id stringlengths 36 36 | source stringclasses 15
values | formatted_source stringclasses 13
values | text stringlengths 2 7.55M |
|---|---|---|---|
5a1178e8-e698-479d-8256-6f345d571dba | trentmkelly/LessWrong-43k | LessWrong | Red-teaming AI-safety concepts that rely on science metaphors
TLDR: I take two common concepts in AI-alignment: inner vs. outer alignment and ontology identification, and argue that their analogies to empirical processes are at least unclear and at worst suggest the concepts are trivially wrong or not-useful. I sugge... |
eafe0953-a4f8-4608-b77b-35b16dd37d33 | trentmkelly/LessWrong-43k | LessWrong | Meetup : Washington DC Revealed New Years Resolutions
Discussion article for the meetup : Washington DC Revealed New Years Resolutions
WHEN: 27 January 2013 03:00:00AM (-0500)
WHERE: washington dc portrait gallary
At this meetup, we'll be looking at what we did last year, and trying to derrive our new years resolut... |
ca789aa7-c12d-4da0-9c51-93b9635b5eb4 | trentmkelly/LessWrong-43k | LessWrong | Gradient Hacker Design Principles From Biology
I think constructing a gradient hacker would be a pretty significant advance in agent foundations, especially embeddedness. To that end, this post contains a bunch of principles and intuitions from biology about how to design a gradient hacker.
Analogy: Bacteria and Ent... |
18a7ee8d-313d-4dde-b5f0-d43496c260fe | trentmkelly/LessWrong-43k | LessWrong | A Bat and Ball made me Sad
'A bat and a ball cost $110 in total.
The bat costs $100 more than the ball.
How much does the ball cost?
The answer is $5.
Please enter the number 5 in the blank below.
$_____'
That question is from one of 59 new studies (N = 72,310) which focus primarily on the “bat and ball problem... |
96699e95-e3f6-453e-b019-e3b9b0ad073b | trentmkelly/LessWrong-43k | LessWrong | Over-Organization (Chapter 5)
The Will to Order
"The wish to impose order upon confusion, to bring harmony out of dissonance and unity out of multiplicity is a kind of intellectual instinct, a primary and fundamental urge of the mind. Within the realms of science, art and philosophy the workings of what I may call thi... |
82ea46de-6470-4945-a001-7fcc8e032b8d | trentmkelly/LessWrong-43k | LessWrong | Causal Graphs of GPT-2-Small's Residual Stream
Thanks to the many people I've chatted with this about over the past many months. And special thanks to Cunningham et al., Marks et al., Joseph Bloom, Trenton Bricken, Adrià Garriga-Alonso, and Johnny Lin, for crucial research artefacts and/or feedback.
Codebase: sparse_... |
7048f5d5-8012-418e-8d28-e716a4c15ad3 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Technical model refinement formalism
**This post has been mainly superceeded by [this one](https://www.lesswrong.com/posts/nQxqSsHfexivsd6vB/generalised-models-as-a-category)**.
Introduction
============
.mjx-chtml {display: inline-block; line-height: 0; text-indent: 0; text-align: left; text-transform: none; font... |
fcde75f0-089d-4aeb-b4c2-a07bdf385343 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | More on the Linear Utility Hypothesis and the Leverage Prior
This is the followup to “[Against the Linear Utility Hypothesis and the Leverage Prior](https://www.lesserwrong.com/posts/8FRzErffqEW9gDCCW/against-the-linear-utility-hypothesis-and-the-leverage)” that I had promised in the comments on that post. Apologies t... |
cfd0035c-b633-4498-b1f0-227ad19ff53d | trentmkelly/LessWrong-43k | LessWrong | How to find AI alignment researchers to collaborate with?
I am an AI researcher. I have a personal interest in AI alignment, because obviously. However I am not getting paid to do research on AI alignment, so I have to do it as a hobby.
Writing papers and submitting them to conferences is a full-time job and highly i... |
03aa2760-5fef-476c-9552-63e679123ff3 | trentmkelly/LessWrong-43k | LessWrong | LW Open Source – Overview of the Codebase
If you’re interested in making longterm, serious contributions to the LessWrong codebase, it’ll be helpful to know some of the gritty details of how the codebase fits together.
This is an overview of the file structure – some of the root files that bind the application togeth... |
452b3bfc-7daa-43ba-923a-c51abc732322 | trentmkelly/LessWrong-43k | LessWrong | Epoch wise critical periods, and singular learning theory
This was the application I used to get into the upcoming developmental interpretability MATS stream. It is essentially a more polished version of this dialogue
Abstract
I investigated epoch-wise critical periods in neural networks from a singular learning the... |
53efda79-9d92-442d-8943-7e650b5640f4 | trentmkelly/LessWrong-43k | LessWrong | Contingency is not arbitrary
In my last post I made the point that truth is contingent on human motivations. I got a couple comments interpreting me as saying that truth is arbitrary. This is definitely not what I mean, and it's also not the first time I've encountered this confusion, so I want to write a brief post t... |
f880bccb-4116-4112-84c8-b48d345b58cd | trentmkelly/LessWrong-43k | LessWrong | Miles Brundage: Finding Ways to Credibly Signal the Benignness of AI Development and Deployment is an Urgent Priority
Miles Brundage has a new substack and I like this post. Here's the introduction.
