text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Recreating logic in type theory
This is the second entry in my series on type theory. In this post, we will talk about how to recreate logic in type theory.
As you might know, in order to define [ZFC set theory](https://en.wikipedia.org/wiki/Zermelo%E2%80%93Fraenkel_set_theory), you first define first-order (FO) lo... | https://www.lesswrong.com/posts/MHBPPv8NfsSe3KiMF/recreating-logic-in-type-theory |
# What readings did you consider best for the happy parts of the secular solstice?
I'm currently thinking about creating a new rationalist ritual day that focuses just on happiness. I'm thinking about reusing some elements that worked well at the solstice. Which readings did you enjoy very much during the happy parts ... | https://www.lesswrong.com/posts/v2odsk4nzcjNGwmwt/what-readings-did-you-consider-best-for-the-happy-parts-of |
# Think wider about the root causes of progress
Too much discussion of the Industrial Revolution is myopic, focused narrowly on a few highlights such as steam and coal. The IR was part of broader trends that are wider in scope and longer in time than its traditional definition encompasses. To understand anything, it i... | https://www.lesswrong.com/posts/Ahit7vnAp7tGsgGQE/think-wider-about-the-root-causes-of-progress |
# Not Getting Hacked
A lot of people get hacked or phished, and being completely secure is somewhere between difficult and impossible, especially if someone trying hard to hack you in particular. But there are several simple things you can do that decrease your risk a lot:
* Use a password manager that fills in fie... | https://www.lesswrong.com/posts/ESp3SsuqWrht7SApu/not-getting-hacked |
# Naive Set Theory, Halmos
*Epistemic status: Unorganized thoughts of mine generated after reading this book -- not a concerted attempt to convey any single particular mathematical insight.*
Set theory is the mathematical study of collections. These collections -- *sets* \-\- can be either the collection of nothing (... | https://www.lesswrong.com/posts/munnmo4c2rpQbbxLt/naive-set-theory-halmos |
# Applied Linear Algebra Lecture Series
Over the past couple months, I gave weekly lectures on applied linear algebra. The lectures cover a grab-bag of topics which I've needed to know for my own work, but which typically either aren't covered in courses or are covered only briefly in advanced courses which use them (... | https://www.lesswrong.com/posts/LKAXgTen4Xbqb8eZY/applied-linear-algebra-lecture-series |
# Take 13: RLHF bad, conditioning good.
*As a writing exercise, I'm writing an AI Alignment Hot Take Advent Calendar - one new hot take, written ~~every day~~ some days for 25 days. I have now procrastinated enough that I probably have enough hot takes.*
*Edit - I should have cited* [*Buck's recent post*](https://www... | https://www.lesswrong.com/posts/AXpXG9oTiucidnqPK/take-13-rlhf-bad-conditioning-good |
# Notes on Meta's Diplomacy-Playing AI
_Disclaimer: I'm not an expert at machine learning, AI safety or Diplomacy, so there may be errors here, though hopefully no major ones. For previous discussion of CICERO on here, see the comments in [this post](https://www.... | https://www.lesswrong.com/posts/oT8fmwWddGwnZbbym/notes-on-meta-s-diplomacy-playing-ai |
# Mental acceptance and reflection
Naive investigation of the world often works, especially when we have direct access, fast feedback loops, and can try many times. Sports, games, and even parts of science often have properties. When we investigate, say, microorganisms, we can directly experiment on what we are intere... | https://www.lesswrong.com/posts/tJyuHfRpENXYjCJit/mental-acceptance-and-reflection |
# Secular Solstice for children
Sometimes the Boston [Secular Solstice](https://secularsolstice.com/) has been hosted at my (Julia's) house because my husband runs the music and physical setup. While the adults are doing a long solstice downstairs, I’ve sometimes run a version for our kids upstairs. This year I did... | https://www.lesswrong.com/posts/Wzzr2g4imrtJTma3L/secular-solstice-for-children |
# Staring into the abyss as a core life skill
Recently I’ve been thinking about how all my favorite people are great at a skill I’ve labeled in my head as “staring into the abyss.”^[1](#fn:1)^
Staring into the abyss means thinking reasonably about things that are uncomfortable to contemplate, like arguments against y... | https://www.lesswrong.com/posts/vzfz4AS6wbooaTeQk/staring-into-the-abyss-as-a-core-life-skill |
# Response to Holden’s alignment plan
*This work was supported by the Monastic Academy for the Preservation of Life on Earth. You can support my work directly* [*here*](http://alexflint.io/donate.html)*.*
*I will give a short presentation of this work followed by discussion on Saturday (Dec 24) at 9am Pacific / 12p... | https://www.lesswrong.com/posts/CBvebC9FgSMtsD5T9/response-to-holden-s-alignment-plan |
