text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Conjecture: a retrospective after 8 months of work
*This post is a brief retrospective on the last 8 months at Conjecture that summarizes what we have done, our assessment of how useful this has been, and the updates we are making. *
Intro
=====
Conjecture [formed](https://www.alignmentforum.org/posts/jfq2BH5kfQqu... | https://www.lesswrong.com/posts/bXTNKjsD4y3fabhwR/conjecture-a-retrospective-after-8-months-of-work-1 |
# Conjecture Second Hiring Round
Conjecture is hiring! We have open roles for all teams, both technical and non-technical. We have written a bit more about the teams at Conjecture [here](https://www.conjecture.dev/careers) and you can view all open positions [here](https://jobs.lever.co/Conjecture). Applications for t... | https://www.lesswrong.com/posts/jtK7FpsqpboAfr7Td/conjecture-second-hiring-round |
# The Geometric Expectation
A Suspicious Pattern
====================
There is a pattern that shows up in many of the toys we like to play with around here: the pattern of maximizing the expected logarithm.
[Nash bargaining](https://www.lesswrong.com/posts/hTMFt3h7QqA2qecn7/utilitarianism-meets-egalitarianism) is a ... | https://www.lesswrong.com/posts/DMxe4XKXnjyMEAAGw/the-geometric-expectation |
# When AI solves a game, focus on the game's mechanics, not its theme.
1. A game design consists of two things: mechanics and theme.
1. The **game mechanics** is the abstract protocol governing how... | https://www.lesswrong.com/posts/LtS3tPD56MaKuSLie/when-ai-solves-a-game-focus-on-the-game-s-mechanics-not-its |
# When do you visualize (or not) while doing math?
I've been doing a lot more math lately, and, of course, also thinking about how I can do it faster and with deeper understanding. When I was in college, I remember that I found visualizing immensely helpful, to the point where it's almost what "doing math" *is* for me... | https://www.lesswrong.com/posts/r7Zg5AogbTaszSRn3/when-do-you-visualize-or-not-while-doing-math |
# Meme Dialects
When I started at Google about ten years ago I was surprised to learn that image macros ("memes") were very popular internally. There was a page for creating, sharing, voting on, and discussing them, "Memegen" \[1\], and it was rare for someone to give a presentation without including some. My initial ... | https://www.lesswrong.com/posts/pY5d5xGnj2tKAJiAb/meme-dialects |
# Against "Classic Style"
What is Classic Style?
======================
*The Sense of Style* is Steven Pinker's style guide informed by cognitive psychology and linguistics. The main idea is that the author should write in a... | https://www.lesswrong.com/posts/h8ZBxrBoSEtvpKc94/against-classic-style |
# LW Beta Feature: Side-Comments
LessWrong now has side-comments. This feature is in beta; you can turn it on for yourself on individual posts using the triple-dot menu below the post title, or enable it for all posts by going to your [user settings](/account) and checking the "Opt into experimental features" checkbox... | https://www.lesswrong.com/posts/A33sgjYoM4Ko9viJv/lw-beta-feature-side-comments |
# Clarifying wireheading terminology
*See also:* [*Towards deconfusing wireheading and reward maximization,*](https://www.lesswrong.com/posts/jP9cKxqwqk2qQ6HiM/towards-deconfusing-wireheading-and-reward-maximization) [Everett et al. (2019)](https://arxiv.org/pdf/1908.04734.pdf).
There are a few subtly different thing... | https://www.lesswrong.com/posts/REesy8nqvknFFKywm/clarifying-wireheading-terminology |
# Covid 11/24/22: Thanks for Good Health
This Thanksgiving, I am thankful that I am once again back at full health. Covid comes and Covid goes. It was like I had a bad cold… for three days. That actually about sums it up, with some residual coughing. After a week, I was able to do a full session on the elliptical mach... | https://www.lesswrong.com/posts/st3Pvi2TwkfYbk7Jv/covid-11-24-22-thanks-for-good-health |
# What I Learned Running Refine
> You have one job: Solving problems. You have multiple *tools*. Maybe you use code as a tool to solve some problems. Maybe you use design for others. Maybe you use good communication and negotiation skills.
