text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Covid 8/4/22: Rebound
I’ll be at Future Forum over the weekend. If you’re here, come and say hello. I don’t plan on scheduling a lot of meetings because I hate scheduling, but talking to people is the plan. I have a few people that I know I want to see, but mostly I’m going to be improvising, and whatever happens ha... | https://www.lesswrong.com/posts/ejBK8pJ2yK3d8odqe/covid-8-4-22-rebound |
# Interpretability isn’t Free
*Epistemic status: This post uses a simple and intuitive model*[^dw6bvqsknxr]*, which seems obvious in retrospect, but it hasn’t been formalized and I’m not confident in its conclusions.*
Current consensus is that large language models will be a core component of transformative AI system... | https://www.lesswrong.com/posts/eDg7RDnj7Q9oZxjs6/interpretability-isn-t-free |
# What do ML researchers think about AI in 2022?
_Katja Grace, Aug 4 2022_
AI Impacts just finished collecting data from a new survey of ML researchers, as similar to the [2016 one](https://aiimpacts.org/2016-expert-survey-on-progress-in-ai/) as practical, aside from a couple of new questions that seemed too interest... | https://www.lesswrong.com/posts/H6hMugfY3tDQGfqYL/what-do-ml-researchers-think-about-ai-in-2022 |
# Socratic Ducking, OODA Loops, Frame-by-Frame Debugging
**Socratic Ducking**
--------------------
Occasionally, a software developer will get stuck trying to debug a program, walk over to a colleague’s... | https://www.lesswrong.com/posts/CJGKkTjWLGCwhkjGY/socratic-ducking-ooda-loops-frame-by-frame-debugging |
# Fiber arts, mysterious dodecahedrons, and waiting on “Eureka!”
Part 1: The anomaly
-------------------
This story starts, as many stories do, with my girlfriend 3D-printing me a supernatural artifact. Specifically, one of my favorite SCPs... | https://www.lesswrong.com/posts/FfPukic3Qskd9ZAkk/fiber-arts-mysterious-dodecahedrons-and-waiting-on-eureka |
# Running a Basic Meetup
This is a guide for running a basic meetup.
If you’re interested in running a meetup (ACX, LessWrong, probably even EA) then there exist more detailed guides – I would recommend [Kaj’s](https://www.lesswrong.com/s/CXJTAqQXdEqWNkjqj/p/qMuAazqwJvkvo8teR), but others exist. This is not intended... | https://www.lesswrong.com/posts/veF5t2TEqtuvrgMzY/running-a-basic-meetup |
# The Pragmascope Idea
Pragma (Greek): thing, object.
A “pragmascope”, then, would be some kind of measurement or visualization device which shows the “things” or “objects” present.
I currently see the pragmascope as *the* major practical objective of [work on natural abstractions](https://www.lesswrong.com/posts/cy... | https://www.lesswrong.com/posts/gdEDPHjCY5DKsMsvE/the-pragmascope-idea |
# Calibration Trivia
**Summary:** A game of trivia where you answer factual questions about the world, but stating how sure you are that you’re right and trying to be well calibrated.
**Tags:** Large, Repeatable
**Purpose:** Calibration Trivia is designed to practice proper calibration – recognizing when you're ve... | https://www.lesswrong.com/posts/2RXNmrSLEphNXMwKQ/calibration-trivia |
# Cambist Booking
**Summary:** A [Cambist](https://en.wiktionary.org/wiki/cambist) is a manual showing exchange rates of different currencies and measurements. [Dutch Booking](https://en.wikipedia.org/wiki/Dutch_book) is when a sequence of trades would leave one party strictly worse off and another strictly better off... | https://www.lesswrong.com/posts/fXMWri2XgMFNf2yBP/cambist-booking |
# Convergence Towards World-Models: A Gears-Level Model
1\. Intuitions
--------------
One of the more solid results in agency theory is the [generalized definition of power-seeking](https://www.lesswrong.com/s/fSMbebQyR4wheRrvk/p/6DuJxY8X45Sco4bS2). Power-seeking is the tendency of agents to move towards states with ... | https://www.lesswrong.com/posts/HzSdYWvdrdQqG9tqW/convergence-towards-world-models-a-gears-level-model |
# The Falling Drill
**Summary:** The Falling Drill is intended to practice being wrong in a comfortable practice environment, intended to help us handle being wrong out in the world easier and with less stress.
**Tags:** Small, repeatable
**Purpose:** The first thing you learn in martial arts is the ability to fal... | https://www.lesswrong.com/posts/BdAfJ3xnonTQZMxtb/the-falling-drill |
# Two Kids Crosswise
When traveling, we sometimes end up somewhere with the older two kids sharing a double bed. Initially we would put them down the normal way, as the bed was made to be used, but recently we realized it goes much better if you think of the bed as being sideways:
[, [Abram](https://www.alignmentforum.org/users/abramdemski), [Josiah](https://josiahlopezwild.com/), [Daniel](https://www.danielherrmann.ca/)) posts on ... | https://www.lesswrong.com/posts/rQDYQrDjPGqjrf8Mk/bridging-expected-utility-maximization-and-optimization |