# Lead in Chocolate?
Consumer Reports recently tested a bunch of dark chocolate for lead and cadmium, and found that:
> For 23 of the bars, eating just an ounce a day would put an adult over a level that public health authorities and CR's experts say may be harmful for at least one of those heavy metals. Five of the ... | https://www.lesswrong.com/posts/DdJZ9MFK8xPLpxz6k/lead-in-chocolate |
# Racing through a minefield: the AI deployment problem
_Click lower right to download or find on Apple Podcasts, Spotify, Stitcher, etc._
In previous pieces, I argue... | https://www.lesswrong.com/posts/PABtHv8X28jJdxrD6/racing-through-a-minefield-the-ai-deployment-problem |
# China Covid #4
China has now decided to let Covid take its course, so it makes sense to spin developments there off into their own posts for a bit. It is hard to get a good read on what is happening, because China’s core strategy was to stop controlling the spread of Covid and instead to put tight controls on its _C... | https://www.lesswrong.com/posts/PTWxhnENh38Gm7cti/china-covid-4 |
# Covid 12/22/22: Reevaluating Past Options
This is the non-China section of this week’s update. Despite there being essentially zero good statistical data about the real situation in China, things are clearly moving fast, and I have assembled what useful information I have been able to find. If cases are rising as qu... | https://www.lesswrong.com/posts/DnobPFEZwEjrsSgfp/covid-12-22-22-reevaluating-past-options |
# December 2022 updates and fundraising
_Harlan Stewart and _Katja Grace_*_, _22 December, 2022_
News
----
### New Hires and role changes
In 2022, the AI Impacts team has grown from two to seven full time staff. Out of more than 250 applicants, we hired Elizabeth Santos as Operations Lead, Harlan Stewart as Researc... | https://www.lesswrong.com/posts/KbNyr2n5tW5uRoypq/december-2022-updates-and-fundraising |
# Some Notes on the mathematics of Toy Autoencoding Problems
Anthropic's recent mechanistic interpretability paper, [Toy Models of Superposition](https://transformer-circuits.pub/2022/toy_model/index.html), helps to demonstrate the conceptual richness of very small feedforward neural networks. Even when being trained ... | https://www.lesswrong.com/posts/dodEwdiphp5TJZjJj/some-notes-on-the-mathematics-of-toy-autoencoding-problems |
# Let’s think about slowing down AI
**Averting doom by not building the doom machine**
--------------------------------------------------
If you fear that someone will build a machine that will seize control of the world and annihilate humanity, then one kind of response is to try to build further machines that will ... | https://www.lesswrong.com/posts/uFNgRumrDTpBfQGrs/let-s-think-about-slowing-down-ai |
# Article Review: Discovering Latent Knowledge (Burns, Ye, et al)
\[This is a cross-post from my blog at [aizi.substack.com](https://aizi.substack.com/).\]
Large Language Models are hell-demons summoned from the data dimension. You feed a giant machine approximately everything ever written, and suddenly it can [have ... | https://www.lesswrong.com/posts/WtKGLJQfjCWTm7tFK/article-review-discovering-latent-knowledge-burns-ye-et-al |
# "Search" is dead. What is the new paradigm?
For over two decades internet search (which has been largely synonymous with Google for most of that time) has been the main way to augment the human brain with the digital information outside of it. First the virtual assistants and now LLMs are challenging and changing th... | https://www.lesswrong.com/posts/zjSdrbYfCKcM2giZE/search-is-dead-what-is-the-new-paradigm |
# Epigenetics of the mammalian germline
*This is a crosspost from* [*my Substack*](https://denovo.substack.com) *and is is Part 2 of my epigenetics post series. Part 1 is* [*here*](https://denovo.substack.com/p/what-is-epigenetics)*.*
The year was 1984, and scientists were wondering: why is sperm, and not just eggs, ... | https://www.lesswrong.com/posts/oY8BPGiknEkAEYDti/epigenetics-of-the-mammalian-germline |
# On sincerity
Cross-posted from [my website](https://joecarlsmith.com/2022/12/23/on-sincerity). Audio version [here](https://joecarlsmithaudio.buzzsprout.com/2034731/11922489-on-sincerity), or search "Joe Carlsmith Audio" on your podcast app.
> *Nearby is the country they call life.*
>
> *You will know it by its se... | https://www.lesswrong.com/posts/WDwGqLd2Puedouogq/on-sincerity |
# ChatGPT understands, but largely does not generate Spanglish (and other code-mixed) text
*Content format: Commented chat screenshots, and then some thoughts on their implications.*
*Epistemic status: Exploratory*
Introduction
============
[Code-mixing](https://en.wikipedia.org/wiki/Code-mixing) is the ad-hoc mixi... | https://www.lesswrong.com/posts/TZPbm3BRkWWTm9ecC/chatgpt-understands-but-largely-does-not-generate-spanglish |
# Two Truths and a Prediction Market
**Summary**: Play some guessing games using a prediction market, to get used to how those markets work.
**Tags**: Small to Medium, Repeatable
**Purpose**: Prediction markets are a relatively popular tool in rationalist spaces, but uncommon in the general population. This is an op... | https://www.lesswrong.com/posts/DER2WdFhppzdAbKkh/two-truths-and-a-prediction-market |
# To err is neural: select logs with ChatGPT
This post includes some portions of my logs with ChatGPT, where I
asked it basic arithmetic questions and it made several mistakes. I
have used **bold** for cases where it made incorrect statements, and
*italics* for cases where it made correct statements that directly
cont... | https://www.lesswrong.com/posts/w7c2iKjsDrNqoymxQ/to-err-is-neural-select-logs-with-chatgpt |
# Why is "Argument Mapping" Not More Common in EA/Rationality (And What Objections Should I Address in a Post on the Topic?)
