text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Advice for interacting with busy people
Over the last year, I have turned from somebody with a lot of time and relatively few social contacts to maintain into someone quite busy and well-networked. I'm by no means close to the top of the busyness hierarchy. But, I've climbed high enough to learn some lessons that I ... | https://www.lesswrong.com/posts/v8xoqiwaYu9krmjsY/advice-for-interacting-with-busy-people |
# AI #10: Code Interpreter and Geoff Hinton
The big capabilities news this week is a new ChatGPT mode (that I do not have access to yet) called Code Interpreter. It lets you upload giant data files, analyzes them automatically, can even write papers about its findings, many are impressed.
The big discourse news is th... | https://www.lesswrong.com/posts/fbc54dEWuuugLbgHv/ai-10-code-interpreter-and-geoff-hinton |
# Clarifying and predicting AGI
*This post is a slightly-adapted summary of two twitter threads,* [*here*](https://twitter.com/RichardMCNgo/status/1643310525697105935) *and* [*here*](https://twitter.com/RichardMCNgo/status/1640568775018975232?lang=en)*.*
The t-AGI framework
-------------------
As we get closer to AG... | https://www.lesswrong.com/posts/BoA3agdkAzL6HQtQP/clarifying-and-predicting-agi |
# [Linkpost]Transformer-Based LM Surprisal Predicts Human Reading Times Best with About Two Billion Training Tokens
From the abstract:
> Recent psycholinguistic studies have drawn conflicting conclusions about the relationship between the quality of a language model and the ability of its surprisal estimates to predi... | https://www.lesswrong.com/posts/Y54nc3kkeXMLm5Wje/linkpost-transformer-based-lm-surprisal-predicts-human |
# Google "We Have No Moat, And Neither Does OpenAI"
This (purported) leak is incredibly scary if you're worried about AI risk due to AI capabilities developing too fast.
Firstly, it suggests that open-source models are improving rapidly because people are able to iterate on top of each other's improvements and try ou... | https://www.lesswrong.com/posts/A2sGgpYTbkkveoXbn/google-we-have-no-moat-and-neither-does-openai |
# Recursive Middle Manager Hell: AI Edition
Two-sentence summary
=====================
In [Recursive Middle Manager Hell](https://www.lesswrong.com/posts/pHfPvb4JMhGDr4B7n/recursive-middle-manager-hell), Raemon argues that organisations get increasingly warped (immoral/maze-like/misaligned/...) as they gain more lev... | https://www.lesswrong.com/posts/jzHeZt3xFnGTudH5M/recursive-middle-manager-hell-ai-edition |
# Who regulates the regulators? We need to go beyond the review-and-approval paradigm
**IRBs**
--------
Scott Alexander [reviews](https://astralcodexten.substack.com/p/book-review-from-oversight-to-overkill) a book about institutional review boards (IRBs), the panels that review the ethics of medical trials: *From Ov... | https://www.lesswrong.com/posts/uYFbtgAoRFWpLTfTW/who-regulates-the-regulators-we-need-to-go-beyond-the-review |
# A brief collection of Hinton's recent comments on AGI risk
Since I've seen some people doubt whether (original popularizer of the backpropagation algorithm and one of the original developers of deep learning) Geoff Hinton is *actually* concerned about AGI risk (as opposed to e.g. the NYT spinning an anti-tech agenda... | https://www.lesswrong.com/posts/bLvc7XkSSnoqSukgy/a-brief-collection-of-hinton-s-recent-comments-on-agi-risk |
# How to get good at programming
*Epistemic status: very confident*
See also: A closely related [post](https://gwern.net/unseeing) by Gwern, another related [post](https://www.lesswrong.com/posts/nEBbw2Bc2CnN2RMxy/gears-level-models-are-capital-investments) by John, and some interesting [slides](https://www.cs.dartmo... | https://www.lesswrong.com/posts/LTypqBMTSmRrrhb2v/how-to-get-good-at-programming |
# Why not use active SETI to prevent AI Doom?
Let's assume that Eliezer is right: soon we'll have an AGI that is very likely to kill us all. (personally, I think Eliezer is right).
There are several ways to reduce the risk, in particular: speeding up alignment research and slowing down capabilities research, by vario... | https://www.lesswrong.com/posts/eExFvqojsagFm4S6e/why-not-use-active-seti-to-prevent-ai-doom |
# What can we learn from Bayes about reasoning?
What are the *qualitative* lessons we can learn about logic and reasoning from Bayesian epistemology, that is, from taking Bayes' rule as a mathematical model for thought (even if it is considered a simplified formalism that we often can't implement?)
I've seen at least... | https://www.lesswrong.com/posts/PPfb4nc8ewcTtFWmJ/what-can-we-learn-from-bayes-about-reasoning |
# Kingfisher Live CD Process
Our live album is [out today](https://kingfisherband.bandcamp.com/album/kingfisher-recorded-live)! We'll have physical CDs around the end of May, and it should be on streaming services at some point. If you backed the Kickstarter and haven't heard from us about your rewards, let me know!
... | https://www.lesswrong.com/posts/uQTELjcgzpKAZq58b/kingfisher-live-cd-process |
# Linkpost for Accursed Farms Discussion / debate with AI expert Eliezer Yudkowsky
Eliezer has been on a podcast tour of late. This appears to be the latest one on YouTube.