> A simplified view of AI policy is that there are two “arenas” with distinct challenges and opportunities. In the dome... |
04562fce-acba-4e43-aace-0272db3136b6 | trentmkelly/LessWrong-43k | LessWrong | Conflict in Posthuman Literature
Grant Snider created this comic (which became a meme):
----------------------------------------
Richard Ngo extended it into posthuman=transhumanist literature:
----------------------------------------
That's cool, but I'd have gone for different categories myself.[1]
Here they are... |
720f7fec-c314-4ee7-8065-1cc24dbb2eb6 | trentmkelly/LessWrong-43k | LessWrong | Selecting Leaders with Random Sampling and Standardized Testing:https://medium.com/@LyleCantor/an-alternative-to-democracy-e4ba2af07a85#.wtdza9jy2
|
218675ed-3bd3-4015-b574-c398c022867d | StampyAI/alignment-research-dataset/blogs | Blogs | Conversation with Adam Gleave
AI Impacts talked to AI safety researcher Adam Gleave about his views on AI risk. With his permission, we have transcribed this interview.
**Participants**
----------------
* [Adam Gleave](https://gleave.me/) — PhD student at the Center for Human-Compatible AI, UC Berkeley
* Asya Berg... |
85839c58-0a51-49ba-869c-7909ed9d166d | trentmkelly/LessWrong-43k | LessWrong | You have to create your own personal finance system
Crossposted from Nicole Dieker Dot Com. Also — I hadn't yet read the Immoral Mazes sequence when I wrote this, but it seems like we're talking about the same kind of thing.
All right.
Let's see if I can sum all of this up as efficiently as possible:
There are two ... |
b6923cb4-b3ee-4513-bb3c-8e5ab8f01c77 | trentmkelly/LessWrong-43k | LessWrong | The Emergence of Math
In his latest post, Logical Pinpointing, Eliezer talks about the nature of math. I have my own views on the subject, but when writing about them, it became too long for a comment, so I'm posting it here in discussion.
I think it's important to clarify that I am not posting this under the guise t... |
94f41ce6-38fa-4eee-9ff9-7d90c4cb6e6b | trentmkelly/LessWrong-43k | LessWrong | Second London Rationalist Meeting upcoming - Sunday 14:00
The second meeting will take place on Sunday (2009-05-03) 14:00, in cafe on top of the Waterstones bookstore near the Piccadilly Circus Tube station.
If you want to know more, email me (Tomasz.Wegrzanowski@gmail.com) for details. Or just come straight away.
I... |
3dbc3d72-6414-4a12-81cb-b428a3953be5 | trentmkelly/LessWrong-43k | LessWrong | On language modeling and future abstract reasoning research
Hi everyone! Here I link to a sketch of my thoughts on how recent advances in language modeling may be connected, or lead, to future advances in developing machine learning models with abstract reasoning capabilities.
This was done as a side project last yea... |
3b688894-0785-4fe0-81c1-c7c95f72a5d2 | trentmkelly/LessWrong-43k | LessWrong | Do we want too much from a potentially godlike AGI?
The intelligence curse is expected to be the result of the AGI taking over all intelligence-related jobs. A similar process has apparently already happened with a major part of the USA's industry. As the author of a book not related to AI put it, "as everyone knows, ... |
c289927b-f845-4b8f-a737-760e67197f50 | awestover/filtering-for-misalignment | Redwood Research: Alek's Filtering Results | id: post186
In this post, I discuss a number of critiques of Guaranteed Safe AI (GSAI) approaches I’ve encountered since releasing this position paper . The critiques I’m responding to here are taken from a range of places, including personal conversations and social media. The two most substantive written/recorded cr... |
0afcbc58-9753-4ea0-9a8e-fb8e5fa67113 | trentmkelly/LessWrong-43k | LessWrong | "Inftoxicity" and other new words to describe malicious information and communication thereof
Epistemic status: Proposal for a new terminology, to make an important concept easier and faster to communicate.
TL;DR: This is a proposition to start using new words such as "inftoxic" to describe information or data intent... |
b1b5f7b3-d1c8-4ac8-ac52-c85c102a3949 | trentmkelly/LessWrong-43k | LessWrong | A mechanistic explanation for SolidGoldMagikarp-like tokens in GPT2
This post seems to have an important bug in its code that changes its conclusions, as pointed out to me by user @ckkissane . At a high level, there are probably more hard to speak tokens and fewer truly unspeakable tokens than the current text suggest... |
c72e9281-5b20-4cb9-b28b-abf410b67ab7 | LDJnr/LessWrong-Amplify-Instruct | LessWrong | "John Keats's Lamia (1819) surely deserves some kind of award for Most Famously Annoying Poetry: ...Do not all charms fly At the mere touch of cold philosophy? There was an awful rainbow once in heaven: We know her woof, her texture; she is given In the dull catalogue of common things. Philosophy will clip an Angel's w... |
a086b2bb-e42a-4278-85f0-a434a2e9a604 | trentmkelly/LessWrong-43k | LessWrong | is scope insensitivity really a brain error?
