love
We have long been waiting for a version of this story, where someone hacks together the technology to use Generative AI to work the full stack of the dating apps on their behalf, ultimately finding their One True Love.
Or at least, we would, if it turned out he is Not Making This Up.
Fun question: Gi... | https://www.lesswrong.com/posts/QxAFoEdtsmK783jzM/one-true-love |
# Constituency-sized AI congress?
I just had the idea for a constituency-sized AI congress. Each member of the constituency would have their personal debate agent trained on their preferences and values. The agents would debate tirelessly at superhuman speed to develop proposals which represented the best available wi... | https://www.lesswrong.com/posts/QuuuxHwvHiEwcjaDd/constituency-sized-ai-congress |
# What’s ChatGPT’s Favorite Ice Cream Flavor? An Investigation Into Synthetic Respondents
Introduction
============
Welcome to a deep dive into the fascinating world of artificial intelligence (AI), where we are comparing synthetic responses from advanced large language models (LLMs) with real human survey responses ... | https://www.lesswrong.com/posts/2hsrPrsJNgNxBSFs4/what-s-chatgpt-s-favorite-ice-cream-flavor-an-investigation |
# Close the Gates to an Inhuman Future: How and why we should choose to not develop superhuman general-purpose artificial intelligence
This is a revised version of a paper I put out late last year (including a new table about capabilities lacking in current AI and how we might get them.)
Interested in reactions and p... | https://www.lesswrong.com/posts/3Z8bpv3xepLmyAaQp/close-the-gates-to-an-inhuman-future-how-and-why-we-should |
# Scenario planning for AI x-risk
*This post is part of a series by* [*Convergence Analysis*](https://www.convergenceanalysis.org/)*. In it, I’ll motivate and review some methods for applying scenario planning methods to AI x-risk strategy. Feedback and discussion are welcome.*
Summary
=======
AI is a particularly d... | https://www.lesswrong.com/posts/foqm3fKfDuRHtwjuH/scenario-planning-for-ai-x-risk |
# Dreams of AI alignment: The danger of suggestive names
Let's not forget the old, well-read post: [Dreams of AI Design](https://www.lesswrong.com/posts/p7ftQ6acRkgo6hqHb/dreams-of-ai-design). In that essay, Eliezer correctly points out errors in imputing meaning to nonsense by using suggestive names to describe the n... | https://www.lesswrong.com/posts/yxWbbe9XcgLFCrwiL/dreams-of-ai-alignment-the-danger-of-suggestive-names |
# A Strange ACH Corner Case
It turns out ACH transactions can fail because your bank has too many people out for the holidays. Which isn't great if you're trying to get your donation in before the end of the year!
* * *
Juila and I were a bit late in deciding where we wanted to donate this year. We intended to sort ... | https://www.lesswrong.com/posts/HWcb36GDePprFB4pJ/a-strange-ach-corner-case |
# The lattice of partial updatelessness
*In this post, I explore the fundamental conceptual structure of updateless commitments, by constructing a formalism for them.*
I say your decision procedure *updates* on piece of information $p$ if $p$ is logically/computationally upstream of your decisions/actions. Or equival... | https://www.lesswrong.com/posts/Jn43ZrzKZ86pd5HPM/the-lattice-of-partial-updatelessness |
# Sam Altman’s Chip Ambitions Undercut OpenAI’s Safety Strategy
*If you enjoy this, please consider subscribing to my* [*Substack*](https://garrisonlovely.substack.com/)*.*
Sam Altman has said he thinks that developing artificial general intelligence (AGI) could lead to human extinction, but OpenAI is trying to build... | https://www.lesswrong.com/posts/pEAHbJRiwnXCjb4A7/sam-altman-s-chip-ambitions-undercut-openai-s-safety |
# And All the Shoggoths Merely Players
_[Setting: a suburban house. The interior of the house takes up most of the stage; on the audience's right, we see a wall in cross-section, and a front porch. **Simplicia** enters stage left and rings the doorbell.]_
**Doomimir**: _[opening the door]_ Well? What do you want?
**... | https://www.lesswrong.com/posts/8yCXeafJo67tYe5L4/and-all-the-shoggoths-merely-players |
# Experimenting With Footboard Piezos
In playing for contra dances it's rare to have a drummer. You mostly see them at big events where the finances work out for a fourth or fifth musician: in trio, or especially a duo, you generally can't allocate anyone just to drums. What you do see a lot of, however, is a musician... | https://www.lesswrong.com/posts/YkbueixbmeWbuENNJ/experimenting-with-footboard-piezos |
# What's the theory of impact for activation vectors?
[Activation vectors](https://www.lesswrong.com/posts/5spBue2z2tw4JuDCx/steering-gpt-2-xl-by-adding-an-activation-vector) are really, really cool, but what is the theory of impact for this work?