I recently published [an article](https://georgetownsecuritystudiesreview.org/2022/11/30/complexity-demands-adaptation-two-proposals-for-facilitating-better-debate-in-international-relations-and... | https://www.lesswrong.com/posts/bHXpbf6jXc4bgyrrp/why-is-argument-mapping-not-more-common-in-ea-rationality |
# Loose Threads on Intelligence
*Epistemic Status: Unfinished deep-dive into the nature of intelligence*[^21jca9qojju]*. I* [*committed*](https://www.lesswrong.com/posts/ypTdBWxYdvTy64QF7/research-principles-for-6-months-of-ai-alignment-studies) *to writing down my research path, but three weeks in I don't have a cohe... | https://www.lesswrong.com/posts/sky64wuH7dyJrwsaT/loose-threads-on-intelligence |
# Durkon, an open-source tool for Inherently Interpretable Modelling
*or, what abstractapplic has been tinkering with for the past two years, and how it can be applied to fictional card games*
Motivation
==========
Interpretability of ML models is important. If people don’t understand how a model makes its predictio... | https://www.lesswrong.com/posts/GQPJjccM4hTDhn5KF/durkon-an-open-source-tool-for-inherently-interpretable |
# Löb's Lemma: an easier approach to Löb's Theorem
Related to: [Löb's Theorem for implicit reasoning in natural language: Löbian party invitations](https://www.lesswrong.com/posts/QsZ3ycfRYs2ps5sNA/loeb-s-theorem-for-implicit-reasoning-in-natural-language)
tl;dr: Löb's Theorem is much easier to grok if you separate t... | https://www.lesswrong.com/posts/AkGvmJ6WE5sXjuwnC/loeb-s-lemma-an-easier-approach-to-loeb-s-theorem |
# Content and Takeaways from SERI MATS Training Program with John Wentworth
Introduction
============
I participated in the training program for John Wentworth’s stream of [SERI MATS](https://www.serimats.org/), and overall, it was a very good experience! This post is meant to convey the content of the workshops, so ... | https://www.lesswrong.com/posts/kpkxKDpiRn6BNArFm/content-and-takeaways-from-seri-mats-training-program-with |
# The case against AI alignment
Trigger warning: Discussion of seriously horrific shit. Honestly, everything is on the table here so if you're on the lookout for trigger warnings you should probably stay away from this conversation.
Any community of people which gains notability will attract criticism. Those who advo... | https://www.lesswrong.com/posts/CtXaFo3hikGMWW4C9/the-case-against-ai-alignment |
# Non-Elevated Air Purifiers
Last year for Christmas gatherings at my dad's house I put together [wall-mounted filter cubes](https://www.jefftk.com/p/elevated-air-purifier-cubes), to get them up and out of the way. This time around this seemed like too much hassle, so I put one together on top of the piano, one in a b... | https://www.lesswrong.com/posts/HAhMEvDNhudRm39Xg/non-elevated-air-purifiers |
# Three reasons to cooperate
In this post I’ll discuss three reasons to cooperate in a [truly one-shot](https://www.lesswrong.com/posts/HFyWNBnDNEDsDNLrZ/the-true-prisoner-s-dilemma#) prisoner’s dilemma:
* **Kindness**: You care about the other player.
* **Correlation**. Your decision is correlated with the other... | https://www.lesswrong.com/posts/mm8sFBpPH3Bb2NhGg/three-reasons-to-cooperate |
# Contra Steiner on Too Many Natural Abstractions
Reply To: "[Take 4: One problem with natural abstractions is there's too many of them.](https://www.lesswrong.com/posts/NK4XxyrjFWt83m3dx/take-4-one-problem-with-natural-abstractions-is-there-s-too)"
Epistemic Status
----------------
Unconfident. I don't yet have a f... | https://www.lesswrong.com/posts/kFPAMuYa7FhKwHt7f/contra-steiner-on-too-many-natural-abstractions |
# Shared reality: a key driver of human behavior
I once asked Robin Hanson if he really thought status-seeking was such a dominant driver of human behavior. I said humans had dozens of factors motivating their behavior, it was crazy to claim there was One Big Thing. He replied (something to the effect of) "well, even ... | https://www.lesswrong.com/posts/73QyjLymEak4L8RDC/shared-reality-a-key-driver-of-human-behavior |
# A hundredth of a bit of extra entropy
There are two ways to calculate the amount of information in one term of a continued fraction:
* The entropy of the Gauss-Kuzmin distribution is about 3.4325 bits.
* Twice the logarithm of the Khinchin-Lévy constant is about 3.4237 bits.
These differ by about 0.0088 bits. ... | https://www.lesswrong.com/posts/DWH5XcqeXqQjXXyyo/a-hundredth-of-a-bit-of-extra-entropy |
# Hyperfinite graphs ~ manifolds
Manifolds are better described as 'shapes'.
Hyperfinite is an unlimited number, often an integer. A whole number larger than any standard n.
Imagine a graph with hyperfinitely many vertices and edges, where the edges have infinitesimal weights. If I were nearer my iPad I'd draw it.