# Where are the red lines for AI?
*Thanks to Daniel Kokotajlo, Jan Hendrik Kirchner, Remmelt Ellen, Berbank Green, Otto Barten and Olaf Voß for helpful suggestions and comments.*
As [AI alignment remains terribly difficult](https://www.lesswrong.com/posts/3pinFH3jerMzAvmza/on-how-various-plans-miss-the-hard-bits-of-t... | https://www.lesswrong.com/posts/eStLg3uhHmzjCqWDm/where-are-the-red-lines-for-ai |
# COVID-19 Group Testing Post-mortem?
As Covid is for the most part over and tests have not been in short supply for a very long time now, eliminating most of the rationale for complicated testing procedures meant to optimize test count/cost, it would be interesting for anyone who was involved in this to do a retrospe... | https://www.lesswrong.com/posts/A99DoJDf26PA3v3yH/covid-19-group-testing-post-mortem |
# Gears-Level Understanding, Deliberate Performance, The Strategic Level
**Gears-Level Understanding**
-----------------------------
It seems to be important to distinguish between two kinds of knowing: the knowing that comes from listening to trusted sources, and the knowing that comes from seeing why the world coul... | https://www.lesswrong.com/posts/T8piFGywHFd4ax9yx/gears-level-understanding-deliberate-performance-the |
# Counterfactuals are Confusing because of an Ontological Shift
I recently became [unstuck](https://www.lesswrong.com/posts/QbLQpqojpwEGnf8bc/getting-unstuck-on-counterfactuals) on counterfactuals. I now believe that counterfactuals are confusing, in large part, because they entail preserving our values through an ont... | https://www.lesswrong.com/posts/MiWBC2A2cADspausG/counterfactuals-are-confusing-because-of-an-ontological |
# Rant on Problem Factorization for Alignment
This post is the second in what is likely to become a series of uncharitable rants about alignment proposals (previously: [Godzilla Strategies](https://www.lesswrong.com/posts/DwqgLXn5qYC7GqExF/godzilla-strategies)). In general, these posts are intended to convey my underl... | https://www.lesswrong.com/posts/tmuFmHuyb4eWmPXz8/rant-on-problem-factorization-for-alignment |
# "Just hiring people" is sometimes still actually possible
Graham's design paradox is a maxim about organizations that says that if key management isn't competent at ${X}, it's often impossible for an organization to hire people who are, regardless of how well resourced that organization is. It was once described thu... | https://www.lesswrong.com/posts/jNnpv4xnFii4xkN2g/just-hiring-people-is-sometimes-still-actually-possible |
# «Boundaries», Part 2: trends in EA's handling of boundaries
*This is Part 2 of my* [*«Boundaries» Sequence*](https://www.lesswrong.com/s/LWJsgNYE8wzv49yEc) *on LessWrong, and is also available on the* [*EA Forum*](https://forum.effectivealtruism.org/posts/QFBLbfxgb2XK6tRcv/boundaries-part-2-trends-in-ea-s-handling-o... | https://www.lesswrong.com/posts/vnJ5grhqhBmPTQCQh/boundaries-part-2-trends-in-ea-s-handling-of-boundaries |
# Announcing the Introduction to ML Safety course
TLDR
====
We're announcing a new course designed to introduce students with a background in machine learning to the most relevant concepts in empirical ML-based AI safety. The course is available publicly [here](https://course.mlsafety.org/).
Background
==========
A... | https://www.lesswrong.com/posts/4F8Bg8Z5cePTBofzo/announcing-the-introduction-to-ml-safety-course |
# Metaculus and medians
_Cross-posted from [my personal blog](https://blog.rossry.net/metaculus-median/)._
## (1)
Should I expect monkeypox to be a big deal for the world? Well, fortunately, [Metaculus](https://www.metaculus.com/) has a pair of questions that ask users to predict how many infections and deaths there... | https://www.lesswrong.com/posts/TFFvLyfFiSCz88opm/metaculus-and-medians |
# PredictIt is closing due to CFTC changing its mind
[https://www.predictit.org/platform-announcements](https://www.predictit.org/platform-announcements)
> The staff of the Commodity Futures Trading Commission (CFTC) has [withdrawn](https://www.cftc.gov/PressRoom/PressReleases/8567-22) the No-Action letter (NAL) issu... | https://www.lesswrong.com/posts/HK768onceku2vvnwn/predictit-is-closing-due-to-cftc-changing-its-mind |
# My advice on finding your own path
tl;dr - 4 steps
---------------
1. Give yourself permission to take the future seriously
2. Be willing to imagine successful or ambitious outcomes
3. Give yourself the gift of focused time and space
4. Write draw sketch diagram or whatever you need to empty your mind
Intro
--... | https://www.lesswrong.com/posts/X7Gr6EBpQ8jprqkgh/my-advice-on-finding-your-own-path |
# Why I Am Skeptical of AI Regulation as an X-Risk Mitigation Strategy
*Epistemic Signpost: I'm not an expert in policy research, but I think most of the points here are straightforward.*
1\. Regulations are ineffective at preventing bad behaviors
-----------------------------------------------------------
Here's a ... | https://www.lesswrong.com/posts/zQnzhGLDp2PSvAjcW/why-i-am-skeptical-of-ai-regulation-as-an-x-risk-mitigation |
# Six weeks doesn’t make a habit
I occasionally get asked some version of “How long do I need to practice my habit before it will stick?” The coachee has often heard some magic number – 21 days, 6 weeks, 9 weeks – that will make a new habit automatic and robust. (I hear surprising different numbers, giving how simila... | https://www.lesswrong.com/posts/X7HmDvwgwCJ58HfzB/six-weeks-doesn-t-make-a-habit |
# A Data limited future
This is a prediction of what I think the near future might be like.