* Is the hope that activation vectors will allow us to actually gain... | https://www.lesswrong.com/posts/Gx74coCsm7WtAmHYC/what-s-the-theory-of-impact-for-activation-vectors |
# GPT2XL_RLLMv3 vs. BetterDAN, AI Machiavelli & Oppo Jailbreaks
*This post contains a significant amount of harmful content; please read with caution. Lastly, I want to apologize to all the AI labs and owners of language models I attacked in the process of creating this post. We have yet to solve the alignment problem... | https://www.lesswrong.com/posts/vZ5fM6FtriyyKbwi9/gpt2xl_rllmv3-vs-betterdan-ai-machiavelli-and-oppo |
# How do you actually obtain and report a likelihood function for scientific research?
I just read the article [https://arbital.com/p/likelihoods\_not\_pvalues/,](https://arbital.com/p/likelihoods_not_pvalues/,) but this left me with one big question: what is the actual sequence of actions a researcher would take to r... | https://www.lesswrong.com/posts/tgcTdaMsThQiJArGa/how-do-you-actually-obtain-and-report-a-likelihood-function |
# Carl Shulman On Dwarkesh Podcast June 2023
This is basically just an exposition of the content of Dwarkesh Patel, I thought it might be useful to have a slightly edited full version here. Full credit to Dwarkesh and Carl.
I find it easier to ingest the information this way and it is a bit quicker too. I did not fin... | https://www.lesswrong.com/posts/BdPjLDG3PBjZLd5QY/carl-shulman-on-dwarkesh-podcast-june-2023 |
# What are the deciding factors of human cognitive endurance?
Cal Newport says something to the effect that the median capacity for high-intensity focus is about four hours per day in *Deep Work.* I think this can bounce around over time for individuals – especially regarding psychological dispositions like bipolar or... | https://www.lesswrong.com/posts/iiRBTXfyfSKFdvPbu/what-are-the-deciding-factors-of-human-cognitive-endurance |
# What are the known difficulties with this alignment approach?
Assume you have a world-model that is nicely factored into spatially localized variables that contain interesting-to-you concepts. (Yes, that's a big assumption, but are there any known difficulties with the proposal if we grant this assumption?)
Pick so... | https://www.lesswrong.com/posts/n9ht4aPNpxenibnw8/what-are-the-known-difficulties-with-this-alignment-approach |
# Skepticism About DeepMind's "Grandmaster-Level" Chess Without Search
Update: Authors commented below that their unpublished results show that actually the bot is as good as described; many thanks to them.
[The paper](https://arxiv.org/abs/2402.04494)
First, addressing misconceptions that could come from the title ... | https://www.lesswrong.com/posts/PXRi9FMrJjyBcEA3r/skepticism-about-deepmind-s-grandmaster-level-chess-without |
# Thoughts on "The Offense-Defense Balance Rarely Changes"
_Personal Views Only_
Introduction
------------
I thought Maxwell Tabarrok’s _[The Offense-Defense Balance Rarely Changes](https://forum.effectivealtruism.org/posts/oHkJziKoP47PBSizr/the-offense-defense-balance-rarely-changes)_ was a good contribution to the... | https://www.lesswrong.com/posts/doBDpwEx48L54izSo/thoughts-on-the-offense-defense-balance-rarely-changes |
# On the Proposed California SB 1047
[California Senator Scott Wiener](https://twitter.com/Scott_Wiener/status/1755650108287578585) of San Francisco introduces [SB 1047](https://t.co/JWaOLP44Iu) [to regulate AI](https://www.washingtonpost.com/technology/2024/02/08/california-legislation-artificial-intelligence-regulat... | https://www.lesswrong.com/posts/oavGczwcHWZYhmifW/on-the-proposed-california-sb-1047 |
# Tort Law Can Play an Important Role in Mitigating AI Risk
Epistemic status: This is a summary of an 80-page law review article that took me ~6 months to research and write. I am highly confident about my characterization of existing U.S. tort law and its limits. This characterization is only contested by other legal... | https://www.lesswrong.com/posts/5e7TrmH7mBwqpZ6ek/tort-law-can-play-an-important-role-in-mitigating-ai-risk |