... | https://www.lesswrong.com/posts/L2RT5NctYYt37MdsA/hyperfinite-graphs-manifolds |
# Simplified Level Up
Vienna Teng's [Level Up](https://viennateng.bandcamp.com/track/level-up) is a great song, but when we wanted to use it in our secular solstice ( [retrospective](https://www.jefftk.com/p/boston-solstice-2022-retrospective)) it wasn't going to work as-is. The problem is she's a professional singer ... | https://www.lesswrong.com/posts/WCQGMgrYxsDBCS2Y6/simplified-level-up |
# Take 14: Corrigibility isn't that great.
*As a writing exercise, I'm writing an AI Alignment Hot Take Advent Calendar - one new hot take, written ~~every day~~ some days for 25 days.*
*It's the end (I saved a tenuous one for ya')! Kind of disappointing that this ended up averaging out to one every 2 days, but this ... | https://www.lesswrong.com/posts/MWBFnH225LgRxfHgd/take-14-corrigibility-isn-t-that-great |
# The Meditation on Winter
Preface
-------
For the first several winter solstice ceremonies that I ran, I wrote a new, heartfelt speech for the moment of darkness – essentially the "keynote" speech of the event. The speeches were fueled by grief and difficult challenges that I was facing. They were good speeches. Peo... | https://www.lesswrong.com/posts/pnhjfkcBpzGp7gFTJ/the-meditation-on-winter |
# ChatGPT is our Wright Brothers moment
The Wright Brother’s first flight was a joke by modern flying standards. Even though proving heavier than air flight was viable was monumental, many people at the time could not see how a machine that only can fly a short distance, a few feet off the ground, with the pilot in an... | https://www.lesswrong.com/posts/3TjfowjTcHEL56Nyp/chatgpt-is-our-wright-brothers-moment |
# Accurate Models of AI Risk Are Hyperexistential Exfohazards
(Where "an exfohazard" is [information which leads to bad outcomes if known by a large fraction of society](https://www.lesswrong.com/posts/yET7wbjjJZtpz6NF3/don-t-use-infohazard-for-collectively-destructive-info).)
Let us suppose that we've solved the tec... | https://www.lesswrong.com/posts/xAzKefLsYdFa4SErg/accurate-models-of-ai-risk-are-hyperexistential-exfohazards |
# How evolutionary lineages of LLMs can plan their own future and act on these plans
TLDR
----
LLM lineages can plan their future and act on these plans, using the internet as the storage of event memory. “We” are not guaranteed to “out-OODA” them, even if their OODA loop will be six months or one year because the OO... | https://www.lesswrong.com/posts/ddR8dExcEFJKJtWvR/how-evolutionary-lineages-of-llms-can-plan-their-own-future |
# YCombinator fraud rates
Summary
=======
I estimate that 1-2% of $100M+ YCombinator-backed companies have faced serious allegations of fraud.
Details
=======
* I’m interested in better understanding the base rates of fraud in high-growth companies. YCombinator is a convenience sample for “high-growth companies”,... | https://www.lesswrong.com/posts/ji4cvzir2qdpALLHo/ycombinator-fraud-rates |
# [Hebbian Natural Abstractions] Mathematical Foundations
**TL;DR:** *We showed how Hebbian learning with weight decay could enable a) feedforward circuits (one-to-many) to extract the first principal component of a barrage of inputs and b) recurrent circuits to amplify signals which are present across multiple input ... | https://www.lesswrong.com/posts/EASv46FpehppAFHSm/hebbian-natural-abstractions-mathematical-foundations |
# It's time to worry about online privacy again
As we all know, if you have nothing to hide, you have nothing to fear. Nobody cares about your private life. You are not an important geopolitical target. Nobody's going to spy on you to know [what weird pornography you watch](https://sea.ign.com/final-fantasy-vii-remake... | https://www.lesswrong.com/posts/PEmAi3PFjXezJz3js/it-s-time-to-worry-about-online-privacy-again |
# Concrete Steps to Get Started in Transformer Mechanistic Interpretability
***Disclaimer**: This post mostly links to resources I've made. I feel somewhat bad about this, sorry! Transformer MI is a pretty young and small field and there just aren't many people making educational resources tailored to it. *[*Some*](ht... | https://www.lesswrong.com/posts/9ezkEb9oGvEi6WoB3/concrete-steps-to-get-started-in-transformer-mechanistic |
# Causal abstractions vs infradistributions
**Summary**: I illustrate the relationship between Wentworth-style causal abstractions and infradistributions, and how they both deal with nonrealizability by throwing away information. If you have a basic intuition for causal abstractions, this might help you understand inf... | https://www.lesswrong.com/posts/zxhdxx7o453uq2zWr/causal-abstractions-vs-infradistributions |
# Mlyyrczo
This blog post is not intended for direct human consumption. The intended method for interacting with the information contained within is **indirectly**, via an LLM (large language model) trained on this document. LLMs are not yet [2022-12-25] powerful enough to emulate mlyyrczo correctly. This document is ... | https://www.lesswrong.com/posts/GqyQSwYrryc4e2hgf/mlyyrczo |
# Analogies between Software Reverse Engineering and Mechanistic Interpretability
These are notes taken during a call with [Itay Yona](https://mentaleap.ai/), an expert in software/hardware reverse engineering (SRE). Itay gave me an excellent distillation of key ideas and mindsets in the field, and we discussed analog... | https://www.lesswrong.com/posts/tsYcsZAkKsqLXC3Bu/analogies-between-software-reverse-engineering-and |
# Looking Back on Posts From 2022
I am taking stock of my first year on Substack, and my first year with the resources necessary to focus on attempting to be a public intellectual of sorts.