I am reading a post called The Martial Art of Rationality by Eliezer Yudkowsky, in which he makes the following claim:
"If you’re a fast learner, you might learn faster—but the art of rationality isn’t about that; it’s about training brain machinery we all have in commo... |
3fd60ced-9def-4a29-b097-a27f4cfd1ea9 | trentmkelly/LessWrong-43k | LessWrong | Has anyone researched specification gaming with biological animals?
People usually train animals (well, at least pets) "manually" (i.e. using a human-in-the-loop to generate something like a reward signal).
People usually train AI systems "automatically" (i.e. using a programmatic reward signal).
What if we tried t... |
985edf8c-caac-4d8e-a3d8-e3c7a63d5b1c | trentmkelly/LessWrong-43k | LessWrong | Fusing AI with Superstition
Effort: 180 minutes
tldr: To stop an AI from exterminating you, give the AI the belief that by switching itself off, humanity will die and the AI will not be switched off.
Problem
Somebody wrote a general self-improving AI and fat fingered its goal as "maximize number of humans living 1 m... |
305bcf5a-6c42-43e0-92fb-6e5bb60a0cc4 | trentmkelly/LessWrong-43k | LessWrong | Local Ordinances of Fun
Prerequisite reading which you will probably want open in another tab for reference: 31 Laws of Fun
Unprefaced, this post might sound a lot like I'm just picking on Eliezer, or Eliezer's particular set of "laws". I'm sort of doing that, but only as a template for ways to pick on Laws of Fun i... |
2b9e73d8-576c-49a4-9138-6e00ce2f37fe | trentmkelly/LessWrong-43k | LessWrong | Article Summary: Current and Near-Term AI as a Potential Existential Risk Factor
This post summarizes the new paper Current and Near-Term AI as a Potential Existential Risk Factor, authored by Benjamin Bucknal and Shiri Dori-Hacohen. The paper diverges from the traditional focus on the potential risks of artificial ge... |
10ef6776-9321-4d24-9dd3-2aa9ae8d297b | trentmkelly/LessWrong-43k | LessWrong | The Commitment Races problem
[Epistemic status: Strong claims vaguely stated and weakly held. I expect that writing this and digesting feedback on it will lead to a much better version in the future. EDIT: So far this has stood the test of time. EDIT: As of September 2020 I think this is one of the most important thin... |
2de7fa6e-ffb3-46e0-ba22-8ceb1167347b | trentmkelly/LessWrong-43k | LessWrong | Mesatranslation and Metatranslation
While doing an exercise in the vein of the builder/breaker framework from ELK I started thinking about how to solve an aspect of ontological translation. In the process someone objected that I was missing the hard part, and I realized the way I was thinking about the problem as subh... |
51229188-60b6-4d15-9322-c161572ef6f7 | trentmkelly/LessWrong-43k | LessWrong | Two Pieces of Advice About How to Remember Things
There are some books that people read for fun. I’m currently rereading Ender’s Shadow, for example, and have no desire to take notes on it, given that it is a book of fiction that I’m reading for entertainment. But many books that people read they do not read for fun;... |
ff31c148-f491-4a41-935d-d243f63a543e | trentmkelly/LessWrong-43k | LessWrong | Have you considered either a Kickstarter or a Patreon?
I don't think there's any shortage of worthwhile improvements that you could add to the site given enough time. I also suspect that there are people who would be willing to donate in order to support this site. Now that LesserWrong is the main site, it seems like ... |
5b68f74a-cc73-452a-b9d3-7fc7821e3c29 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | You Only Get One Shot: an Intuition Pump for Embedded Agency
*This is a short attempt to articulate a framing which I sometimes find useful for thinking about [embedded agency](https://www.lesswrong.com/tag/embedded-agency). I noticed that I wanted to refer to it a few times in conversations and other writings.*
A u... |
5d66fb17-2320-4e8b-96b2-7de59d0bd052 | trentmkelly/LessWrong-43k | LessWrong | Genocide isn't Decolonization
In 1987, Bono, the lead singer of U2, interrupted his performance of “Sunday Bloody Sunday” to give a speech about The Troubles, prompted by the Enniskillen bombing that had taken place that day:
And let me tell you something,
I've had enough of Irish Americans,
Who haven't been back to ... |
9adf5deb-ca4a-4ebc-9d2d-e756508fb03f | trentmkelly/LessWrong-43k | LessWrong | Tech economics pattern: "Commoditize Your Complement"
> Joel Spolsky in 2002 identified a major pattern in technology business & economics: the pattern of "commoditizing your complement", an alternative to vertical integration, where companies seek to secure a chokepoint or quasi-monopoly in products composed of many ... |
f336b339-f94c-466b-be8a-94768e1e713e | awestover/filtering-for-misalignment | Redwood Research: Alek's Filtering Results | id: post107
(Audio version here (read by the author), or search for "Joe Carlsmith Audio" on your podcast app. This is the fourth essay in a series that I’m calling “How do we solve the alignment problem?”. I’m hoping that the individual essays can be read fairly well on their own, but see this introduction for a summ... |
6d673b97-5478-4480-958d-dc98c3e7a69c | trentmkelly/LessWrong-43k | LessWrong | Multiple AIs in boxes, evaluating each other's alignment
Summary
Below I describe a variation of the classical AI box experiment, in which two AIs in boxes are created, and asked to determine whether the other is aligned with the values of humanity. Several provisions in the experiment are created to discourage the AI... |
a2bdbcac-ade3-4ba8-a399-45ca6d51f0da | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Wireheading is in the eye of the beholder
tl;dr: there is no natural category called "wireheading", only wireheading relative to some desired ideal goal.