# Natural abstractions are observer-dependent: a conversation with John Wentworth
The conversation
----------------
**Martín:**
Any write-up on the thing that "[natural abstractions](https://www.alignmentforum.org/tag/natural-abstraction) depend on your goals"? As in, pressure is a useful abstraction because we car... | https://www.lesswrong.com/posts/CJjT8GMitsnKc2wgG/natural-abstractions-are-observer-dependent-a-conversation-1 |
# I played the AI box game as the Gatekeeper — and lost
Eliezer Yudkowsky's AI box game is a two-hour conversation carried out through text. One person plays the part of the "Gatekeeper" and the other is the "AI". If, at any point, the Gatekeeper types the phrase `you are out`, the AI wins. If the Gatekeeper can go t... | https://www.lesswrong.com/posts/Bnik7YrySRPoCTLFb/i-played-the-ai-box-game-as-the-gatekeeper-and-lost |
# Interpreting Quantum Mechanics in Infra-Bayesian Physicalism
*This work was inspired by a question by Vanessa Kosoy, who also contributed several of the core ideas, as well as feedback and mentorship.*
## Abstract
We outline a *computationalist* interpretation of quantum mechanics, using the framework of infra-Baye... | https://www.lesswrong.com/posts/GctPwQhrvwBpzyt9q/interpreting-quantum-mechanics-in-infra-bayesian-physicalism |
# My cover story in Jacobin on AI capitalism and the x-risk debates
> Google cofounder Larry Page [thinks](https://www.vanityfair.com/news/2023/09/artificial-intelligence-industry-future#:~:text=While%20Page%20stays,soon%20as%20possible.%E2%80%9D) superintelligent AI is “just the next step in evolution.” In fact, Page... | https://www.lesswrong.com/posts/QmdkBz2R2zoPzBzgk/my-cover-story-in-jacobin-on-ai-capitalism-and-the-x-risk |
# Where is the Town Square?
I am seeking crowdsourced wisdom.
Suppose I[^sa06s0ni0ld] want to influence public opinion on a complicated, nuanced topic.[^l4imxgu66c]
And further suppose that one of the ways I want to do that is by participating actively in public discourse: posting short-form content, reacting to oth... | https://www.lesswrong.com/posts/2KR2wtQHqEQTT7TBJ/where-is-the-town-square |
# Lsusr's Rationality Dojo
> Why aren’t there dojos that teach rationality?
>
> ―[*The Martial Art of Rationality*](https://www.lesswrong.com/posts/teaxCFgtmCQ3E9fy8/the-martial-art-of-rationality) by Eliezer Yudkowsky
For the last 6 months, I've been running a dojo that teaches rationality.
Why
===
I was at an AC... | https://www.lesswrong.com/posts/9BK5kSC88ddZ8kLmP/lsusr-s-rationality-dojo |
# Technologies and Terminology: AI isn't Software, it's... Deepware?
A few weeks ago, I (David) tried to argue that AI wasn’t software. In retrospect, I think I misstated the case by skipping some inferential steps, and based on some feedback from that article, and on an earlier version of this article, with a large a... | https://www.lesswrong.com/posts/kSq5qiafd6SqQoJWv/technologies-and-terminology-ai-isn-t-software-it-s-deepware |
# Causality is Everywhere
*Tl;dr: Everything in our world is causally connected in countless ways, which means causal effects are everywhere, and practically anything has systematic effects on almost everything else – it’s just that most of these effects are very close to (but not quite) zero. Studies often fail to me... | https://www.lesswrong.com/posts/dyHiGjdbH8dK26i2m/causality-is-everywhere |
# More on the Apple Vision Pro
Previously: [On the Apple Vision Pro](https://thezvi.substack.com/p/on-the-apple-vision-pro)
The reviews are coming in. What say the people, other than the complaining about the two to three hour battery life?
Then later I’ll get to my own thoughts after the demo.