The results are a mixed bag. Things are progressing, but slowly. Everything takes longer than one thinks or expects. Finding use... | https://www.lesswrong.com/posts/tEXdeAcp8mKtuNCZ5/looking-back-on-posts-from-2022 |
# An overview of some promising work by junior alignment researchers
We’re all familiar with [ELK](https://www.lesswrong.com/posts/qHCDysDnvhteW7kRd/arc-s-first-technical-report-eliciting-latent-knowledge), [natural abstractions](https://www.lesswrong.com/posts/cy3BhHrGinZCp3LXE/testing-the-natural-abstraction-hypothe... | https://www.lesswrong.com/posts/jcFSEbXEfKgMwETqw/an-overview-of-some-promising-work-by-junior-alignment |
# Coherent extrapolated dreaming
*This work was supported by the Monastic Academy for the Preservation of Life on Earth. You can support my work* [*here*](http://alexflint.io/donate.html)*.*
*I will give a short presentation of this work followed by discussion on Wednesday Dec 28 at 12pm Pacific / 3pm Eastern.* [*R... | https://www.lesswrong.com/posts/vdXNPzuh3fwgykvKY/coherent-extrapolated-dreaming |
# Safety of Self-Assembled Neuromorphic Hardware
*Authors: Can Rager*[^ad0v81y8lb]* and Kyle Webster, PhD *
Summary
=======
The scalability of modern computing hardware is limited by physical bottlenecks and high energy consumption. These limitations could be addressed by neuromorphic hardware (NMH) which is inspire... | https://www.lesswrong.com/posts/ngqFnDjCtWqQcSHXZ/safety-of-self-assembled-neuromorphic-hardware |
# Theodicy and the simulation hypothesis, or: The problem of simulator evil
*Philosophy Bear here. At the moment I'm composing an anthology of all the work I've done on the topic of AI. Simultaneously, as I edit those works for the anthology, I thought it would be a good idea to crosspost the here, as I've never shar... | https://www.lesswrong.com/posts/PbfFFs6koYqueaH8o/theodicy-and-the-simulation-hypothesis-or-the-problem-of |
# Are men harder to help?
[Of Boys and Men](https://www.brookings.edu/book/of-boys-and-men/) is a new book about how the outcomes of American men and boys are lagging in several ways. Most of it concerns aggregate trends that are irrefutable, such as the gender gap in college graduation. But books like these are often... | https://www.lesswrong.com/posts/nqwudd5NxbpEpZhkK/are-men-harder-to-help |
# Announcing: The Independent AI Safety Registry
Are you an independent AI Safety Researcher who would like to get in touch with other people in the same position? [Fill out](https://forms.gle/Sd2YgsnVqLDdrrSQ9) your location, work details, and collaboration interests, and you'll be added to a public google sheets fil... | https://www.lesswrong.com/posts/mXaugZyivQN3Eg8G3/announcing-the-independent-ai-safety-registry |
# Avoiding perpetual risk from TAI
Stephen Casper, [scasper@mit.edu](mailto:scasper@mit.edu). Thanks to Rose Hadshar and Daniel Dewey for some discussions and feedback.
The goal of this post is to sort through some questions involving the difficulty of avoiding perpetual risk from transformative AI. Feedback in the ... | https://www.lesswrong.com/posts/FfTxEf3uFPsZf9EMP/avoiding-perpetual-risk-from-tai |
# Slightly against aligning with neo-luddites
To summarize,
* When considering whether to delay AI, the choice before us is not merely whether to accelerate or decelerate the technology. We can choose what *type* of regulations are adopted, and some options are much better than others.
* Neo-luddites do not funda... | https://www.lesswrong.com/posts/9MZdwQ7u53oaRiBYX/slightly-against-aligning-with-neo-luddites |
# Air-gapping evaluation and support
*This blog post was written fast to communicate a concept I think is important. I may edit this post for legibility later.*
I think evaluation and support mechanisms should be somewhat “[air-gapped](https://en.wikipedia.org/wiki/Air_gap_(networking)),” or isolated, in their inform... | https://www.lesswrong.com/posts/ehLR9HeXB5TMp9Y4v/air-gapping-evaluation-and-support |
# Can we efficiently distinguish different mechanisms?
(_This post is an elaboration on “tractability of discrimination” as introduced in section III of_ [_Can we efficiently explain model behaviors?_](https://ai-alignment.com/can-we-efficiently-explain-model-behaviors-92b83c2acd5a) _For an overview of the general pla... | https://www.lesswrong.com/posts/JLyWP2Y9LAruR2gi9/can-we-efficiently-distinguish-different-mechanisms |
# Against Agents as an Approach to Aligned Transformative AI
Epistemic Status
----------------
Written (very) quickly[^saflr6cg0l].
* * *
Thesis
------
The creation of superintelligent agents — even aligned ones — would with high probability entail a kind of existential catastrophe[^gfjvryeh4zt].