Suppose the trends in scaling laws roughly continue. Deep learning can do anything if it has enough data. But getting that data is hard.
So large language models get better, but not that much better. They are using most of th... | https://www.lesswrong.com/posts/gqqhYijxcKAtuAFjL/a-data-limited-future |
# Dwarves & D.Sci: Data Fortress
*This is an entry in the 'Dungeons & Data Science' series, a set of puzzles where players are given a dataset to analyze and an objective to pursue using information from that dataset. *
**STORY (skippable)**
=====================
You stare out the window of your office in the dwarve... | https://www.lesswrong.com/posts/wmG6kdcacNoB3HWJs/dwarves-and-d-sci-data-fortress |
# Boston Rents Over Time II
Since 2011, and monthly since 2013, I've maintained a [map of Boston-area rents](https://www.jefftk.com/apartment_prices), scraping [Padmapper](https://www.padmapper.com/apartments/boston-ma). I looked at how they were changing over time in [in 2015](https://www.jefftk.com/p/boston-rents-ov... | https://www.lesswrong.com/posts/Sn6oJ3irvS28j6DXr/boston-rents-over-time-ii |
# [AMA] Announcing Open Phil’s University Group Organizer and Century Fellowships [x-post]
\[Crossposted from the EA Forum. Particularly relevantly to LessWrong, I'm interested in funding more rationality groups at universities.\]
Open Philanthropy recently launched two new fellowships intended to provide funding... | https://www.lesswrong.com/posts/7mc7NMoa77ogNZXpY/ama-announcing-open-phil-s-university-group-organizer-and |
# Newcombness of the Dining Philosophers Problem
Before finding this site, I didn't normally associate the Prisoner's Dilemma or other questions of Decision Theory with the field of Artificial Intelligence. After all, when writing software I don't have to negotiate to get the program to cooperate with me; rather, it o... | https://www.lesswrong.com/posts/jiAjwzeSvotv9wcJo/newcombness-of-the-dining-philosophers-problem |
# Most Ivy-smart students aren't at Ivy-tier schools
\[I posted this [on the EA Forum](https://forum.effectivealtruism.org/posts/czjZgTBFxPtDnPTBv/most-ivy-smart-students-aren-t-at-ivy-tier-schools) as well, and may as well leave the following note in :)\]
### *Note for the EA Forum *
I've decided to share this her... | https://www.lesswrong.com/posts/QfPSm2dug9NfBeFXd/most-ivy-smart-students-aren-t-at-ivy-tier-schools |
# Paper reading as a Cargo Cult
I have come across various people (including my past self) who meet up regularly to study, e.g., alignment forum posts and discuss them. This helps people bond over their common believes, fears, and interests, which I think is good, but *in no way is this ever going to lead anyone to fi... | https://www.lesswrong.com/posts/EhapyH46hDec3JvT4/paper-reading-as-a-cargo-cult |
# Lamentations, Gaza and Empathy
# Lamentations
Today is the fast of Av, when Jews mourn the destruction of their temples in Jerusalem and their exile. We fast, sit on low stools, and read the book of lamentations (Eichah).
This book, traditionally ascribed to the prophet Jeremiah, consists of 5 poems mourning the d... | https://www.lesswrong.com/posts/aZbdMHqTKH5RaFN26/lamentations-gaza-and-empathy |
# Jack Clark on the realities of AI policy
Some choice picks:
> The real danger in Western AI policy isn't that AI is doing bad stuff, it's that governments are so unfathomably behind the frontier that they have no notion of \_how\_ to regulate, and it's unclear if they \_can\_
>
> Many AI policy teams in industry a... | https://www.lesswrong.com/posts/R3tXGhSCgYbp3kXm2/jack-clark-on-the-realities-of-ai-policy |
# Careful with Caching
A friend recently shared this graphic on Facebook:
[](https://www.jefftk.com/prison-vs-college-wrong-big.jpg)
(I've added the "wrong" overlay.)
This is clearly incorrect: there's no way Massachusetts has more prisons than colleges. (MA ac... | https://www.lesswrong.com/posts/EohTT3htbgQi2vtQP/careful-with-caching |
# The lessons of Xanadu
One of my all-time favorite articles is “[The Curse of Xanadu](https://www.wired.com/1995/06/xanadu/),” by Gary Wolf, which ran in WIRED Magazine in 1995. On the surface, it’s a piece of tech history, a story of a dramatic failure. But look closer, and you can find deep philosophical insight.