I feel like this one didn't go so well. It's also over two hours, so maybe don't watch this one if you have better things to do, but for those th... | https://www.lesswrong.com/posts/t7qwNyHfp6Jc4mdHJ/linkpost-for-accursed-farms-discussion-debate-with-ai-expert |
# CHAT Diplomacy: LLMs and National Security
> "The view, expressed by almost all competent atomic scientists, that there was no "secret" about how to build an atomic bomb was thus not only rejected by influential people in the U.S. political establishment, but was regarded as a treasonous plot."
*Robert Oppenheimer... | https://www.lesswrong.com/posts/cneXDpqQnPncwnXMo/chat-diplomacy-llms-and-national-security |
# What Jason has been reading, May 2023: “Protopia,” complex systems, Daedalus vs. Icarus, and more
This is a monthly feature. As usual, I’ve omitted recent blog posts and such, which you can find in my [links digests](https://rootsofprogress.org/writing#twitter-digest).
**John Gall,** [***The Systems Bible***](https... | https://www.lesswrong.com/posts/xSMpgESsRznsgumFq/what-jason-has-been-reading-may-2023-protopia-complex |
# A smart enough LLM might be deadly simply if you run it for long enough
**TL;DR**: I introduce a new potential threat: a smart enough LLM, even if it’s myopic, not fine-tuned with any sort of RL, “non-agentic”, and was prompt-engineered into imitating aligned and helpful scientists who are doing their best to hel... | https://www.lesswrong.com/posts/PfoJLZg2e4xXt2Q6a/a-smart-enough-llm-might-be-deadly-simply-if-you-run-it-for |
# Orthogonal's Formal-Goal Alignment theory of change
We recently announced [Orthogonal, an agent foundations alignment research organization. In this post, I give a thorough explanation of the formal-goal alignment framework, the motivation behind it, and the theory of change it fits in.
The overall shape of what w... | https://www.lesswrong.com/posts/4XcADCLDDguyej2N7/orthogonal-s-formal-goal-alignment-theory-of-change |
# Explaining “Hell is Game Theory Folk Theorems”
I, along with many commenters, found the explanation in [Hell is Game Theory Folk Theorems](https://www.lesswrong.com/posts/d2HvpKWQ2XGNsHr8s/hell-is-game-theory-folk-theorems) somewhat unclear. I am re-explaining some of the ideas from that post here. Thanks to [jessic... | https://www.lesswrong.com/posts/FE4R38FQtnTNzeGoz/explaining-hell-is-game-theory-folk-theorems |
# Naturalist Collection
There’s a kind of rhythm to the four phases of naturalism. To me, they feel a bit like oscillation between zooming out and zooming in, or like inhaling and exhaling. The first phase ([Locating Fulcrum Experiences](https://www.lesswrong.com/s/zLib3j2Fdnnx3aP3F/p/Gs6oMWEFsL4gGocBf)) has a broad e... | https://www.lesswrong.com/posts/7oAENKMsud2qQBXDj/naturalist-collection |
# You don't need to be a genius to be in AI safety research
The aim of this article is to share my experiences within the AI safety community to those in research and operations in the field, with the goal of creating a more inclusive, supportive, and positive environment. The intention of this article is to allow tho... | https://www.lesswrong.com/posts/WZZnKP9rXg9sSx9du/you-don-t-need-to-be-a-genius-to-be-in-ai-safety-research |
# My preferred framings for reward misspecification and goal misgeneralisation
TL;DR
=====
* I was feeling quite dissatisfied with a bunch of categories like “inner and outer misalignment” and “reward misspecification and goal misgeneralisation”
* I think most of this dissatisfaction stems from them not being abl... | https://www.lesswrong.com/posts/uxumdiuip7oWCAgaA/my-preferred-framings-for-reward-misspecification-and-goal |
# Estimating Norovirus Prevalence
How many people currently have Norovirus? What about in June 2021, to pick a historical month? How could we get an estimate?
The CDC says there are [19-21M cases annually](https://www.cdc.gov/norovirus/trends-outbreaks/burden-US.html) but Norovirus is strongly seasonal; you're much m... | https://www.lesswrong.com/posts/XLqQ4jYoNZ5xuxTFG/estimating-norovirus-prevalence |
# The Broader Fossil Fuel Community
In his blog post, Why Not Slow AI Progress?,[^xe4cnk37kse] Scott Alexander notes that AI Safety researchers and AI Capabilities researchers seem to be part of the same community. This is kind of weird: fossil fuel companies and environmentalists seem to be much more socially distinc... | https://www.lesswrong.com/posts/3Dqdw4gf8FK4vhyAK/the-broader-fossil-fuel-community |
# Long Covid Risks: 2023 Update
Back in 2021 I wrote a [post](https://acesounderglass.com/2021/08/30/long-covid-is-not-necessarily-your-biggest-problem/) estimating the risk of long covid. Recently a client hired me to do an update, focusing on changes induced by Paxlovid and vaccination. This was a <5h project and th... | https://www.lesswrong.com/posts/2FWiBko2WkD8Ldcz3/long-covid-risks-2023-update |