* * *
Introduct... | https://www.lesswrong.com/posts/xLsyFKsyaagxao9s2/against-agents-as-an-approach-to-aligned-transformative-ai |
# What policies have most thoroughly crippled (otherwise-promising) industries or technologies?
(cropped from an image by DALL-E 2)
In order to seriously consider promoting policies aimed at slowing dow... | https://www.lesswrong.com/posts/CFPjvozs5oLXG6oEw/what-policies-have-most-thoroughly-crippled-otherwise |
# Fractional Resignation
When you're playing games you generally want to spend your time exploring interesting situations, which are typically ones where there are a range of outcomes. If someone is definitely going to win it's not usually much fun to keep going, and in many gaming cultures when you get to this point ... | https://www.lesswrong.com/posts/7h6kbKBEDeKsQxzsW/fractional-resignation |
# MDPs and the Bellman Equation, Intuitively Explained
Imagine you're a selfish taxi driver who takes passengers around between 3 towns. Each trip has a different probability of catastrophe and different earnings potential. You want to find a repeatable path of trips to maximise profit. No, you can’t just greedily pic... | https://www.lesswrong.com/posts/k3vEcwZYoRPrSXaS8/mdps-and-the-bellman-equation-intuitively-explained |
# Reflections on my 5-month alignment upskilling grant
Five months ago, I received a grant from the[ Long Term Future Fund](https://funds.effectivealtruism.org/funds/far-future) to upskill in AI alignment. As of a few days ago, I was invited to Berkeley for two months of full-time alignment research under [Owain Evans... | https://www.lesswrong.com/posts/wnF9iydYiBMRs2jPg/reflections-on-my-5-month-alignment-upskilling-grant |
# Housing and Transportation Roundup #2
Previously: [#1](https://thezvi.substack.com/p/housing-and-transit-thoughts-and)
In [my look back](https://thezvi.substack.com/p/looking-back-on-posts-from-2022) I concluded I want to be more focused and long-term oriented. Roundups by default are the opposite of that. This one... | https://www.lesswrong.com/posts/z73dfgvGdcXxYKDs6/housing-and-transportation-roundup-2 |
# Why The Focus on Expected Utility Maximisers?
Epistemic Status
----------------
Unsure[^d0323hkyj3], partially [noticing my own confusion](https://www.lesswrong.com/s/zpCiuR4T343j9WkcK). Hoping [Cunningham's Law](https://en.wikipedia.org/wiki/Ward_Cunningham#%22Cunningham's_Law%22) can help resolve it.
[Related A... | https://www.lesswrong.com/posts/XYDsYSbBjqgPAgcoQ/why-the-focus-on-expected-utility-maximisers |
# Things that can kill you quickly: What everyone should know about first aid
There are things that kill you instantly, like a bullet to the head or a fall from twenty stories. First aid can’t help you there. There are also things that kill you relatively slowly, like a bacterial infection. If you have even hours to l... | https://www.lesswrong.com/posts/sGnPTfjE5JthAStqg/things-that-can-kill-you-quickly-what-everyone-should-know |
# My Reservations about Discovering Latent Knowledge (Burns, Ye, et al)
*\[This is a cross-post from my blog at* [*aizi.substack.com*](https://aizi.substack.com/)*.\]*
*This is my second post on Burns, Ye, et al’s recent preprint* [*Discovering Latent Knowledge in Language Models Without Supervision*](https://arxiv.o... | https://www.lesswrong.com/posts/C53REQuzSk3TTfqgT/my-reservations-about-discovering-latent-knowledge-burns-ye |
# Fundamental Uncertainty: Chapter 5 - How do we know what we know?
*N.B. This is a chapter in a book about truth and knowledge. It is the first draft. I have since revised it. You can find the most up-to-date info/version on the* [*book's website*](https://www.fundamentaluncertainty.com/)*.*
Alice is walking with he... | https://www.lesswrong.com/posts/CLY5uppzHDmvyt5xp/fundamental-uncertainty-chapter-5-how-do-we-know-what-we |
# What AI Safety Materials Do ML Researchers Find Compelling?
We recently ran a small pilot study in which we showed ML researchers (randomly selected NeurIPS / ICML / ICLR 2021 authors) a number of introductory AI safety materials, asking them to answer questions and rate those materials.