#### Reviews and Rea... | https://www.lesswrong.com/posts/HbneNh2NJpQCgfxQA/more-on-the-apple-vision-pro |
# The "context window" analogy for human minds
Years ago I already modeled (and [wrote about](https://www.lesswrong.com/posts/esJWfrGY89QrhGQbo/four-factors-that-moderate-the-intensity-of-emotions)) that emotions are mediated by attention. You have emotions about what's on your mind (or recently on your mind) rather t... | https://www.lesswrong.com/posts/Pav6tbSwocFw6xDTf/the-context-window-analogy-for-human-minds |
# Masterpiece
*Inspired by qntm's* [Lena](https://qntm.org/lena)*. Reading* Lena *first is helpful but not necessary.*
* * *
We’re excited to announce the fourth annual MMindscaping competition! Over the last few years, interest in the art of mindscaping has continued to grow rapidly. We expect this year’s competiti... | https://www.lesswrong.com/posts/Fruv7Mmk3X5EekbgB/masterpiece |
# Useful starting code for interpretability
Want to try your hand at neural network interpretability? A very nice way to get started is to find an existing Python notebook using one or more interpretability techniques, hopefully one written with beginners in mind. In a click or two you can make a copy of it, which you... | https://www.lesswrong.com/posts/iSJrd3TE6Pd3ctyaD/useful-starting-code-for-interpretability |
# An EA used deceptive messaging to advance her project; we need mechanisms to avoid deontologically dubious plans
**TL;DR**: good-hearted EAs lack the mechanisms to not output information that can mislead people. Holly Elmore organized a protest, with the messaging centered on OpenAI changing their documents to "work... | https://www.lesswrong.com/posts/P8CbosuG84dBjEh7C/ea-community-needs-mechanisms-to-avoid-deceptive-messaging |
# Meetup In a Box: Year In Review
**Summary:** Three rounds of questions, where the meetup thinks about their past year in a few frames of reference, then discusses their insights.
**Tags:** Small, Medium, One-Off, Experiment
**Purpose:** Year in Review is something more useful than New Year's Resolutions - more lik... | https://www.lesswrong.com/posts/rSWEfsvpsggxsHiGC/meetup-in-a-box-year-in-review |
# CFAR Takeaways: Andrew Critch
I'm trying to build my own art of rationality training, and I've started talking to various CFAR instructors about their experiences – things that might be important for me to know but which hadn't been written up nicely before.
This is a quick write up of a conversation with Andrew Cr... | https://www.lesswrong.com/posts/Jash4Gbi2wpThzZ4k/cfar-takeaways-andrew-critch |
# Scale Was All We Needed, At First
*This is a hasty speculative fiction vignette of one way I expect we might get AGI by January 2025 (within about one year of writing this). Like similar works by others, I expect most of the guesses herein to turn out incorrect. However, this was still useful for expanding my imagin... | https://www.lesswrong.com/posts/xLDwCemt5qvchzgHd/scale-was-all-we-needed-at-first |
# FTX expects to return all customer money; clawbacks may go away
[New York Times (unpaywalled)](https://www.nytimes.com/2024/02/03/business/dealbook/what-happens-to-ftx-clawback-cases-if-the-company-repays-its-creditors.html?unlocked_article_code=1.VU0.QBLc.h36dKm_EgU8L&smid=url-share):
> When the cryptocurrency exc... | https://www.lesswrong.com/posts/agnARYYJurYMkZWyj/ftx-expects-to-return-all-customer-money-clawbacks-may-go |
# Requirements for a Basin of Attraction to Alignment
**TL;DR:** It has been known for over a decade that that certain agent architectures based on [Value Learning](https://www.lesswrong.com/tag/value-learning) by construction have the very desirable property of having a basin of attraction to full alignment, where if... | https://www.lesswrong.com/posts/EbGkqFNz8y93Ttuwq/requirements-for-a-basin-of-attraction-to-alignment |
# Optimizing for Agency?
It's commonly accepted that pretty much every optimization target results in death. If you optimize for paperclips, humans die. If you optimize for curing cancer, humans die.
What about optimizing for agency?
The way I visualize this is after a superintelligence takeover, and the superintell... | https://www.lesswrong.com/posts/FEGe3HPDD5WPZJAuv/optimizing-for-agency |
# What experiment settles the Gary Marcus vs Geoffrey Hinton debate?
It started with this video of Hinton taking a jab at Marcus: [https://twitter.com/tsarnick/status/1754439023551213845](https://twitter.com/tsarnick/status/1754439023551213845)
And here is Marcu's answer:
[https://garymarcus.substack.com/p/deconstru... | https://www.lesswrong.com/posts/9PfBYpAE6cLXfCLM9/what-experiment-settles-the-gary-marcus-vs-geoffrey-hinton |
#
Examples of governments doing good in house (or contracted) technical research
An open question is the degree to which governments can do good technical research, either in-house or commissioned. Some possible ways in which this could be relevant:
* Evals (although this likely could be contracted in a successful... | https://www.lesswrong.com/posts/iAxxonqPGQkQYhzEJ/examples-of-governments-doing-good-in-house-or-contracted |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.