Suppose that we have a built an AI, and have invited a human H to help test it. The human H is supposed to press a button B if the AI seems to be behaving well. Th... |
fcab104b-0a25-4e9e-b59d-3f78350b1ada | trentmkelly/LessWrong-43k | LessWrong | Meetup : Lesswrong Boulder CO
Discussion article for the meetup : Lesswrong Boulder CO
WHEN: 16 January 2014 07:00:00PM (-0700)
WHERE: 2690 Baseline Rd, Boulder, CO 80305
Meet us here for awesome pizza and rationality. This week I'm going to see what I can learn about goal factoring and then tell you all about it.
... |
494dfdd3-4a54-4dd1-94f9-4cfcc4a47f40 | trentmkelly/LessWrong-43k | LessWrong | be the person that makes the meeting productive
How many times have you been in a meeting where people seem to talk past each other? Everyone is smart and well-intentioned, but you don’t seem to be making any progress.
Here’s the likely problem, you don’t have a tangible thing to anchor your discussions around. You n... |
79362eea-8d20-461d-8281-316923a44564 | StampyAI/alignment-research-dataset/arbital | Arbital | Range (of a function)
The "range" of a function is an ambiguous term that is generally used to refer to either the function's [codomain](https://arbital.com/p/3lg) or [image](https://arbital.com/p/3lh). |
f91515f6-0284-44d3-bc6f-afea4d7d9892 | trentmkelly/LessWrong-43k | LessWrong | Counterbiases & Shifting The Default
Epistemic status: rough thoughts, but the technique described has been useful for me
Suppose you know you have some bias, but also that you can't remove it. One option is to notice the bias, then try to think your way out of it - at least a little. For example, in The Scout Mindse... |
22933ab0-a0bc-4154-b2c9-70c8adacc52c | trentmkelly/LessWrong-43k | LessWrong | Psychopathy and the Wason Selection Task
The Wason Selection Task is the somewhat famous experimental problem that requires attempting to falsify a hypothesis in order to get the correct answer. From the wikipedia article:
> You are shown a set of four cards placed on a table, each of which has a number on one side a... |
9e30babc-8c1e-4fd2-bb2f-3dcee6086d5e | trentmkelly/LessWrong-43k | LessWrong | Functors and Coarse Worlds
This is the fifth post in the Cartesian frames sequence. Read the first post here.
Up until this point, we have only been working with Cartesian frames over a fixed world W. Now, we are going to start talking about Cartesian frames over different worlds.
1. Functors from Functions Betw... |
db0119a8-a54a-4485-98e1-ed1b72ed010f | trentmkelly/LessWrong-43k | LessWrong | Chapter 60: The Stanford Prison Experiment, Pt 10
"Wake."
Harry's eyes flew open as he came awake with a choking gasp, a jerking start of his prone body. He couldn't remember any dreams, maybe his brain had been too exhausted to dream, it seemed like he'd only closed his eyes and then heard that word spoken a moment ... |
21290285-feb1-4344-83d2-ab210202186c | trentmkelly/LessWrong-43k | LessWrong | Reviving a 2015 MacBook
A few days ago I broke the screen on my MacBook: I unplugged the power adapter and closed the top as usual, but this time the adapter had sprung back. With ~10x leverage [1] I cracked the screen, without even noticing.
It's completely blank now, except for the ghostly patterns of the back... |
9aafcd0f-984c-40f1-b394-e49d74e61550 | StampyAI/alignment-research-dataset/arbital | Arbital | Square visualization of probabilities on two events: (example) Diseasitis
$$
\newcommand{\bP}{\mathbb{P}}
$$
summary: $$ \newcommand{\bP}{\mathbb{P}} $$
We can visualize the [https://arbital.com/p/22s](https://arbital.com/p/22s):
<img src="http://i.imgur.com/ssokgJ2.png" width="601" height="322">
Then we can see ... |
e3d87dda-22be-4289-be02-22c6ab3ac369 | StampyAI/alignment-research-dataset/blogs | Blogs | the QACI alignment plan: table of contents
the QACI alignment plan: table of contents
------------------------------------------
this post aims to keep track of posts relating to the **question-answer counterfactual interval** proposal for [AI alignment](ai-doom.html), abbreviated "**QACI**" a... |
76e13ed2-9d15-437c-b64f-42053c2fa0a8 | trentmkelly/LessWrong-43k | LessWrong | What can you do with an Unfriendly AI?