# How much do you believe your results?
*Thanks to Drake Thomas for feedback.*
**I.**
Here’s a fun scatter plot. It has two thousand points, which I generated as follows: first, I drew two thousand x-values from a normal distribution with mean 0 and standard deviation 1. Then, I chose the y-value of each point by ta... | https://www.lesswrong.com/posts/nnDTgmzRrzDMiPF9B/how-much-do-you-believe-your-results |
# Time and Energy Costs to Erase a Bit
[Landauer](https://www.informationphilosopher.com/solutions/scientists/landauer/Landauer-1961.pdf) derived a limit on the amount of energy dissipated in order to erase a bit. This limit is $kT\log 2$, where $k$ is Boltzmann's constant and $T$ is the temperature of the environment... | https://www.lesswrong.com/posts/PyChB935jjtmL5fbo/time-and-energy-costs-to-erase-a-bit |
# On the Loebner Silver Prize (a Turing test)
The Metaculus question "When will the first weakly general AI system be devised, tested, and publicly announced?" has four resolution criteria, and in my opinion, reliably passing a Loebner-Silver-prize-equivalent Turing test is the hardest, since it is the only one that i... | https://www.lesswrong.com/posts/mNR3tQDCxWL9XWv5u/on-the-loebner-silver-prize-a-turing-test |
# Corrigibility, Much more detail than anyone wants to Read
Corrigibility has been variously defined, for example [here](https://intelligence.org/files/Corrigibility.pdf) as:
> We say that an agent is “corrigible” if it tolerates or
> assists many forms of outside correction, including at
> least the following: (... | https://www.lesswrong.com/posts/v3jocJRScqkBGtwvf/corrigibility-much-more-detail-than-anyone-wants-to-read |
# TED talk by Eliezer Yudkowsky: Unleashing the Power of Artificial Intelligence
**Note:** The video is no longer available. It has been set to private. It'll eventually be released on the main TED channel.
This is a TED talk published early by a random TEDx channel.
Tweet by EY: [https://twitter.com/ESYudkowsky/sta... | https://www.lesswrong.com/posts/SWBRYeqTYDKJbbsfr/ted-talk-by-eliezer-yudkowsky-unleashing-the-power-of |
# Properties of Good Textbooks
Heuristics for choosing/writing good textbooks (see also
[here](https://docs.google.com/document/d/e/2PACX-1vTkqKg5IxCmPbw7JqnAWxoypaYNFH3XJd4UgYw4PufP09zzzW6j3v-CYXZkpD83sVrzygvg7gLbjM_Q/pub)):
* Has exercises
* Exercises are interspersed in the text, not in large chunks (better at... | https://www.lesswrong.com/posts/LockA5xWWn39fXQ5i/properties-of-good-textbooks |
# Violin Supports
Violins are a bit of a strange design. Classically you hold them up by pinching them between your chin and shoulder (restricting head motion), though in more casual contexts people sometimes use their left arm (restricting hand motion). Neither of these are great! Why can't the violin stay up on its ... | https://www.lesswrong.com/posts/QjGgXpLpugoMni7yv/violin-supports |
# An anthropomorphic AI dilemma
*[Metadata: crossposted from [https://tsvibt.blogspot.com/2023/01/the-anthropomorphic-ai-dilemma.html](https://tsvibt.blogspot.com/2023/01/the-anthropomorphic-ai-dilemma.html). First completed January 21, 2023.]*
Either generally-human-level AI will work internally like humans work in... | https://www.lesswrong.com/posts/L2h9nAtPqEFK6atSJ/an-anthropomorphic-ai-dilemma |
# Let's look for coherence theorems
In [There are no coherence theorems](https://www.lesswrong.com/posts/yCuzmCsE86BTu9PfA/there-are-no-coherence-theorems) EJT defines a "coherence theorem" as a theorem that proves maximizing expected utility is the only way to play a game that is not dominated. _Ie, any strategy that... | https://www.lesswrong.com/posts/wKk7Jixwvjzzt6Zyq/let-s-look-for-coherence-theorems |
# Statistical models & the irrelevance of rare exceptions
*“I don’t care about this instance - I don’t care about any instances! Life is too short to care about anything but the general case!”*
Yes, the general case is drawn from instances, I’m saying that we shouldn't get caught up in the details unless they really ... | https://www.lesswrong.com/posts/9KXuuFz9h9JpriWQ4/statistical-models-and-the-irrelevance-of-rare-exceptions |
# How to have Polygenically Screened Children
Polygenic screening is a method for modifying the traits of future children via embryo selection. If that sounds like gobbledygook, then think of it a bit like choosing stats for your baby.
That may sound amazing. It may sound like science fiction. It may even sound horri... | https://www.lesswrong.com/posts/yT22RcWrxZcXyGjsA/how-to-have-polygenically-screened-children |
# Understanding mesa-optimization using toy models
Overview
--------
* Solving the problem of mesa-optimization would [probably be easier if we understood how models do search internally](https://www.lesswrong.com/posts/FDjTgDcGPc7B98AES/searching-for-search-4)
* We are training GPT-type models on the toy task of... | https://www.lesswrong.com/posts/svuawhk64eF8fGv6c/understanding-mesa-optimization-using-toy-models |
# Categories of Arguing Style : Why being good among rationalists isn't enough to argue with everyone
*tl;dr : We overgeneralize the way we personally come up with concepts. People, notably outside of LW and inside of Academia, argue by referring themselves to diverse and starkly different ways of coming up with conce... | https://www.lesswrong.com/posts/5sZCTTecuwi2nT5oi/categories-of-arguing-style-why-being-good-among |
# What's wrong with being dumb?
I like to make fun of my dog sometimes for being dumb. Affectionately. I'll say things like "That's not for you dumb dumbs!" when he tries to eat food that will clearly end up giving him a stomach ache and making him nauseous.