... | https://www.lesswrong.com/posts/wr9dH2GjztvCz6pYX/the-lessons-of-xanadu |
# Complexity No Bar to AI (Or, why Computational Complexity matters less than you think for real life problems)
Basically an article about why knowing a problem is in a certain complexity class does not mean AGI can't do that problem. It's some evidence, but usually isn't enough unless 2-3 or more assumptions are righ... | https://www.lesswrong.com/posts/S7K8Pmuh768YiYooW/complexity-no-bar-to-ai-or-why-computational-complexity |
# Can we get full audio for Eliezer's conversation with Sam Harris?
I listened to [this full interview](https://www.samharris.org/podcasts/making-sense-episodes/116-ai-racing-toward-brink) in 2018, and it made me take AI risk way more seriously. I now plan to devote my career to this problem, so people often ask me wh... | https://www.lesswrong.com/posts/Swsy9fK7Gpf3t8Ewm/can-we-get-full-audio-for-eliezer-s-conversation-with-sam |
# How Deadly Will Roughly-Human-Level AGI Be?
> Which is not to say that recursive self-improvement happens before the end of the world; if the first AGI's mind is sufficiently complex and kludgy, it’s entirely possible that the cognitions it implements are able to (e.g.) crack nanotech well enough to kill all humans,... | https://www.lesswrong.com/posts/bvRFtgGymqLRQoj5h/how-deadly-will-roughly-human-level-agi-be |
# Steganography in Chain of Thought Reasoning
Here I give a possible phenomenon of steganography in chain of thought reasoning, where a system doing multi-stage reasoning with natural language encodes hidden information in its outputs that is not observable by humans, but can be used to boost its performance on some t... | https://www.lesswrong.com/posts/yDcMDJeSck7SuBs24/steganography-in-chain-of-thought-reasoning |
# Area under the curve, Eat Dirt, Broccoli Errors, Copernicus & Chaos
**Area Under The Curve**
------------------------
People often try to *maximize*.
They’ll track their productivity, and try to get the number as high as possible. They’ll look at the amount of weight they can lift, and try to push the envelope. Th... | https://www.lesswrong.com/posts/ZCXjmeHNHYiM2uCw4/area-under-the-curve-eat-dirt-broccoli-errors-copernicus-and |
# A sufficiently paranoid paperclip maximizer
*A hypothetical scenario of a non-friendly but sufficiently paranoid AGI saving humanity.*
* * *
The devil was laughing as the last of mankind showed up in his dominion. Then he stopped, because after them came some kind of machines that began to dismantle Hell and t... | https://www.lesswrong.com/posts/A3sJeuKNAQdCm7Ji7/a-sufficiently-paranoid-paperclip-maximizer |
# Interpretability/Tool-ness/Alignment/Corrigibility are not Composable
Interpretability
----------------
I have a decent understanding of how transistors work, at least for purposes of basic digital circuitry. Apply high voltage to the gate, and current can flow between source and drain. Apply low voltage, and curre... | https://www.lesswrong.com/posts/qXtbBAxmFkAQLQEJE/interpretability-tool-ness-alignment-corrigibility-are-not |
# Broad Basins and Data Compression
*Produced As Part Of The SERI ML Alignment Theory Scholars Program 2022 Research Sprint Under* [*John Wentworth*](https://www.lesswrong.com/users/johnswentworth)
**Intro**
---------
The "broad basins induce data compression" argument goes like this: If we train a neural network to... | https://www.lesswrong.com/posts/ybLaw7EuMBeM2dLn9/broad-basins-and-data-compression |
# Encultured AI Pre-planning, Part 1:
Enabling New Benchmarks
*Also available on the* [*EA Forum*](https://forum.effectivealtruism.org/posts/yczkGfcfWoRN6zfrf/encultured-ai-part-1-enabling-new-benchmarks)*.*
*Followed by:* [*Encultured AI, Part 2*](https://www.lesswrong.com/posts/2vxoTfuScspraSJeC/encultured-ai-par... | https://www.lesswrong.com/posts/AR6mfydDJiGksj6Co/encultured-ai-pre-planning-part-1-enabling-new-benchmarks |
# Encultured AI, Part 1 Appendix: Relevant Research Examples
*Also available on the EA Forum.*
*Appendix to:* [*Encultured AI, Part 1: Enabling New Benchmarks*](https://www.lesswrong.com/posts/AR6mfydDJiGksj6Co/encultured-ai-part-1-enabling-new-benchmarks)
*Followed by: Encultured AI, Part 2: Providing a Service*
... | https://www.lesswrong.com/posts/PvuuBN39pmjw6wRpj/encultured-ai-part-1-appendix-relevant-research-examples |
# Experiment: Be my math tutor?
I'm a CS student at a top university who takes AI x-risk incredibly seriously. While university has given me a lot of fun opportunities to grow, I am not entirely satisfied with the rate at which my skill is improving.
If I want to stop human extinction, I'll need to \[go faster\](htt... | https://www.lesswrong.com/posts/EyNiWngTA2RNdnXnt/experiment-be-my-math-tutor |
# General alignment properties
[AIXI](https://en.wikipedia.org/wiki/AIXI) and the genome are both ways of specifying intelligent agents.