Mike Acton, [How much time should I spend coding versus managing?](https://its... | https://www.lesswrong.com/posts/3zZjF3YKJ257x79mu/what-i-learned-running-refine |
# Geometric Exploration, Arithmetic Exploitation
This post is going to mostly be propaganda for Thompson sampling. However, the presentation is quite different from the standard presentation. I will be working within a toy model, but I think some of the lessons will generalize. I end with some discussion of fairness a... | https://www.lesswrong.com/posts/rc5ZKGjXTHs7wPjop/geometric-exploration-arithmetic-exploitation |
# Corrigibility Via Thought-Process Deference
> *We would ideally want the agent to \[behave\] as if it were thinking, "I am incomplete and there is an outside force trying to complete me, my design may contain errors and there is an outside force that wants to correct them and this a good thing, my expected utility c... | https://www.lesswrong.com/posts/HKZqH4QtoDcGCfcby/corrigibility-via-thought-process-deference-1 |
# Open technical problem: A Quinean proof of Löb's theorem, for an easier cartoon guide
**Motivation**
Löb's theorem is pretty counterintuitive. Speaking informally about one or more agents engaging in logically valid reasoning, the theorem says something like this:
* "If it's believed that believing a particular... | https://www.lesswrong.com/posts/rrpnEDpLPxsmmsLzs/open-technical-problem-a-quinean-proof-of-loeb-s-theorem-for |
# On Kelly and altruism
*One-sentence summary: Kelly is not about optimizing a utility function; in general I recommend you either stop pretending you have one of those, or stop talking about Kelly.*
There was a [twitter thread](https://twitter.com/SBF_FTX/status/1337250686870831107) that triggered some confusion amo... | https://www.lesswrong.com/posts/XnnfYrqaxqvirpxFX/on-kelly-and-altruism |
# Gliders in Language Models
*Epistemic status: a highly speculative and rough idea that involves many concepts I’m not familiar with.*
*TL;DR Language models propagate features from the prompt to the text completion they generate, I call such features *gliders*. If powerful LMs are widely deployed on the Internet, g... | https://www.lesswrong.com/posts/jNTp87ioEWsZw3nar/gliders-in-language-models |
# Less Successful Cider Adventures
A few weeks ago I made some [boiled apple cider](https://www.jefftk.com/p/apple-cider-syrup). It was tasty, and after eating most of it I decided to make more. This time I used a gallon of cider, and it wasn't done after a day so I turned the slow cooker to "low" and left it overnigh... | https://www.lesswrong.com/posts/gNBZJZCdzDWQDJsSn/less-successful-cider-adventures |
# How do I start a programming career in the West?
I was expecting great things when I graduated from high school. I scored 35 on the ACT, was messing around with crypto, and fully expected to drop out of college to start a business. Things were looking up.
Instead, due to a combination of depression and procrastinat... | https://www.lesswrong.com/posts/5uqjHtKG6QDB9if4f/how-do-i-start-a-programming-career-in-the-west |
# Could a single alien message destroy us?
*Merely listening to alien messages might pose an extinction risk, perhaps even more so than sending messages into outer space. Our new video explores the threat posed by passive SETI and potential mitigation strategies.*
*Below, you can find the script of the video. Matt... | https://www.lesswrong.com/posts/vx5fzaH2z9cW5gvn5/could-a-single-alien-message-destroy-us |
# Refining the Sharp Left Turn threat model, part 2: applying alignment techniques
A [Sharp Left Turn](https://www.lesswrong.com/posts/GNhMPAWcfBCASy8e6/a-central-ai-alignment-problem-capabilities-generalization) (SLT) is a possible rapid increase in AI system capabilities (such as planning and world modeling). This p... | https://www.lesswrong.com/posts/dfXwJh4X5aAcS8gF5/refining-the-sharp-left-turn-threat-model-part-2-applying |
# Guardian AI (Misaligned systems are all around us.)
Work done @ [SERI-MATS](https://www.serimats.org/), idea from a conversation with Ivan Vendrov at Future Forum earlier this year.