A few years ago I did this in front of my girlfriend's mom.... | https://www.lesswrong.com/posts/yiPjHF7m75LKAkRbA/what-s-wrong-with-being-dumb |
# An artificially structured argument for expecting AGI ruin
Philosopher David Chalmers [asked](https://twitter.com/robbensinger/status/1647340328998039552):
> \[I\]s there a canonical source for "the argument for AGI ruin" somewhere, preferably laid out as an explicit argument with premises and a conclusion?
Unsurp... | https://www.lesswrong.com/posts/QzkTfj4HGpLEdNjXX/an-artificially-structured-argument-for-expecting-agi-ruin |
# LLM cognition is probably not human-like
This post is a collection of thought experiments and observations, meant to distill and explain some claims and observations about GPTs and LLMs made by myself and others.
Stated in its most general form, the claim is: the ability of a system or process to predict something ... | https://www.lesswrong.com/posts/rjghymycfrMY2aRk5/llm-cognition-is-probably-not-human-like |
# Is EDT correct? Does "EDT" == "logical EDT" == "logical CDT"?
**Main question:** Does EDT / "logical EDT"[^pvlsreyzbe] make any wrong decisions, once the "tickle defense" is accounted for? And is there any real distinction between EDT, "logical EDT", and "logical CDT"?
Relatedly, I'm having trouble constructing ... | https://www.lesswrong.com/posts/kowKm25hFxRquEyim/is-edt-correct-does-edt-logical-edt-logical-cdt |
# A technical note on bilinear layers for interpretability
Summary
=======
In this short theoretical note (now on [Arxiv](https://arxiv.org/abs/2305.03452)) I examine bilinear layers, which are MLP layers that take the form
$MLP_{Bilinear}(x) = (W_1 x) \odot (W_2 x)$.
When used in language models, they perform bett... | https://www.lesswrong.com/posts/qm6bhbJmft2LJNzKH/a-technical-note-on-bilinear-layers-for-interpretability |
# What does it take to ban a thing?
Epistemic status: I am not an expert. I just took several things which people banned (child labor, chemical weapons, ozone-depleting substances) and for each just searched for the first article which seriously engages with the question "how did we succeed to ban it?", read it once, ... | https://www.lesswrong.com/posts/cvpSNy32rgNdqqQ9r/what-does-it-take-to-ban-a-thing |
# Dance Profit Sharing
With most endeavors that have a financial component there's a spectrum of approaches that differ on how they assign risk. At the "low risk" end you have agreements to pay a specific amount ("we pay $100/person"), while at the "high risk" end it depends on how things go go ("everyone gets a share... | https://www.lesswrong.com/posts/QacKfd7Efd53AzwBM/dance-profit-sharing |
# Housing and Transit Roundup #4
It’s time for another housing roundup, so I can have a place to address the recent discussions about the local impact of housing construction on housing costs.
#### Does Increasing Local Housing Supply Decrease Local Housing Prices?
[Scott Alexander says](https://astralcodexten.subst... | https://www.lesswrong.com/posts/HKjB5hteptxNGWwst/housing-and-transit-roundup-4 |
# Thriving in the Weird Times: Preparing for the 100X Economy
*Epistemic status: Confident in the general concept, but not in the specific details.*
The world is changing fast. As soon as 2026[^p53t5bf6ljf], we may experience an exponential increase in personal productivity, intelligence, and overall capacity to infl... | https://www.lesswrong.com/posts/x6eGsJGNtH3Sagxwu/thriving-in-the-weird-times-preparing-for-the-100x-economy |
# Inference Speed is Not Unbounded
*\[The intro of this post has been lightly edited since it was first posted to address some comments. I have also changed the title to better reflect my core argument. My apologies if that is not considered good form.\]*
This post will be a summary of some of my ideas on what intell... | https://www.lesswrong.com/posts/Qvec2Qfm5H4WfoS9t/inference-speed-is-not-unbounded |
# Product Endorsement: Apollo Neuro
Short version: This [$310 vibrating bracelet](https://share.apolloneuro.com/x/rdJmKX) dramatically improved my sleep and moderately improved my emotional regulation. The return policy is pretty liberal so if this seems at all appealing I recommend trying it, or one of the cheaper al... | https://www.lesswrong.com/posts/xXW98AmifmuB2xBiY/product-endorsement-apollo-neuro |
# LeCun’s “A Path Towards Autonomous Machine Intelligence” has an unsolved technical alignment problem
Summary
=======
* This post is about the paper [**A Path Towards Autonomous Machine Intelligence**](https://openreview.net/pdf?id=BZ5a1r-kVsf) **(APTAMI) by Yann LeCun**. It’s a high-level sketch of an AI architec... | https://www.lesswrong.com/posts/C5guLAx7ieQoowv3d/lecun-s-a-path-towards-autonomous-machine-intelligence-has-1 |
# Annotated reply to Bengio's "AI Scientists: Safe and Useful AI?"
The AI safety proposal was published by Yoshua Bengio on May 7, 2023.
Below I quote the salient sections of the proposal and comment on them in turn.
Main thesis: safe AI scientists
-------------------------------
> \[…\] I would like to share my th... | https://www.lesswrong.com/posts/kGrwufqxfsyuaMREy/annotated-reply-to-bengio-s-ai-scientists-safe-and-useful-ai |
# Predictable updating about AI risk
(Cross-posted from [my website](https://joecarlsmith.com/2023/05/08/predictable-updating-about-ai-risk). Podcast version [here](https://www.buzzsprout.com/2034731/12809255-predictable-updating-about-ai-risk), or search "Joe Carlsmith Audio" on your podcast app.)