1. Give AIXI a utility function (perhaps over observation histories), and hook it up to an environment, and this pins down a policy.[^iggswxs0y97]
2. Situate the genome in the e... | https://www.lesswrong.com/posts/FMdGt9S9irgxeD9Xz/general-alignment-properties |
# Are ya winning, son?
Hello,
I've been reading through various Newcomblike Problems, in order to get a better understanding of the differences between each Decision Theory. From what I can tell, it seems like each Decision Theory gets evaluated based on whether they are able to "win" in each of these thought experim... | https://www.lesswrong.com/posts/2KvjY6HZ64QXDE2T6/are-ya-winning-son |
# ACX Montreal Meetup August 21 2022 **Venue Changed**
**Venue: **
Due to projected rain, **the venue has been changed to to the nearby** [**Olde Orchard pub & grill**](https://www.google.com/maps/place/Ye+Olde+Orchard+Pub+%26+Grill/@45.5134983,-73.5727146,17z/data=!3m1!4b1!4m5!3m4!1s0x4cc91a35812fae03:0xf5fdb6658701... | https://www.lesswrong.com/events/hvbkcvGcdYCSzwJbz/acx-montreal-meetup-august-21-2022-venue-changed |
# How (not) to choose a research project
Background
==========
*(specific information will be sparse here. This is meant to give context for the Takeaways section of the post)*
Our group (Garrett, Chu, and Johannes) have worked with John Wentworth in the [SERI MATS](https://www.serimats.org/) ~~2 Electric Boogaloo~~... | https://www.lesswrong.com/posts/kDsywodAKgQAAAxE8/how-not-to-choose-a-research-project |
# Project proposal: Testing the IBP definition of agent
Context
-------
Our team in SERI MATS needs to choose a project to work on for the next month. We spent the first two weeks discussing the alignment problem and what makes it difficult, and proposing (lots of) projects to look for one that we think would direct... | https://www.lesswrong.com/posts/LWchvJxSGPKQwueq2/project-proposal-testing-the-ibp-definition-of-agent |
# What are some Works that might be useful but are difficult, so forgotten?
It seems like the internet in general, and social media broadly, e.g. LessWrong, has this problem that there's a lot of data. The data has to be filtered somehow. The main way data is filtered is by (1) recency, (2) backlinks, and (3) upvo... | https://www.lesswrong.com/posts/DtW3sLuS6DaYqJyis/what-are-some-works-that-might-be-useful-but-are-difficult |
# Announcing: Mechanism Design for AI Safety - Reading Group
We’re starting a new reading group for people interested in applying mechanism design tools to technical AI alignment. If you’re interested in joining, you can [apply here](https://docs.google.com/forms/d/1nLMqCfS99dvLWEmGhWUNRFWLN4u4VbWJbneLBL2uJjg/edit) by... | https://www.lesswrong.com/posts/FhqZZFydyQG9WTSKR/announcing-mechanism-design-for-ai-safety-reading-group |
# Team Shard Status Report
*Team Shard is a nebulous alignment research collective, on paper siloed under John Wentworth's SERI MATS program, but in reality extending its many tendrils far across the Berkeley alignment community. "Shard theory" -- a name spoken of in hushed, mildly confused tones at many an EA hangout... | https://www.lesswrong.com/posts/5F6vzHanbhNRTus2T/team-shard-status-report |
# Radio Bostrom: Audio narrations of papers by Nick Bostrom
**We've started a project to record audio narrations of the most important papers by Nick Bostrom.**
The recordings will be added to nickbostrom.com. They will also be distributed as a podcast with its own [website](https://radiobostrom.com/), brand and [Y... | https://www.lesswrong.com/posts/PEmBENgH5DafkHkNB/radio-bostrom-audio-narrations-of-papers-by-nick-bostrom |
# What are some alternatives to Shapley values which drop additivity?
I originally asked this on math.stackexchange; after reading Diffractor's Unifying Bargaining sequence ([Part 1 here](https://www.lesswrong.com/posts/rYDas2DDGGDRc8gGB/unifying-bargaining-notions-1-2)) I'm wondering if there are more insights floati... | https://www.lesswrong.com/posts/FGjDDfAAcbrYHDfFC/what-are-some-alternatives-to-shapley-values-which-drop |
# Progress links and tweets, 2022-08-09
**Links**
---------
* [The challenge of corrosion and how engineers fight it](https://youtu.be/2RbiCOFffRs) (YouTube)
* [Arnold Kling claims that we *over*-consume medical services in the US](https://arnoldkling.substack.com/p/crisis-of-abundance-81)
* [The lessons of Xan... | https://www.lesswrong.com/posts/xtjyKyeLjMxLoNsY4/progress-links-and-tweets-2022-08-09 |
# Disagreements about Alignment: Why, and how, we should try to solve them
Tldr
====
1. AI Alignment does not follow the standard scientific method.
2. Therefore, we might not agree on which research agendas have the potential to align AGI systems.
3. We should work intentionally to build consensus around the foun... | https://www.lesswrong.com/posts/ALDnrFBhsGdogkm4o/disagreements-about-alignment-why-and-how-we-should-try-to |
# Proposal: Consider not using distance-direction-dimension words in abstract discussions
*Epistemic Status: Less Wrong Classic™, a bold and weird proposal to possibly improve your rationality a little bit, maybe.*
Overview
--------
It is common to use words and phrases implying distance, directionality, and dimensi... | https://www.lesswrong.com/posts/BrjuYAJhJpQfKi7cz/proposal-consider-not-using-distance-direction-dimension |
# The Medium Is The Bandage
A man sets himself the task of portraying the world. Through the years he peoples a space with images of provinces, kingdoms, mountains, bays, ships, islands, fishes, rooms, instruments, stars, horses, and people. Shortly before his death, he discovers that that patient labyrinth of lin... | https://www.lesswrong.com/posts/syCvQM38GarFKBWdG/the-medium-is-the-bandage |
# Dissent Collusion
**Summary**: Take turns being confronted with a group that’s presenting a united front, and grow comfortable agreeing or disagreeing with them.