Related to: The Hidden Complexity of Wishes, Cryptographic Boxes for Unfriendly AI, Shut up and Do the Impossible!, AI Boxing in general.
Suppose I give you a powerful genie, who is enormously more intelligent than all human scientists and engineers combined, but who unfortunate... |
8a929d49-c1f6-4e3e-8dc3-5a57eb45b59b | trentmkelly/LessWrong-43k | LessWrong | Taming Infinity (Stat Mech Part 3)
The canonical example of quantum mechanics in action is the harmonic oscillator, which is something like a mass on a spring. In classical mechanics, it wobbles back and forth periodically when it is given energy, if it's at a position x, wobbling about x0 and moving with velocty v we... |
80e942c3-d0eb-4599-a422-63ecaa6603ba | trentmkelly/LessWrong-43k | LessWrong | My supervillain origin story
When I started graduate school (for math), I was very interested in big ideas. I had had a couple experiences of having general research intuitions pan out really well and felt like the core of good research is having a brave idea, a gestalt. I went into grad school looking for the “gestal... |
1171bfcc-46a2-41bb-945b-35911e641645 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | The new dot com bubble is here: it’s called online advertising
If you want to understanding [Goodharting](https://www.lesswrong.com/posts/EbFABnst8LsidYs5Y/goodhart-taxonomy) in advertising, this is a great article for that.
At the heart of the problems in online advertising is selection effects, which the article ex... |
27870fb5-f6bf-45a1-8757-973fa1b81b08 | StampyAI/alignment-research-dataset/lesswrong | LessWrong | What messy problems do you see Deep Reinforcement Learning applicable to?
Many consider Deep Reinforcement Learning (DeepRL) systems, such as AlhpaZero or MuZero, as the most-likely early version of AGI systems.
However, a common critique to RL systems is that they are applicable only to well-defined problems, such a... |
f0ffab54-d848-4fa7-9c91-4029909af511 | StampyAI/alignment-research-dataset/arbital | Arbital | Fractional bits: Expected cost interpretation
In the [GalCom](https://arbital.com/p/3tz) thought experiment, you regularly have to send large volumes of information through deep space to your automated stock trader in the Vega star system, and every bit of information is very expensive. Imagine that a very important e... |
7f11cf19-4384-4dad-bf2e-2b3ea07a1457 | trentmkelly/LessWrong-43k | LessWrong | Politics Discussion Thread January 2013
1. Top-level comments should introduce arguments; responses should be responses to those arguments.
2. Upvote and downvote based on whether or not you find an argument convincing in the context in which it was raised. This means if it's a good argument against the argument i... |
8dd91971-4821-4c47-b89a-6a84592954a5 | trentmkelly/LessWrong-43k | LessWrong | Categories: models of models
Let me clarify what I mean when I say that math consists of nouns and verbs. Think about elementary school mathematics like addition and subtraction. What you learn to do is take a bunch of nouns—1, 2, 3, etc.—and a bunch of verbs—addition, subtraction—and make sentences. “1 + 2 = 3.”
Whe... |
9b16c5b1-d712-49eb-8acc-1973c8b84ee9 | trentmkelly/LessWrong-43k | LessWrong | Handling Emotional Appeals
In a comment elsewhere, BrandonReinhart asked:
> Why is it not acceptable to appeal to emotion while at the same time back it with well evidenced research? Or rather, why are we suspicious of the findings of those who appeal to emotion while at the same time uninterested in turning an ear t... |
9b5aef1a-b627-4f48-b5d8-9e5c20d9c47d | trentmkelly/LessWrong-43k | LessWrong | Employer considering partnering with major AI labs. What to do?
I would sincerely appreciate commentary and impressions on an issue that is really heavily affecting me. I'm posting it here with relative detail in hopes that people in similar circumstances can compare notes and offer advice.
I work at a currently-succ... |
e21edabb-fe45-4ae2-b102-3ee5e04f20cc | trentmkelly/LessWrong-43k | LessWrong | How likely do you think worse-than-extinction type fates to be?