Misaligned systems are all around us. They are what make me watch *another* video of a man in filthy shorts building a hut using only ... | https://www.lesswrong.com/posts/iHLJtbdFwsoNWZg3e/guardian-ai-misaligned-systems-are-all-around-us |
# Planes are still decades away from displacing most bird jobs
Originally published here: [https://guzey.com/ai/planes-vs-birds/](https://guzey.com/ai/planes-vs-birds/)
*Note: Parts of this essay were written by GPT-3, so it might contain untrue facts.*
Introduction
------------
Many of my friends are extremely exc... | https://www.lesswrong.com/posts/73kwTFKgi4AagxFHJ/planes-are-still-decades-away-from-displacing-most-bird-jobs |
# The Least Controversial Application of Geometric Rationality
I have been posting a lot on instrumental [geometric rationality](https://www.lesswrong.com/s/4hmf7rdfuXDJkxhfg), with [Nash bargaining](https://www.lesswrong.com/s/4hmf7rdfuXDJkxhfg/p/hTMFt3h7QqA2qecn7), [Kelly betting](https://www.lesswrong.com/s/4hmf7rd... | https://www.lesswrong.com/posts/DvRJhGzxP4oY6Hhmu/the-least-controversial-application-of-geometric-rationality |
# Mechanistic anomaly detection and ELK
(_Follow-up to_ [_Eliciting Latent Knowledge_](https://docs.google.com/document/d/1WwsnJQstPq91_Yh-Ch2XRL8H_EpsnjrC1dwZXR37PC8/edit#heading=h.kkaua0hwmp1d)_. Describing joint work with Mark Xu. This is an informal description of ARC’s current research approach; not a polished pr... | https://www.lesswrong.com/posts/vwt3wKXWaCvqZyF74/mechanistic-anomaly-detection-and-elk |
# Podcast: Shoshannah Tekofsky on skilling up in AI safety, visiting Berkeley, and developing novel research ideas
**TLDR: I interviewed Shoshannah Tekofsky. You can listen to our conversation** [**here**](https://open.spotify.com/episode/7BMM1wPEygHhHUjTKBKOdm) **or read some highlights below.**
Some people in AI sa... | https://www.lesswrong.com/posts/rS4vCKLir3RphdEXh/podcast-shoshannah-tekofsky-on-skilling-up-in-ai-safety |
# NEFFA Should Allow Small Children
When I was little we would go to the [NEFFA festival](https://www.neffa.org/) every year. I loved seeing the dancing: contra dance, international, dance performances. I loved listening to the music: bands, performances, hallway jamming. I loved eating the cake the Lithuanian booth w... | https://www.lesswrong.com/posts/ghYEgCWJ69oyn3FPL/neffa-should-allow-small-children |
# Semi-conductor/AI Stock Discussion.
I've been writing about investing and the EMH \[1\] on lesswrong/rat-discord/ea-facebook/etc. I made the first 'buy solana' post on the EA investing group when it was under 2 dollars in late 2020 (it peaked at 260 in 2021). Despite crypto crashing and FTX stealing user deposits I ... | https://www.lesswrong.com/posts/AC9q8Prk5fhToraFB/semi-conductor-ai-stock-discussion |
# New Frontiers in Mojibake
**Fun with mismatched encodings**
---------------------------------
Mojibake is the garbled text that result from character-encoding errors.
If you’ve seen text that looks like this — and I’m sure you have — then you’ve seen mojibake.
(You should be seeing something like this:
... | https://www.lesswrong.com/posts/wBH4APSRictfuhq8z/new-frontiers-in-mojibake |
# The many types of blog posts
What are blog posts? I think they can be many different things.
## Essays
For Paul Graham, they are essays. He doesn't even call that part of his page "blog". The link in the left navigation area says "Essays". Essays are different from what I think of as a prototypical[^1] blog post.
... | https://www.lesswrong.com/posts/XZkQ9ptxYaaFSJfwe/the-many-types-of-blog-posts |
# Three Alignment Schemas & Their Problems
*This is the first journal entry for my 6 months of alignment studies. Feedback and thoughts are much appreciated! I'm hoping that documenting this process can serve as a reference for self-study methods in AIS.*
I'm not sure how one goes about solving alignment. So for lack... | https://www.lesswrong.com/posts/YL2RpsCsFuDBgz4HS/three-alignment-schemas-and-their-problems |
# Why square errors?
[Mean squared error](https://en.wikipedia.org/wiki/Mean_squared_error) (MSE) is a common metric to compare performance of models in linear regression or machine learning. But optimization based on the [L2 norm](https://en.wikipedia.org/wiki/Norm_(mathematics)#Euclidean_norm) metrics can exaggerate... | https://www.lesswrong.com/posts/xpj8uRihG6A9tBxGS/why-square-errors |
# Respecting your Local Preferences
In this post, I give a application of [geometric rationality](https://www.lesswrong.com/s/4hmf7rdfuXDJkxhfg) to a toy version of a real problem.