**Tags**: Medium, Repeatable
**Purpose**: Working with groups involves balancing between agreeing with the group consensus when the group is wrong and di... | https://www.lesswrong.com/posts/q7ZHzPpCsFFq4LpkC/dissent-collusion |
# Using GPT-3 to augment human intelligence
A blog post is a very long and complex search query to find fascinating people and make them route interesting stuff to your inbox.
It is like summoning an alien intelligence. I post at night, before crawling into bed next to my kids, and wake in the morning to find my inbo... | https://www.lesswrong.com/posts/spBoxzcaCrqXqyQHq/using-gpt-3-to-augment-human-intelligence |
# How To Go From Interpretability To Alignment: Just Retarget The Search
\[EDIT: Many people who read this post were very confused about some things, which I later explained in [What’s General-Purpose Search, And Why Might We Expect To See It In Trained ML Systems?](https://www.lesswrong.com/posts/6mysMAqvo9giHC4iX/wh... | https://www.lesswrong.com/posts/w4aeAFzSAguvqA5qu/how-to-go-from-interpretability-to-alignment-just-retarget |
# Emergent Abilities of Large Language Models [Linkpost]
I've [argued before](https://forum.effectivealtruism.org/posts/7ZZpWPq5iqkLMmt25/aogara-s-shortform?commentId=NKwpLDfJ4PYZkmtxs) against the view that intelligence is a single coherent concept, and that AI will someday suddenly cross the threshold of general int... | https://www.lesswrong.com/posts/kNAiwLrtn8DcBxGjR/emergent-abilities-of-large-language-models-linkpost |
# How Do We Align an AGI Without Getting Socially Engineered?
(Hint: Box It)
### *Produced during the Stanford Existential Risk Initiative (SERI) ML Alignment Theory Scholars (MATS) Program of 2022, under John Wentworth*
*“Overconfidence in yourself is a swift way to defeat.”*
*\- Sun Tzu*
**TL;DR:** Escape into t... | https://www.lesswrong.com/posts/p62bkNAciLsv6WFnR/how-do-we-align-an-agi-without-getting-socially-engineered |
# How much alignment data will we need in the long run?
This question stands out to me because:
* It should directly affect empirical alignment priorities today
* While it is informed by both theoretical and empirical evidence, it seems tractable for purely theoretical alignment researchers to make progress on to... | https://www.lesswrong.com/posts/qoz2ryN4GDqEWPBnQ/how-much-alignment-data-will-we-need-in-the-long-run-1 |
# The alignment problem from a deep learning perspective
*This report (*[*now available on arxiv*](https://arxiv.org/abs/2209.00626)*) is intended as a concise introduction to the alignment problem for people familiar with machine learning. It translates previous arguments about misalignment into the context of deep l... | https://www.lesswrong.com/posts/KbyRPCAsWv5GtfrbG/the-alignment-problem-from-a-deep-learning-perspective |
# Shard Theory: An Overview
*Generated as part of SERI MATS, Team Shard's research, under John Wentworth.*
*Many thanks to Quintin Pope, Alex Turner, Charles Foster, Steve Byrnes, and Logan Smith for feedback, and to everyone else I've discussed this with recently! All mistakes are my own.*
.
Then I tried to read the paper he talked about, [Every Good Regulator Of A System Must Be... | https://www.lesswrong.com/posts/wTJECdFSAk4843mFj/thoughts-on-the-good-regulator-theorem |
# Covid 8/11/22: The End Is Never The End
On Tuesday morning as I was grabbing croissants, the man in front of me turned around and told me to back away. ‘Because of the pandemic.’ He was the only one in the bakery wearing a mask, yet in every way he gave off the clear vibe that I, not he, was the one being unreasonab... | https://www.lesswrong.com/posts/zGgRknKj7p3zPbado/covid-8-11-22-the-end-is-never-the-end |
# Pendulums, Policy-Level Decisionmaking, Saving State
**Pendulums**
-------------
> "Remember back when we used to talk about pendulums *too little?"*
One excellent, quick-and-dirty models of social change is the mental image of a pendulum swinging back and forth around a central resting point.
Imagine that the... | https://www.lesswrong.com/posts/8FsLyk6oDrhoqWEjB/pendulums-policy-level-decisionmaking-saving-state |
# Introducing Pastcasting: A tool for forecasting practice
TL;DR: Visit [pastcasting.com](https://www.pastcasting.com/) to forecast on already resolved questions you don’t have prior knowledge about to get quick feedback on how you’re doing.