> *"This present m... | https://www.lesswrong.com/posts/bHozHrQD4qxvKdfqq/predictable-updating-about-ai-risk |
# All AGI Safety questions welcome (especially basic ones) [May 2023]
**tl;dr: Ask questions about AGI Safety as comments on this post, including ones you might otherwise worry seem dumb!**
Asking beginner-level questions can be intimidating, but everyone starts out not knowing anything. If we want more people in the... | https://www.lesswrong.com/posts/SFuLQA7guCnG8pQ7T/all-agi-safety-questions-welcome-especially-basic-ones-may |
# H-JEPA might be technically alignable in a modified form
This post is a collection of replies to Steven Byrnes' "[LeCun’s “A Path Towards Autonomous Machine Intelligence” has an unsolved technical alignment problem](https://www.lesswrong.com/posts/C5guLAx7ieQoowv3d/lecun-s-a-path-towards-autonomous-machine-intellige... | https://www.lesswrong.com/posts/umsGb5qkfzD3WarTR/h-jepa-might-be-technically-alignable-in-a-modified-form |
# Yoshua Bengio argues for tool-AI and to ban "executive-AI"
Yoshua Bengio wrote a blogpost yesterday in which he argues for developing "scientist AI", which seems in-structure very similar to historical [Tool-AI](https://www.lesswrong.com/tag/tool-ai) proposals.
For the (IMO) best response to this kind of proposal ... | https://www.lesswrong.com/posts/EP92JhDm8kqtfATk8/yoshua-bengio-argues-for-tool-ai-and-to-ban-executive-ai |
# Announcing “Key Phenomena in AI Risk” (facilitated reading group)
**TLDR: **“Key Phenomena in AI Risk” is a 7 week-long, facilitated reading group. It is aimed at people interested in conceptual AI alignment research, in particular from fields such as philosophy, systems research, biology, cognitive and social scien... | https://www.lesswrong.com/posts/mqvxR9nrXAzRr3ow9/announcing-key-phenomena-in-ai-risk-facilitated-reading |
# When is Goodhart catastrophic?
*Thanks to Aryan Bhatt, Eric Neyman, and Vivek Hebbar for feedback.*
*This post gets more math-heavy over time; we convey some intuitions and overall takeaways first, and then get more detailed. Read for as long as you're getting value out of things!*
TLDR
----
How much should you o... | https://www.lesswrong.com/posts/fuSaKr6t6Zuh6GKaQ/when-is-goodhart-catastrophic |
# Focusing on longevity research as a way to avoid the AI apocalypse
I don't want to die.
You don't want to die.
The people who are potentially going to get us killed by pushing AI capability research don't want to die.
Our most basic goals have been aligned all this time, so it's really tragic that we are in t... | https://www.lesswrong.com/posts/8v3c9bPuSbxEntmKh/focusing-on-longevity-research-as-a-way-to-avoid-the-ai |
# Summaries of top forum posts (1st to 7th May 2023)
This is part of a weekly series summarizing the top posts on the EA and LW forums - you can see the full collection [here.](https://forum.effectivealtruism.org/s/W4fhpuN26naxGCBbN) The first post includes some details on purpose and methodology. Feedback, thoughts, ... | https://www.lesswrong.com/posts/dSjYcrioi8DRpgJ4L/summaries-of-top-forum-posts-1st-to-7th-may-2023 |
# Most people should probably feel safe most of the time
There is an idea that I’ve sometimes heard around rationalist and EA circles, that goes something like “you shouldn’t ever feel safe, because nobody is actually ever safe”. I think there are at least two major variations of this:
1. You shouldn’t ever feel saf... | https://www.lesswrong.com/posts/pPLcrBzcog4wdLcnt/most-people-should-probably-feel-safe-most-of-the-time |
# EconTalk podcast: "Eliezer Yudkowsky on the Dangers of AI"
Youtube link: https://www.youtube.com/watch?v=fZlZQCTqIEo
From the description: "Eliezer Yudkowsky insists that once artificial intelligence becomes smarter than people, everyone on earth will die. Listen as Yudkokwsky speaks with EconTalk's Russ Roberts on... | https://www.lesswrong.com/posts/Kr5xh9Lx9cKRo6Gf8/econtalk-podcast-eliezer-yudkowsky-on-the-dangers-of-ai |
# Quote quiz answer
Here’s the answer to the recent [quote quiz](https://rootsofprogress.org/quote-quiz-drifting-into-dependence):
**The author was Ted Kaczynski, aka the Unabomber.** The quote was taken from his manifesto, “[Industrial Society and Its Future](https://www.washingtonpost.com/archive/politics/1995/09/1... | https://www.lesswrong.com/posts/GtFuyormXs4fuKSHT/quote-quiz-answer |
# Stampy's AI Safety Info - New Distillations #2 [April 2023]
Hey! This is another update from the distillers at the [AI Safety Info](https://aisafety.info/) website (and its more playful clone [Stampy](https://stampy.ai/)).
![](https://39669.cdn.cke-cs.com/rQvD3VnunXZu34m86e5f/images/12a7f4036983b5c41fd9ea43d8179324... | https://www.lesswrong.com/posts/EELddDmBknLyjwgbu/stampy-s-ai-safety-info-new-distillations-2-april-2023 |
# How to Interpret Prediction Market Prices as Probabilities
Or "how to account for fees"
Over at the EA Forum, bob wrote an article "[Don't Interpret Prediction Market Prices as Probabilities](https://forum.effectivealtruism.org/posts/cJc3f4HmFqCZsgGJe/don-t-interpret-prediction-market-prices-as-probabilities)".