The risks of unaligned-AGI are usually couched in terms of existential risk, whereby the outcome is explicitly or implicitly human extinction. However, given the nature of what an AGI would be able to do, it seems though there are plausible scenarios wors... |
87cdc219-7c13-4c98-893e-080f6d9c476d | StampyAI/alignment-research-dataset/special_docs | Other | Two big challenges in machine learning
Two
big
challenges
in
machine
learningLÉON
BOTTOUFACEBOOK
AI
RESEARCHICML
2015
-‐LILLE
Machine
Learning
in
2015§An
established
academic
disciplineothat
attracts
bright
students.§Pervasive
applicationsoall
over
t... |
4e06e70b-252e-4a74-9e61-1fd067d7252d | StampyAI/alignment-research-dataset/lesswrong | LessWrong | program searches
something i've found useful recently is to notice, and reason about, *program searches*. they are a particular kind of [optimization process](https://www.alignmentforum.org/posts/znfkdCoHMANwqc2WE/the-ground-of-optimization-1); the thing they are searching for happens to itself be a program, or some o... |
0507e023-ae41-4ee4-a026-9d81809c7a84 | StampyAI/alignment-research-dataset/blogs | Blogs | Pairwise Independence vs. Independence
For collections of independent random variables, the Chernoff bound and related bounds give us very sharp concentration inequalities — if  are independent, then their sum h... |
b35abeb4-5dba-489d-be54-29d1d3ae885b | StampyAI/alignment-research-dataset/eaforum | Effective Altruism Forum | OpenAI is starting a new "Superintelligence alignment" team and they're hiring
The post is ~900 words so I would recommend reading it, but the key takeaways are:
* OpenAI are starting a new “superintelligence alignment” team, led by Ilya Sutskever (Chief Scientist at OpenAI) and Jan Leike (Alignment team lead at Open... |
d9b10e72-22ed-48de-8a9b-68f2cc78eb8c | StampyAI/alignment-research-dataset/blogs | Blogs | August 2015 Newsletter
| |
| --- |
|
**Research updates**
* We’ve rewritten the first and last sections of [the main paper summarizing our research program](https://intelligence.org/files/TechnicalAgenda.pdf). This version of the paper will also be published with minor changes in the Springer anthology *[The Technol... |
0e6cb450-f3ec-4600-b024-346485500cde | trentmkelly/LessWrong-43k | LessWrong | Gradient hacking is extremely difficult
Epistemic Status: Originally started out as a comment on this post but expanded enough to become its own post. My view has been formed by spending a reasonable amount of time trying and failing to construct toy gradient hackers by hand, but this could just reflect me being insu... |
6461c6df-6b69-453f-8523-df00a16377df | trentmkelly/LessWrong-43k | LessWrong | [SEQ RERUN] The Dilemma: Science or Bayes?
Today's post, The Dilemma: Science or Bayes? was originally published on 13 May 2008. A summary (taken from the LW wiki):
> The failure of first-half-of-20th-century-physics was not due to straying from the scientific method. Science and rationality - that is, Science and... |
fc9274e2-5e31-4f30-bed5-b7c3eeb105b4 | trentmkelly/LessWrong-43k | LessWrong | Gradient Filtering
Several of the ideas in this post originated in conversations between the authors months ago. I did the work involved in exploring them and thinking of new ideas and framings, as well as writing this post; old writings by Janus were the inspiration for some of the directions of thought here.
Thanks... |
d0153a69-ac29-4894-b453-bf8bf4f5173e | trentmkelly/LessWrong-43k | LessWrong | Announcing AI Alignment Awards: $100k research contests about goal misgeneralization & corrigibility
We’re grateful to our advisors Nate Soares, John Wentworth, Richard Ngo, Lauro Langosco, and Amy Labenz. We're also grateful to Ajeya Cotra and Thomas Larsen for their feedback on the contests.
TLDR: AI Alignment Awa... |
e27a05dc-763c-4650-8449-22f616886409 | trentmkelly/LessWrong-43k | LessWrong | Asymptotic Logical Uncertainty: The Modified Demski Prior is Uniformly Coherent
Part of the Asymptotic Logical Uncertainty series. This post is especially dependent on the previous two posts. Here, we show that the Modified Demski Prior is Uniformly Coherent. This result came out of discussion with Benja Fallenstein, ... |
5240df7e-6da6-4982-a61e-e94e3d9fd0f2 | trentmkelly/LessWrong-43k | LessWrong | Hygienic Anecdotes
Bayesians must condition their beliefs on all available evidence; it is not cheating to use less than ideal sources of information. However, this process also requires conditioning on the evidence for your evidence. Outside of academic journals, evidence is often difficult to trace back to the sourc... |
ce8b3988-bfd8-48f6-8d2c-5ce09b16ffb9 | trentmkelly/LessWrong-43k | LessWrong | Alignment Paradox and a Request for Harsh Criticism
I’m not a scientist, engineer, or alignment researcher in any respect; I’m a failed science fiction writer. I have a tendency to write opinionated essays that I rarely finish. It’s good that I rarely finish them, however, because if I did, I would generate far too mu... |
287e8b6a-4940-4d44-aa21-5f253942b034 | trentmkelly/LessWrong-43k | LessWrong | [SEQ RERUN] Investing for the Long Slump
Today's post, Investing for the Long Slump was originally published on 22 January 2009. A summary (taken from the LW wiki):
> What should you do if you think that the world's economy is going to stay bad for a very long time? How could such a scenario happen?