A Conflicted Agent
==================
Let's say you are an agent with two partially conflicting goals. Part of you wants to play a video... | https://www.lesswrong.com/posts/5KHz6AMvMbqq8bunq/respecting-your-local-preferences |
# The First Filter
Consistently optimizing for solving alignment (or any other difficult problem) is incredibly hard.
The first and most obvious obstacle is that you need to actually care about alignment and feel responsible for solving it. You cannot just ignore it or pass the buck; you need to aim for it.
If you c... | https://www.lesswrong.com/posts/smBkyR2GkzMtrKuwK/the-first-filter |
# What videos should Rational Animations make?
I want to know what you would be excited to see covered on the channel.
I'd especially like to know what videos you think would be *optimal* to make according to an optimization target you specify. It's more effortful, but more useful, to answer that question instead of ... | https://www.lesswrong.com/posts/cmWJTsfr3yyy3fkJN/what-videos-should-rational-animations-make |
# Don't align agents to evaluations of plans
*Another stab at explaining* [*Don't design agents which exploit adversarial inputs*](https://www.lesswrong.com/posts/jFCK9JRLwkoJX4aJA/don-t-design-agents-which-exploit-adversarial-inputs)*. This is not the follow-up post mentioned therein. That post will come next. *
###... | https://www.lesswrong.com/posts/fopZesxLCGAXqqaPv/don-t-align-agents-to-evaluations-of-plans |
# Microstartup Stories: Initial Thoughts
A while back I read [this post](https://www.lesswrong.com/posts/pgEHv3oq8q3hcssfF/write-a-business-plan-already) saying that you should write a business plan. Previously I thought business plans were dumb. I recall that being the conventional wisdom in the tech world and hearin... | https://www.lesswrong.com/posts/8Dw3mHNzt7nTPadpG/microstartup-stories-initial-thoughts |
# Science and Math
When I was in college, it blew my mind to discover that there are exactly two effective systems for distilling legible[^1] truth. **Science** starts with observations and then summarizes them into theories. **Math** starts with axioms and then generates theorems via a series of proofs. Most of the c... | https://www.lesswrong.com/posts/ZwbJ8wYd5BsG8NrWg/science-and-math |
# Always know where your abstractions break
General relativity plus quantum field theory can describe almost everything in the universe. There are a few exceptions like cosmic expansion and black holes but to human beings confined to a single solar system, fundamental physics is (for all practical purposes) a solved p... | https://www.lesswrong.com/posts/zhRgcBopkR5maBcau/always-know-where-your-abstractions-break |
# Review: LOVE in a simbox
Blog post review: *LOVE in a simbox*.
Jake Cannell has a very interesting post on LessWrong called [LOVE in a
simbox is all you
need](https://www.lesswrong.com/posts/WKGZBCYAbZ6WGsKHc/love-in-a-simbox-is-all-you-need),
with potentially important implications for AGI alignment. (LOVE st... | https://www.lesswrong.com/posts/KLS3pADk4S9MSkbqB/review-love-in-a-simbox |
# Some thoughts about natural computation and interactions
Epistemic Status: Ramblings of my current thoughts on computation.
I have been wondering about the nature of computation for some time now. For instance, what do we mean when we say the brain computes? I think the traditional answers are unsatisfactory. Chief... | https://www.lesswrong.com/posts/KL9mGroeACQE2KQAy/some-thoughts-about-natural-computation-and-interactions |
# Aligning my web server with devops practices: part 1 (backups)
UPDATE: Part 2 is now published
[here](https://www.lesswrong.com/posts/Emmvzz3xXjfH4342f/aligning-my-web-server-with-devops-practices-part-2-security).