In... | https://www.lesswrong.com/posts/7HRpCJdotxdcqbAYh/how-to-interpret-prediction-market-prices-as-probabilities |
# A Search for More ChatGPT / GPT-3.5 / GPT-4 "Unspeakable" Glitch Tokens
Inspired by [the](https://www.lesswrong.com/posts/aPeJE8bSo6rAFoLqg/solidgoldmagikarp-plus-prompt-generation) [recent](https://www.lesswrong.com/posts/Ya9LzwEbfaAMY8ABo/solidgoldmagikarp-ii-technical-details-and-more-recent) [posts](https://www.... | https://www.lesswrong.com/posts/kmWrwtGE9B9hpbgRT/a-search-for-more-chatgpt-gpt-3-5-gpt-4-unspeakable-glitch |
# AI Safety Newsletter #5: Geoffrey Hinton speaks out on AI risk, the White House meets with AI labs, and Trojan attacks on language models
Welcome to the AI Safety Newsletter by the [Center for AI Safety](https://www.safe.ai/). We discuss developments in AI and AI safety. No technical background required.
Subscribe ... | https://www.lesswrong.com/posts/X8yhso2noTKXfFE8s/ai-safety-newsletter-5-geoffrey-hinton-speaks-out-on-ai-risk |
# Result Of The Bounty/Contest To Explain Infra-Bayes In The Language Of Game Theory
A month and a half ago I [announced](https://www.lesswrong.com/posts/ig62gqK2xRPv79Wsr/usd500-bounty-contest-explain-infra-bayes-in-the-language-of) a $500 bounty/prize to explain infrabayes work in the language of game theory. Other... | https://www.lesswrong.com/posts/XCDwfr8LER8XnDaCv/result-of-the-bounty-contest-to-explain-infra-bayes-in-the |
# Exciting New Interpretability Paper!
There's a pretty exciting new interpretability paper, which hasn't really received the requisite attention because it's not billed as such.
This [paper](https://arxiv.org/pdf/2302.07253.pdf) modifies the transformer architecture so that a forward pass minimizes a specifically en... | https://www.lesswrong.com/posts/K7PrLpccS6D5Ptfbb/exciting-new-interpretability-paper |
# Making Up Baby Signs
Babies are often mentally ready to start talking before they have enough control over their mouths, at a time when they may have enough control over their arms and hands to make simple gestures; a lot of otherwise verbal families have had good experiences with signing. If you look around online ... | https://www.lesswrong.com/posts/EtrEhm5g7ZoPPn4Pw/making-up-baby-signs |
# Solving the Mechanistic Interpretability challenges: EIS VII Challenge 1
We solved the first (edit: and [second](https://www.alignmentforum.org/posts/k43v47eQjaj6fY7LE/solving-the-mechanistic-interpretability-challenges-eis-vii-1)) Mechanistic Interpretability challenge that [Stephen Casper](https://www.alignmentfor... | https://www.lesswrong.com/posts/sTe78dNJDGywu9Dz6/solving-the-mechanistic-interpretability-challenges-eis-vii |
# Have you heard about MIT's "liquid neural networks"? What do you think about them?
I came across [this video](https://www.youtube.com/watch?v=iRXZ5vQ6mGE) by MIT CSAIL.
Here is the article they are talking about: [https://www.science.org/doi/10.1126/scirobotics.adc8892](https://www.science.org/doi/10.1126/sciroboti... | https://www.lesswrong.com/posts/jAfNzdJHcZB2DWPBt/have-you-heard-about-mit-s-liquid-neural-networks-what-do |
# Progress links and tweets, 2023-05-09
**Announcements**
-----------------
* [Stripe Press is publishing *Poor Charlies Almanack*](https://www.stripe.press/poor-charlies-almanack) (via [@patrickc](https://twitter.com/patrickc/status/1654508051326701575)). Also, Londoners: [sign up for events from Stripe Press and ... | https://www.lesswrong.com/posts/fvswAk4W8DGcNESWo/progress-links-and-tweets-2023-05-09 |
# Stopping dangerous AI: Ideal US behavior
*Disclaimer: this post doesn't have the answers. Moreover, it's an unfinished draft. Hopefully a future version will be valuable, but that will only occur if I revise/rewrite it. For now you're better off reading sources linked from *[*AI policy ideas: Reading list*](https://... | https://www.lesswrong.com/posts/Ybw7LfZWPRbEEKa5s/stopping-dangerous-ai-ideal-us-behavior |
# Red teaming: challenges and research directions
*This post is a follow-up to* [*Safety Standards: a framework for AI regulation*](https://www.lesswrong.com/posts/j2LD87wT3dpr4m7rs/safety-standards-a-framework-for-ai-regulation)*. In the previous post, I claimed that competent red-teaming organizations will be essent... | https://www.lesswrong.com/posts/daPKk4coxx5xha7A3/red-teaming-challenges-and-research-directions |
# Gradient hacking via actual hacking
In this post, I begin with a thought experiment and then pose a question: is there a threat model for gradient hacking which accounts for the possibility of [side-channel attacks](https://en.wikipedia.org/wiki/Side-channel_attack) or other [vulnerabilities](https://www.cvedetails.... | https://www.lesswrong.com/posts/jrtpmdk68R2yZ7ufv/gradient-hacking-via-actual-hacking |
# Naturalist Experimentation
Ordinarily, when someone is trying to solve a problem, experimentation is where they *begin*. “Things aren’t going right? Try changing something, and see what happens!” In naturalism, though, experimentation comes at the very end, if it happens at all.