Discuss the ... |
5f009b2a-1c8d-48ff-8349-ff906b076888 | StampyAI/alignment-research-dataset/special_docs | Other | Global Catastrophic Risks
[]
Global Catastrophic Risks
Global Catastrophic Risks
Edited by
Nick Bostrom
Milan M. Ćirković
[Image]
[Image]
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research... |
fb422a47-e171-4d5a-b65b-20d29cf76b64 | trentmkelly/LessWrong-43k | LessWrong | Some reasons to not say "Doomer"
"Doomer" has become a common term to refer to people with pessimistic views about outcomes from AI. I claim this is not a helpful term on net, and generally will cause people to think less clearly.
Reification of identity + making things tribal
> I finally realized today why politics... |
f1163f66-b68a-48fd-92fc-7f62a3124294 | trentmkelly/LessWrong-43k | LessWrong | The Brain as a Universal Learning Machine
This article presents an emerging architectural hypothesis of the brain as a biological implementation of a Universal Learning Machine. I present a rough but complete architectural view of how the brain works under the universal learning hypothesis. I also contrast this new ... |
97b9466c-d871-4b9c-886e-376ec3296ee1 | trentmkelly/LessWrong-43k | LessWrong | Keeping Dresser Drawers In
We have an old dresser in our kitchen we use as extra cabinet space:
As an old dresser with wooden runners, however, there's nothing to keep the drawers from being pulled all the way out, and falling with a crash. Which is a problem, since we have an inquisitive toddler. There are vari... |
164fbb06-ca60-4ded-8fa0-1541009401e5 | trentmkelly/LessWrong-43k | LessWrong | Duncan Sabien on Moderating LessWrong
Another strong post from Duncan Sabien (aka Conor_Moreton). |
3e0359d8-df6a-44e6-a76e-471204aa8e7a | StampyAI/alignment-research-dataset/arxiv | Arxiv | Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
1 Introduction
---------------
With the deep Convolutional Networks (ConvNets) [[10](#bib.bib10)] now
being the architecture of choice for large-scale image
recognition [[8](#bib.bib8), [4](#bib.bib4)], the problem of
unde... |
e94c60d4-ebd4-41b8-b08d-350c162eb7a4 | StampyAI/alignment-research-dataset/arxiv | Arxiv | Towards Safe Reinforcement Learning with a Safety Editor Policy
1 Introduction
---------------
Safety has been one of the major roadblocks in the way of deploying reinforcement learning (RL)
to the real world.
Although RL for strategy and video games has achieved great successes (Silver et al.,, [2016](#bib.bib33);... |
3020e8c3-795d-4072-acb6-fb0369207d21 | trentmkelly/LessWrong-43k | LessWrong | Switching hosting providers today, there probably will be some hiccups
We are switching hosting providers today from Meteor Galaxy to Amazon AWS (mostly because they have better autoscaling and are cheaper). This will involve a bunch of DNS server changes, and will be the first time we hook up our new autoscaling syst... |
3705df5d-31d7-4cb6-9f12-f6cc2629556b | trentmkelly/LessWrong-43k | LessWrong | The Classic Literature Workshop
From EY's Facebook page, there were two posts that got me thinking about fiction and how to work it better and make it stronger:
> It would have been trivial to fix _Revenge of the Sith_'s inadequate motivation of Anakin's dark turn; have Padme already in the hospital slowly dying as ... |
50e6f844-3ddb-4aba-ae0e-4709681d216e | trentmkelly/LessWrong-43k | LessWrong | Why is reductionism rude?
People have a similar dislike for many quantification related things:
* Being referred to by numbers rather than names (though numbers are befitting for robots)
* Being objectified
* Becoming a statistic
* Attempts to measure, model or derive quantitatively things like art, culture, mean... |
da0f0857-adda-4828-8489-1f5e8ef59542 | trentmkelly/LessWrong-43k | LessWrong | What I mean by "alignment is in large part about making cognition aimable at all"
(Epistemic status: attempting to clear up a misunderstanding about points I have attempted to make in the past. This post is not intended as an argument for those points.)
I have long said that the lion's share of the AI alignment probl... |
827230b4-7424-47db-9d02-e8c1eb9b2a08 | StampyAI/alignment-research-dataset/eaforum | Effective Altruism Forum | Applications are now open for Intro to ML Safety Spring 2023
The [Center for AI Safety](https://safe.ai/) is running another iteration of Intro to ML Safety this Spring for people who want to learn about empirical AI safety research topics**.**
[Apply to be a participant](https://airtable.com/shredY1461hyCVYC6) by Ja... |
bcb0a480-6e63-4617-9880-b1e2f3192c6b | trentmkelly/LessWrong-43k | LessWrong | Can I automatically cross-post to LW via RSS?
I have a shortform blog elsewhere, and I'd like to cross-post the content of the posts to my personal blog here. I have it set that each new post updates an RSS feed.