We may try to coordinate multiplayer sessions in our [Discord](https://disco... | https://www.lesswrong.com/posts/FuGHKKMcAL2GYbxrn/introducing-pastcasting-a-tool-for-forecasting-practice |
# Language models seem to be much better than humans at next-token prediction
\[Thanks to a variety of people for comments and assistance (especially Paul Christiano, Nostalgebraist, and Rafe Kennedy), and to various people for playing the game. Buck wrote the top-1 prediction web app; Fabien wrote the code for the pe... | https://www.lesswrong.com/posts/htrZrxduciZ5QaCjw/language-models-seem-to-be-much-better-than-humans-at-next |
# My summary of the alignment problem
I've been looking into the AI alignment problem last couple of days and came up with the following summary of what problems there are and why. Also, I'd prefer using the umbrella name of Human alignment problem, as AI alignment is just a subset of it.
* The problem is that we d... | https://www.lesswrong.com/posts/Db9XuHGpSrwCzXn9Q/my-summary-of-the-alignment-problem |
# Encultured AI Pre-planning, Part 2:
Providing a Service
*Also available on the* [*EA Forum*](https://forum.effectivealtruism.org/posts/MWWZQ8C655iT9zzRd/encultured-ai-part-2-providing-a-service)*.*
*Preceded by:* [*Encultured AI Pre-planning, Part 1: Enabling New Benchmarks*](https://www.lesswrong.com/posts/AR6mf... | https://www.lesswrong.com/posts/2vxoTfuScspraSJeC/encultured-ai-pre-planning-part-2-providing-a-service |
# Seriously, what goes wrong with "reward the agent when it makes you smile"?
Suppose you're training a huge neural network with some awesome future RL algorithm with clever exploration bonuses and a self-supervised pretrained multimodal initialization and a recurrent state. This NN implements an embodied agent which ... | https://www.lesswrong.com/posts/22xf8GmwqGzHbiuLg/seriously-what-goes-wrong-with-reward-the-agent-when-it |
# Troll Timers
**Summary**: A modification to any two player board game, where you play on a very fast clock with infrequent chances to pause and think things through more carefully.
**Tags**: Small, repeatable
**Purpose**: Two common flaws in thinking that both relate to time management. Sometimes we spend too long... | https://www.lesswrong.com/posts/fCg3pLZqthXsGznHP/troll-timers |
# Seeking PCK (Pedagogical Content Knowledge)
*Author's note: This was originally included in the flash class section and has been broken out on advice from readers, to be placed at the front of the handbook. It was usually a full class at the start of the workshop, and was in the flash class section merely because n... | https://www.lesswrong.com/posts/CsNtMunxsZnvumYLe/seeking-pck-pedagogical-content-knowledge |
# What are some good arguments against building new nuclear power plants?
It seems there is a consensus in the EA community that building new nuclear power plants is a net positive for the world. For example, [this video](https://www.youtube.com/watch?v=EhAemz1v7dQ) by Kurzgesagt summarizes some arguments in support o... | https://www.lesswrong.com/posts/3aBdJkGobmaymtLyF/what-are-some-good-arguments-against-building-new-nuclear |
# Anti-squatted AI x-risk domains index
I heard about a recent case of a relevant domain being squatted, and figured that the people who buy up domains then sell them at a markup might start to notice the growth of interest in AI x-risk and claim some domains which would otherwise host good projects. I've bought the f... | https://www.lesswrong.com/posts/q5xoDf5aHjtsa2QAf/anti-squatted-ai-x-risk-domains-index |
# Argument by Intellectual Ordeal
There once was an extremely annoying child named Jacob. One time during recess, Jacob decided to take up the entire playground bench by lying down on it horizontally. A classmate came up to him and said something like: "You're taking up the whole bench! That's not fair!" Jacob respond... | https://www.lesswrong.com/posts/B6KP9mB746MkSAHSx/argument-by-intellectual-ordeal |
# Refining the Sharp Left Turn threat model, part 1: claims and mechanisms
This is our current distillation of the [sharp left turn](https://www.lesswrong.com/posts/GNhMPAWcfBCASy8e6/a-central-ai-alignment-problem-capabilities-generalization) threat model and an attempt to make it more concrete. We will discuss our un... | https://www.lesswrong.com/posts/usKXS5jGDzjwqv3FJ/refining-the-sharp-left-turn-threat-model-part-1-claims-and |
# Timelines explanation post part 1 of ?
> "Correct prediction isn't enough. You must follow that insight with effective action that actually makes a difference"
\- paraphrase of Holden Karnofsky from recent 80 thousand hours podcast
see also: [Timelines post 2](https://www.lesswrong.com/posts/wgcFStYwacRB8y3Y... | https://www.lesswrong.com/posts/jQXu9tefKfRroegrb/timelines-explanation-post-part-1-of |
# Oversight Misses 100% of Thoughts The AI Does Not Think
Problem: an overseer won’t see the AI which kills us all thinking about how to kill humans, not because the AI conceals that thought, but because the AI doesn’t think about how to kill humans in the first place. The AI just kills humans as a side effect of what... | https://www.lesswrong.com/posts/98c5WMDb3iKdzD4tM/oversight-misses-100-of-thoughts-the-ai-does-not-think |
# What is estimational programming? Squiggle in context
*This post does not contain the canonical QURI opinion, but I am a contractor there. I want to thank early commenters Vivian Belenky and Ozzie Gooen. The section concerning probabilistic programming is readily skippable to make things easier for nonprogrammers.*
... | https://www.lesswrong.com/posts/qg4yy4je3sFm7Z49L/what-is-estimational-programming-squiggle-in-context |
# Dissolve: The Petty Crimes of Blaise Pascal
> VLADIMIR: \[H\]e'd have to think it over.