There’s a balance to be kept in natu... | https://www.lesswrong.com/posts/gg4q9QS7tfEhX7hyK/naturalist-experimentation |
# New OpenAI Paper - Language models can explain neurons in language models
Summary by OpenAI: We use GPT-4 to automatically write explanations for the behavior of neurons in large language models and to score those explanations. We release a dataset of these (imperfect) explanations and scores for every neuron in GPT... | https://www.lesswrong.com/posts/2qTxffyqeR4gbpEua/new-openai-paper-language-models-can-explain-neurons-in |
# In defence of epistemic modesty [distillation]
This is a distillation of [*In defence of epistemic modesty*](https://www.lesswrong.com/posts/SJKowjkGF7z2sS98f/in-defence-of-epistemic-modesty), a 2017 essay by Gregory Lewis. I hope to make the essay’s key points accessible in a quick and easy way so more people engag... | https://www.lesswrong.com/posts/H3rjMpndZkqEQhS6k/in-defence-of-epistemic-modesty-distillation |
# Class-Based Addressing
When we have multiple ships built along similar designs, it's customary to refer to them by the initial example of the design. For example, the last American battleships are called [Iowa-class](https://en.wikipedia.org/wiki/Iowa-class_battleship), after their [lead ship](https://en.wikipedia.o... | https://www.lesswrong.com/posts/otndrzDMMP5jDDqAS/class-based-addressing |
# AGI-Automated Interpretability is Suicide
Backstory: I wrote this post about 12d ago, then a user here pointed out that this could be capability exfohazard since it could give the bad idea of having the AGI look at itself, so I took it down. Well I don’t have to worry about that anymore since now we have proof that ... | https://www.lesswrong.com/posts/pQqoTTAnEePRDmZN4/agi-automated-interpretability-is-suicide |
# Roadmap for a collaborative prototype of an Open Agency Architecture
Deger Turan, Tantum Collins, [davidad](https://www.alignmentforum.org/users/davidad) and [Charbel-Raphael Segerie](https://www.alignmentforum.org/users/charbel-raphael) contributed to the writing of this post.
Andrew Critch, Ben Goldhaber, Ozzie ... | https://www.lesswrong.com/posts/pHJtLHcWvfGbsW7LR/roadmap-for-a-collaborative-prototype-of-an-open-agency |
# A Corrigibility Metaphore - Big Gambles
I present here a helpful analogy to understand the[*corrigibility problem*](https://www.lesswrong.com/tag/corrigibility) and the challenge raised by MIRI in their [proposal](https://intelligence.org/files/Corrigibility.pdf). This analogy simplifies greatly some challenges of c... | https://www.lesswrong.com/posts/pwPo7L2RjNvYDAo7g/a-corrigibility-metaphore-big-gambles |
# Mental Health and the Alignment Problem: A Compilation of Resources (updated April 2023)
*This is a post about mental health and disposition in relation to the alignment problem. It compiles a number of resources that address how to maintain wellbeing and direction when confronted with existential risk. *
*Many peo... | https://www.lesswrong.com/posts/pLLeGA7aGaJpgCkof/mental-health-and-the-alignment-problem-a-compilation-of |
# Better debates
*[Crossposted from [https://tsvibt.blogspot.com/2023/05/better-debates.html](https://tsvibt.blogspot.com/2023/05/better-debates.html).]*
When two people disagree about a proposition even though they've thought about it alot, the disagreement is often hard to resolve. There's a gulf of data, concepts,... | https://www.lesswrong.com/posts/fPgn8ZMtsr6az6D9t/better-debates |
# AI interpretability could be harmful?
A superhuman ethical AI might want to model adversaries and their actions, e.g., model which bioweapons an adversary might develop and prepare response plans and antidotes. If such predictions are done in interpretable representations, they could themselves be used by an adversa... | https://www.lesswrong.com/posts/CRrkKAafopCmhJEBt/ai-interpretability-could-be-harmful |
# How should one feel morally about using chatbots?
Personally, I feel some confusion.
On the one hand, ones personal usage of a chatbot probably wouldn't be more than a drop in the bucket in terms of producing bad consequences such as increasing the demand for chatbots and marginally pushing AI capabilities forward.... | https://www.lesswrong.com/posts/xZKFitRF8vyg6a2KX/how-should-one-feel-morally-about-using-chatbots |
# The Academic Field Pyramid - any point to encouraging broad but shallow AI risk engagement?
I am an adjunct instructor in history. Recently I was teaching the early Cold War in my American History survey class for my community college students and decided to go off on a 20 minute digression in class about how the 19... | https://www.lesswrong.com/posts/GDwCLZnyMC475NpK3/the-academic-field-pyramid-any-point-to-encouraging-broad |
# How should we think about the decision relevance of models estimating p(doom)?
To illustrate what I mean, switching from p(doom) to timelines:
* The recent post [AGI Timelines in Governance: Different Strategies for Different Timeframes](https://www.lesswrong.com/posts/6CjnFcsRHJesR9MEA/agi-timelines-in-governa... | https://www.lesswrong.com/posts/ubiSawqhHe2GAHXWX/how-should-we-think-about-the-decision-relevance-of-models |