Summary
-------
We selecte... | https://www.lesswrong.com/posts/gpk8dARHBi7Mkmzt9/what-ai-safety-materials-do-ml-researchers-find-compelling |
# Getting up to Speed on the Speed Prior in 2022
*This post was written under the mentorship of Evan Hubinger, as a part of the *[*Stanford Existential Risks Institute ML Alignment Theory Scholars (MATS) program*](https://www.alignmentforum.org/posts/8vLvpxzpc6ntfBWNo/seri-ml-alignment-theory-scholars-program-2022)*.*... | https://www.lesswrong.com/posts/bC5xd7wQCnTDw7Kyx/getting-up-to-speed-on-the-speed-prior-in-2022 |
# In Defense of Wrapper-Minds
Recently, there's been a [strong push](https://www.lesswrong.com/s/nyEFg3AuJpdAozmoX) against "[wrapper-minds](https://www.lesswrong.com/posts/dKTh9Td3KaJ8QW6gw/why-assume-agis-will-optimize-for-fixed-goals)" as a framework. It's argued that there's no specific reason to think that all su... | https://www.lesswrong.com/posts/wdC8fH8kHffYn3kNa/in-defense-of-wrapper-minds |
# 200 Concrete Open Problems in Mechanistic Interpretability: Introduction
### **EDIT 19/7/24**: This sequence is now two years old, and fairly out of date. I hope it's still useful for historical reasons, but I no longer recommend it as a reliable source of problems worth working on, eg it doesn't at all discuss Spar... | https://www.lesswrong.com/posts/LbrPTJ4fmABEdEnLf/200-concrete-open-problems-in-mechanistic-interpretability |
# 200 COP in MI: The Case for Analysing Toy Language Models
*This is the second post in a sequence called 200 Concrete Open Problems in Mechanistic Interpretability. *[*Start here*](https://www.alignmentforum.org/posts/LbrPTJ4fmABEdEnLf/200-concrete-open-problems-in-mechanistic-interpretability)*, then read in any ord... | https://www.lesswrong.com/posts/GWCgZrzWCZCuzGktv/200-cop-in-mi-the-case-for-analysing-toy-language-models |
# Rock-Paper-Scissors Can Be Weird
Let's consider a symmetrical, two-player, zero-sum game, where each player must simultaneously choose one of three moves: A, B, or C. After each player choses their move, a winner is randomly selected with odds based on what the two players selected. (Note that because its symmetrica... | https://www.lesswrong.com/posts/mvy5DazH83Liknzp4/rock-paper-scissors-can-be-weird |
# Large Language Models Suggest a Path to Ems
TL:DR:
- Whole brain emulation from first principles isn't happening
- LLMs are pretty good/human like
- This suggests neural nets are flexible enough to implement human like cognition despite alien architecture used
- Training a (mostly) human shaped neural net on brai... | https://www.lesswrong.com/posts/eNwCXXLzEhWv6aznf/large-language-models-suggest-a-path-to-ems |
# Against John Searle, Gary Marcus, the Chinese Room thought experiment and its world
Preface:
--------
* I’m putting my existing work on AI on Less Wrong, and editing as I go, in preparation to publishing a collection of my works on AI in a free online volume. If this content interests you, you could always follow ... | https://www.lesswrong.com/posts/DPShFqQWMSjWBshLw/against-john-searle-gary-marcus-the-chinese-room-thought |
# Privacy Tradeoffs
There's a lot of privacy-related writing that looks like:
1. Introduce a risk due to insufficient privacy.
2. Describe things you can do to increase your privacy.
For example, I recently came across "[It's time to worry about Internet privacy again](https://malmesbury.substack.com/p/its-time-to... | https://www.lesswrong.com/posts/HCJTSSz3dtKzchqKm/privacy-tradeoffs |
# Where do you get your capabilities from?
In both alignment and capabilities research, people end up discussing the effects and feasibility of different AI architectures, and when this happens, I tend to focus on the question: Where do you get your capabilities from?
[Capabilities are possible because the world has ... | https://www.lesswrong.com/posts/fLHyBpAptikZHi8nB/where-do-you-get-your-capabilities-from |
# Don't feed the void. She is fat enough!
My posts. I write them. I post them. And that's it. In my head, I am not writing them to someone. And I don't expect anything back. There are obvious problems with this model. When I am writing an email to someone, I think very carefully about what to write. When I want to exp... | https://www.lesswrong.com/posts/dBetFzujLck2BJNAB/don-t-feed-the-void-she-is-fat-enough |
# My scorched-earth policy on New Year’s resolutions
(forgive me if these concepts are too basic; this was originally written for normies)
Welp, at this time of year everyone either writes a “How to set and keep your New Year’s resolution” article, or a “Why you shouldn’t set New Year’s resolutions” article.
This is... | https://www.lesswrong.com/posts/SajoAeNYDBmqyCZFK/my-scorched-earth-policy-on-new-year-s-resolutions |
# CFP for Rebellion and Disobedience in AI workshop
Hi everyone!
I'm co-organizing a workshop on a really interesting topic that's very relevant for AI safety. We call it "Rebellion and Disobedience in AI". If you're doing work that could be relevant for us, please submit it! If you have questions or want to discuss... | https://www.lesswrong.com/posts/GyZR24j8buR7D2n8D/cfp-for-rebellion-and-disobedience-in-ai-workshop |
# Internal Interfaces Are a High-Priority Interpretability Target
**tl;dr**: ML models, like all software, and like [the NAH](https://www.lesswrong.com/posts/vDGvHBDuMtcPd8Lks/public-static-what-is-abstraction) would predict, must consist of several specialized "modules". Such modules would form interfaces between eac... | https://www.lesswrong.com/posts/nwLQt4e7bstCyPEXs/internal-interfaces-are-a-high-priority-interpretability |
# Entrepreneurship ETG Might Be Better Than 80k Thought
Summary
=======
1. In 2014, [Ryan Carey at 80k estimated](https://80000hours.org/2014/05/how-much-do-y-combinator-founders-earn/) that YCombinator-backed startup founders averaged $2.5M/year
2. I repeat his analysis, and find that this number is now substantia... | https://www.lesswrong.com/posts/Rggqa2wkWnJCt9tHN/entrepreneurship-etg-might-be-better-than-80k-thought |
# Covid 12/29/22: Next Up is XBB.1.5
A massive waive of cases continues to hit China, as one would expect given last week.
There is also a new variant, XBB.1.5, that is spreading rapidly starting in New York, so it’s time once again for our favorite game show, ‘how much trouble will this new variant cause?’