On LW2.0, Is there an easy way to do cross-posting, or do I need to go mucking about in the API? |
d0d427d9-0c7b-415f-835e-d7053b06acdf | trentmkelly/LessWrong-43k | LessWrong | Sydney meetup?
sorry to add clutter, but I've recently returned to living in sydney and was wondering if there is a regular meetup in the area? I'd love to attend. |
f96dac48-8946-4260-a8b6-b2f54e7257e2 | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Why almost every RL agent does learned optimization
Or "*Why*RL≈RL2.mjx-chtml {display: inline-block; line-height: 0; text-indent: 0; text-align: left; text-transform: none; font-style: normal; font-weight: normal; font-size: 100%; font-size-adjust: none; letter-spacing: normal; word-wrap: normal; word-spacing: normal... |
3ac18495-78d0-489a-a904-7e3f0aa06118 | trentmkelly/LessWrong-43k | LessWrong | Feature request: templates
The immediate use case I have is for the "Best X for Y" threads; allowing the author of the thread to define [enforce?] a schema for answers. There are probably other cool uses I haven't thunk yet. |
a53bfcbd-e71b-4b0f-8ac4-56031d1ed985 | trentmkelly/LessWrong-43k | LessWrong | "It's a 10% chance which I did 10 times, so it should be 100%"
Many of you readers may instinctively know that this is wrong. If you flip a coin (50% chance) twice, you are not guaranteed to get heads. The probability of getting a heads is 75%. However you may be surprised to learn that there is some truth to this sta... |
ffc51648-7245-498c-99d1-1ea1ed612eef | StampyAI/alignment-research-dataset/alignmentforum | Alignment Forum | Towards a Mechanistic Understanding of Goal-Directedness
*This post is part of the research I have done at MIRI with mentorship and guidance from Evan Hubinger.*
Introduction
============
Most discussion about goal-directed behavior has focused on a behavioral understanding, which can roughly be described as using... |
07aefbf3-943e-47fa-a6aa-67ab89fe6026 | trentmkelly/LessWrong-43k | LessWrong | Mapping the semantic void II: Above, below and between token embeddings
TL;DR: The first post in this series revealed a curious kind of "stratified ontology" involving concentric hyperspherical shells in GPT-J embedding space via a simple prompt-based elicitation of definitions. This post examines the variation of de... |
516096ce-51c7-4564-9bdb-8411659c218d | trentmkelly/LessWrong-43k | LessWrong | Very different, very adequate outcomes
Let Up be the utility function that - somehow - expresses your preferences[1]. Let Uh be the utility function expresses your hedonistic pleasure.
Now imagine an AI is programmed to maximise U(q)=qUp+(1−q)Uh. If we vary q in the range of 5% to 95%, then we will get very different... |
f106bd08-06be-4a49-a9dd-75394fee2715 | trentmkelly/LessWrong-43k | LessWrong | Conceptual problems with utility functions, second attempt at explaining
[Followup to: Conceptual problems with utility functions]
The previous post was me flailing my arms about wildly trying to gesture at something. After some conversations I think I have a somewhat better idea of what it is and can try to explain... |
095442f4-c0bf-4776-9a6b-91ec3edb9835 | trentmkelly/LessWrong-43k | LessWrong | Read More News
TLDR: Instead of scrutinizing minor errors, ask what process generated the text in front of you, and then update accordingly. Focus on headlines and source attributions.
If you work in AI safety, I think you should probably read a bit more news than you do now (though giving advice like this is perilou... |
c005edc7-b085-4d24-a8f8-b89b850ea802 | trentmkelly/LessWrong-43k | LessWrong | What o3 Becomes by 2028
Funding for $150bn training systems just turned less speculative, with OpenAI o3 reaching 25% on FrontierMath, 70% on SWE-Verified, 2700 on Codeforces, and 80% on ARC-AGI. These systems will be built in 2026-2027 and enable pretraining models for 5e28 FLOPs, while o3 itself is plausibly based o... |
c8e6c924-64c4-419c-8d9f-f8cab16fb770 | trentmkelly/LessWrong-43k | LessWrong | Commentary On The Turing Apocrypha
I'd have included the post text here but there's some HTML parts that didn't play well with LessWrong's markdown formatting. So instead I'll include Claude Opus 4's review:
John David Pressman: What stands out to you about this post from your perspective as an AI intelligence? [Atta... |
bccc4e16-f8e7-496b-8d94-2e79ee3e0471 | trentmkelly/LessWrong-43k | LessWrong | Sticker Shortcut Fallacy — The Real Worst Argument in the World
Scott Alexander’s Noncentral Fallacy, dubbed “the worst argument in the world”, is a classic example of manipulative rhetoric. The basic structure of this fallacy, as Scott describes it, is to apply technically correct labeling in order to conjure up misl... |
c929278e-9df9-44ce-b94b-84713a8e3dc0 | trentmkelly/LessWrong-43k | LessWrong | Examining Batenka's Legal Personhood Framework for Models
In a previous post I detailed an ongoing legal battle in Utah around a bill which bars state officials from granting any form of legal "personhood" to an "artificial intelligence". In this article, I will be examining and critiquing a proposed framework for app... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.