>
> ESTRAGON: In the quiet of his home.
>
> VLADIMIR: Consult his family.
>
> ESTRAGON: His friends.
>
> VLADIMIR: His agents.
>
> ESTRAGON: His correspondents.
>
> VLADIMIR: His books.
>
> ESTRAGON: His bank accoun... | https://www.lesswrong.com/posts/hf6tH3HJoNeiYdhwi/dissolve-the-petty-crimes-of-blaise-pascal |
# DeepMind alignment team opinions on AGI ruin arguments
We had some discussions of the [AGI ruin arguments](https://www.lesswrong.com/posts/uMQ3cqWDPHhjtiesc/agi-ruin-a-list-of-lethalities) within the DeepMind alignment team to clarify for ourselves which of these arguments we are most concerned about and what the im... | https://www.lesswrong.com/posts/qJgz2YapqpFEDTLKn/deepmind-alignment-team-opinions-on-agi-ruin-arguments |
# How to bet against civilizational adequacy?
Coal prices are at historical highs (2x to 4x normal prices depending on the kind of coal), but coal miner stocks are not. They're trading at historically low multiples, around 1x-2x spot FCF, meaning they can make their enterprise value in less than 2 years worth of after... | https://www.lesswrong.com/posts/xRqczuoYQuLPN82iZ/how-to-bet-against-civilizational-adequacy |
# Gradient descent doesn't select for inner search
**TL;DR:** Gradient descent won't select for inner search processes because they're not compute & memory efficient.
**Slightly longer TL;DR**: A key argument for mesa-optimization is that as we search over programs, we will select for "search processes with simple ob... | https://www.lesswrong.com/posts/TdesHi8kkyokQdDoQ/gradient-descent-doesn-t-select-for-inner-search |
# Building a Bugs List prompts
**Prompt 0:**
Think about the way computer programmers talk about “bugs” in the program, or “feature requests” that would make a given app or game much better. Bugs are things-that-are-bad: frustrations, irritations, frictions, problems. Feature requests are things-that-could-be-great... | https://www.lesswrong.com/posts/LaYB3CPMnDNzh2via/building-a-bugs-list-prompts |
# Appendix: Hamming Questions
*Author's note: these questions were usually the prework for an activity called "Hamming Circles," in which four or five people sat down and spent 20-30 minutes with each person's Hamming Question at the center of a slow, intimate discussion. The Hamming Circles activity is not fully des... | https://www.lesswrong.com/posts/RafYaSYQ2uLmYgwhk/appendix-hamming-questions |
# Appendix: Jargon Dictionary
**80-20**
To “80-20” something is to obtain most of the result (80%) with only a small proportion of the work (20%). This expression originates with the Pareto principle which states that for many events, 80% of the effects come from 20% of the causes (e.g. most of a com- pany’s sales co... | https://www.lesswrong.com/posts/fbv9FWss6ScDMJiAx/appendix-jargon-dictionary |
# Steelmining via Analogy
*This post has been written for the first* [*Refine*](https://www.lesswrong.com/posts/D7epkkJb3CqDTYgX9/refine-an-incubator-for-conceptual-alignment-research-bets) *blog post day, at the end of a week of readings, discussions, and exercises about epistemology for doing good conceptual researc... | https://www.lesswrong.com/posts/MfCDfuBHXL5ijJFco/steelmining-via-analogy |
# How I think about alignment
*This was written as part of the first *[*Refine*](https://www.lesswrong.com/posts/5uiQkyKdejX3aEHLM/how-to-diversify-conceptual-alignment-the-model-behind) *blog post day. Thanks for comments by Chin Ze Shen, Tamsin Leake, Paul Bricman, Adam Shimi.*
Magic agentic fluid/force
===========... | https://www.lesswrong.com/posts/9bpACZn6kG2Ec6CPu/how-i-think-about-alignment |
# Shapes of Mind and Pluralism in Alignment
*This post is part of the work done at* [*Conjecture*](https://conjecture.dev)*.*
*This post has been written for the first* [*Refine*](https://www.alignmentforum.org/posts/5uiQkyKdejX3aEHLM/how-to-diversify-conceptual-alignment-the-model-behind) *blog post day, at the end ... | https://www.lesswrong.com/posts/RrirwtP7cNmHtJRxE/shapes-of-mind-and-pluralism-in-alignment |
# I missed the crux of the alignment problem the whole time
*This post has been written for the first *[*Refine*](https://www.alignmentforum.org/posts/5uiQkyKdejX3aEHLM/how-to-diversify-conceptual-alignment-the-model-behind) *blog post day, at the end of the week of readings, discussions, and exercises about epistemol... | https://www.lesswrong.com/posts/dmcJr2NBg5QuGSHFC/i-missed-the-crux-of-the-alignment-problem-the-whole-time |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.