The Twit... | https://www.lesswrong.com/posts/uKfAJF2DijxcQneHm/covid-12-29-22-next-up-is-xbb-1-5 |
# 200 COP in MI: Looking for Circuits in the Wild
*This is the third post in a sequence called 200 Concrete Open Problems in Mechanistic Interpretability. *[*Start here*](https://www.alignmentforum.org/posts/LbrPTJ4fmABEdEnLf/200-concrete-open-problems-in-mechanistic-interpretability)*, then read in any order. If you ... | https://www.lesswrong.com/posts/XNjRwEX9kxbpzWFWd/200-cop-in-mi-looking-for-circuits-in-the-wild |
# Why and how to write things on the Internet
Recently I noticed that most existing “why you should write a blog” articles ([e.g.](https://sites.google.com/site/steveyegge2/you-should-write-blogs)) have at least one of two shortcomings, according to me:
* They mostly focus on _counterarguments_ to _not_ starting a ... | https://www.lesswrong.com/posts/KHh7txJyECK9AriZq/why-and-how-to-write-things-on-the-internet |
# Why not spend more time looking at human alignment?
As we approach AGI, we also approach the risk of alignment failure: due to either mismatch between intended goals and specified goals ('outer misalignment'), or mismatch between specified goals and emergent goals ('inner misalignment'), we end up with a catastrophi... | https://www.lesswrong.com/posts/bepFsfCx3LiauFtdB/why-not-spend-more-time-looking-at-human-alignment-1 |
# A Year of AI Increasing AI Progress
In July, I made [a post](https://www.lesswrong.com/posts/W3tZacTRt4koHyxbr/examples-of-ai-increasing-ai-progress) about AI being used to increase AI progress, along with [this spreadsheet](http://rsi.thomaswoodside.com/) that I've been updating throughout the year. Since then, I h... | https://www.lesswrong.com/posts/camG6t6SxzfasF42i/a-year-of-ai-increasing-ai-progress |
# But is it really in Rome? An investigation of the ROME model editing technique
*Thanks to Andrei Alexandru, Joe Collman, Michael Einhorn, Kyle McDonell, Daniel Paleka, and Neel Nanda for feedback on drafts and/or conversations which led to useful insights for this work. In addition, thank you to both William Saunder... | https://www.lesswrong.com/posts/QL7J9wmS6W2fWpofd/but-is-it-really-in-rome-an-investigation-of-the-rome-model |
# Announcing The Filan Cabinet
Happy holidays! Some months ago, I launched a new podcast called [The Filan Cabinet](https://thefilancabinet.com/), but forgot to announce the podcast on this blog. Today, I rectify that mistake.
In some ways, the podcast is similar to [AXRP - the AI X-risk Research Podcast](https://axr... | https://www.lesswrong.com/posts/PfmygQyvzcjG48HAK/announcing-the-filan-cabinet |
# Progress links and tweets, 2022-12-29
**Links**
---------
* [City size is determined by transportation technology](https://unchartedterritories.tomaspueyo.com/p/transportation-technology-shapes-cities) (Tomas Pueyo)
[](https://pbs.twimg.com/media/Fkg7L8bVUAAk53... | https://www.lesswrong.com/posts/CepZgbkFpnEWgagjW/progress-links-and-tweets-2022-12-29 |
# Language models are nearly AGIs but we don't notice it because we keep shifting the bar
* I’m putting my existing work on AI on Less Wrong, and editing as I go, in preparation to publishing a collection of my works on AI in a free online volume. If this content interests you, you could always follow my Substack, it'... | https://www.lesswrong.com/posts/mypCA3AzopBhnYB6P/language-models-are-nearly-agis-but-we-don-t-notice-it |
# More ways to spot abysses
I liked Ben Kuhn’s [‘Staring into the abyss as a core life skill’](https://www.benkuhn.net/abyss/).
I’d summarize as:
1. If you are making a major error—professionally, romantically, religiously, etc—it can be hard to look at that fact and correct.
2. However it’s super important. Evide... | https://www.lesswrong.com/posts/gw5iTQwCRMgRt3Rip/more-ways-to-spot-abysses |
# Things I carry almost every day, as of late December 2022
Here we see things I carry in my pockets regularly.
To the left is my phone case, with two bandaids in it. This is a stand-in for the ... | https://www.lesswrong.com/posts/5BAD7Cq2yrrYnyi3x/things-i-carry-almost-every-day-as-of-late-december-2022 |
# Models Don't "Get Reward"
*In terms of content, this has a lot of overlap with* [*Reward is not the optimization target*](https://www.lesswrong.com/posts/pdaGN6pQyQarFHXF4/reward-is-not-the-optimization-target)*. I'm basically rewriting a part of that post in language I personally find clearer, emphasising what I th... | https://www.lesswrong.com/posts/TWorNr22hhYegE4RT/models-don-t-get-reward |
# Worldly Positions archive, briefly with private drafts
I realized it was hard to peruse past [Worldly Positions](https://worldlypositions.tumblr.com) posts without logging in to Tumblr, which seemed pretty bad. So I followed Substack’s instructions to import the archives into [world spirit sock stack](https://worlds... | https://www.lesswrong.com/posts/bwqvNnLF4dpNr9vfz/worldly-positions-archive-briefly-with-private-drafts |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.