# Why are we so complacent about AI hell?
In my mind, interventions against s-risks from AI seem like the impartial[^6356xxgakdq] top priority of our time, being more tractable[^0n9srpionsa], important[^9ht9xep78po], and neglected[^13rwv8cpt9i] than alignment. Hence I’m surprised that they’re not as central as alignme... | https://www.lesswrong.com/posts/5Jmhdun9crJGAJGyy/why-are-we-so-complacent-about-ai-hell |
# Separating the "control problem" from the "alignment problem"
I'm often confused by how "loss of control" is somewhat similar in meaning to "misalignment" in the context of AGI risk. An average person might reasonably equate the two, or feel like there's an large overlap of meaning between the two words. Here's my a... | https://www.lesswrong.com/posts/ykccy6LXfmTjZcp6S/separating-the-control-problem-from-the-alignment-problem |
# Contra Caller Signs
A contra dance stage can be a bit chaotic and noisy, especially during a dance. It's a good fit for gestures, and over time callers and musicians have settled on a few covering the main things they need to communicate. I couldn't find a listing of these anywhere, so here are the main ones:
* C... | https://www.lesswrong.com/posts/yvomXJeQzswcRmseB/contra-caller-signs |
# Fatebook for Slack: Track your forecasts, right where your team works
Announcing [Fatebook for Slack](https://fatebook.io/for-slack) \- a Slack bot designed to help high-impact orgs build a culture of forecasting.
With Fatebook, you can ask a forecasting question in your team's Slack:
 and I ([link](https://twitter.com/azsantosk/status/1655998142544846865)).
Conditions are similar to [this question](https://www.metaculus.com/quest... | https://www.lesswrong.com/posts/sHcHriEoJAGcRt3oC/i-bet-usd500-on-ai-winning-the-imo-gold-medal-by-2026 |
# AI #11: In Search of a Moat
Remember the _start_ of the week? That’s when everyone was talking about a leaked memo from a Google employee, saying that neither Google nor OpenAI had a moat and the future belonged to open source models. The author was clearly a general advocate for open source in general. If he is rig... | https://www.lesswrong.com/posts/yWtJL6kPnLRxiKp2B/ai-11-in-search-of-a-moat |
# Advice for newly busy people
After writing "[Advice for interacting with busy people](https://www.lesswrong.com/posts/v8xoqiwaYu9krmjsY/advice-for-interacting-with-busy-people)", I was asked to write a follow-up on advice for newly busy people. So, here's a quick list of tools and mental models that help me prioriti... | https://www.lesswrong.com/posts/RaNhnNjExip36NMxM/advice-for-newly-busy-people |
# Alignment, Goals, and The Gut-Head Gap: A Review of Ngo. et al.
(Cross-posted from my [Substack](https://theviolethour.substack.com/p/alignment-goals-and-the-gut-head). Also on the [EA Forum](https://forum.effectivealtruism.org/posts/2yyZqRParGeLEja5u/alignment-goals-and-the-gut-head-gap-a-review-of-ngo-et-al))
If ... | https://www.lesswrong.com/posts/9fjSi2KzyzXAQxuza/alignment-goals-and-the-gut-head-gap-a-review-of-ngo-et-al |
# Thoughts on LessWrong norms, the Art of Discourse, and moderator mandate
A couple of weeks ago I asked [Should LW have an official list of norms?](https://www.lesswrong.com/posts/gugkWsfayJZnicAew/should-lw-have-an-official-list-of-norms) and I appreciate the responses there. Here I want to say what I'm currently th... | https://www.lesswrong.com/posts/6CM7rcnTBjoeE9M8S/thoughts-on-lesswrong-norms-the-art-of-discourse-and |
# Three Iterative Processes
Throughout all four phases of naturalism, there are three processes that are constantly repeating in the background: Story revision, POU loops, and a few kinds of oscillation that together I think of as "breathing".
Story Revision
--------------
Story revision begins at the very start of ... | https://www.lesswrong.com/posts/P56T5QXzrHpPWAmrJ/three-iterative-processes |
# Formulating the AI Doom Argument for Analytic Philosophers
Recently, [David Chalmers has asked on Twitter](https://twitter.com/davidchalmers42/status/1647333812584562688?s=20) for a canonical formulation of the AI Doom Argument. After some back and forth with EY, DC came up with a [suggestion](https://l.facebook.com... | https://www.lesswrong.com/posts/TC7GhGaKFqTtQH9Aq/formulating-the-ai-doom-argument-for-analytic-philosophers |
# Input Swap Graphs: Discovering the role of neural network components at scale
*This post was written as part of the work done at *[*Conjecture*](https://www.conjecture.dev/)*.*
*You can try input swap graphs in a* [*collab notebook*](https://colab.research.google.com/drive/1iZ0nB0aaQSkJRyfAP4DDLKLsbUTfP_V5#scrollTo... | https://www.lesswrong.com/posts/ZSYo97kcfwtFdpcwe/input-swap-graphs-discovering-the-role-of-neural-network |
# Infinite-width MLPs as an "ensemble prior"
**Summary: **A simple toy model suggests that infinitely wide MLPs[^iunvz264mpj] generalize in an "ensemble-ish" way which is exponentially less data-efficient than Solomonoff induction. It's probably fixable by different initializations and/or regularizations, so I note ... | https://www.lesswrong.com/posts/CQMhLujqMpQ78Ru3R/infinite-width-mlps-as-an-ensemble-prior |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.