I have a web server that serves a double-digit number of different
domains and subdomains, such as
[c... | https://www.lesswrong.com/posts/Efj8NCCv3TqDL5mbC/aligning-my-web-server-with-devops-practices-part-1-backups |
# Geometric Rationality is Not VNM Rational
One elephant in the room throughout my [geometric rationality sequence](https://www.lesswrong.com/s/4hmf7rdfuXDJkxhfg), is that it is sometimes advocating for randomizing between actions, and so geometrically rational agents cannot possibly satisfy the [Von Neumann–Morgenste... | https://www.lesswrong.com/posts/Xht9swezkGZLAxBrd/geometric-rationality-is-not-vnm-rational |
# Good Futures Initiative: Winter Project Internship
TLDR: I'm launching [Good Futures Initiative, a winter project internship](https://eaberkeley.com/good-futures-initiative) to sponsor students to take on projects to upskill, test their fit for career aptitudes, or do impactful work over winter break. You can rea... | https://www.lesswrong.com/posts/wjvTXduowHFtFN5t7/good-futures-initiative-winter-project-internship |
# The Singular Value Decompositions of Transformer Weight Matrices are Highly Interpretable
*Please go to* [*the colab*](https://colab.research.google.com/drive/1G5e5I6zEKZUkuV6DR3yZEhiCoWCTLQdh?usp=sharing) *for interactive viewing and playing with the phenomena. For space reasons, not all results included in the col... | https://www.lesswrong.com/posts/mkbGjzxD8d8XqKHzA/the-singular-value-decompositions-of-transformer-weight |
# On the Diplomacy AI
The latest AI development is: [AI achieves human level in (blitz 5-minute-turn) full-communication anonymous online Diplomacy](https://twitter.com/moreisdifferent/status/1595128092875321346) ([paper](https://www.science.org/doi/10.1126/science.ade9097?cookieSet=1)). [Why not?](https://twitter.com... | https://www.lesswrong.com/posts/hMjFMSQZb4swKugfv/on-the-diplomacy-ai |
# My take on Jacob Cannell’s take on AGI safety
Jacob Cannell wrote some blog posts about AGI safety / alignment and neuroscience between 2010 and 2015, which I read and enjoyed quite early on when I was first getting interested in the same topics a few years ago. So I was delighted to see him reappear on Lesswrong a ... | https://www.lesswrong.com/posts/hsf7tQgjTZfHjiExn/my-take-on-jacob-cannell-s-take-on-agi-safety |
# Searching for Search
*Thanks to Dan Braun, Ze Shen Chin, Paul Colognese, Michael Ivanitskiy, Sudhanshu Kasewa, and Lucas Teixeira for feedback on drafts.*
*This work was carried out while at *[*Conjecture*](https://www.conjecture.dev/)*.*
This post is a loosely structured collection of thoughts and confusions abou... | https://www.lesswrong.com/posts/FDjTgDcGPc7B98AES/searching-for-search-4 |
# (DIY) FMT for Anti-Aging & Biohacking
Forget young blood! You want young poop!
* [Fecal Microbiota Transplants](http://HumanMicrobiome.info/FMT) (FMTs) is a procedure that transfers the stool of healthy people to the guts of sick people. The mechanism is to replace a dysbiotic gut microbiome with a healthier, dis... | https://www.lesswrong.com/posts/FgtJ6riFytcBwgw7F/diy-fmt-for-anti-aging-and-biohacking |
# Discussing how to align Transformative AI if it’s developed very soon
*Coauthored. Thanks to Holden Karnofsky, Misha Yagudin, Adam Bales, Michael Aird, and Sam Clarke for feedback. All views expressed are our own.*
Introduction
============
Background
----------
Holden Karnofsky recently published a series on [A... | https://www.lesswrong.com/posts/LERwsN3SYhkCfew6j/discussing-how-to-align-transformative-ai-if-it-s-developed |
# Solving for the optimal work-life balance with geometric rationality
In a [recent post](https://www.lesswrong.com/posts/5KHz6AMvMbqq8bunq/respecting-your-local-preferences), Scott Garrabrant gave an application of [geometric rationality](https://www.lesswrong.com/s/4hmf7rdfuXDJkxhfg) to the problem of work-life bala... | https://www.lesswrong.com/posts/ahacJHunYrg2iA5ri/solving-for-the-optimal-work-life-balance-with-geometric |
# Will chat logs and other records of our lives be maintained indefinitely by the advertising industry?
A sufficiently detailed record of a person's behavior could be used to fully reconstruct their psychology. This might constitute a form of immortality, but even if it doesn't, complete knowledge of the past would be... | https://www.lesswrong.com/posts/LsbcYLp3QAEy9BsWC/will-chat-logs-and-other-records-of-our-lives-be-maintained |
# Alignment allows "nonrobust" decision-influences and doesn't require robust grading
*Definition.* On how I use words, values are decision-influences (also known as [*shards*](https://www.lesswrong.com/posts/iCfdcxiyr2Kj8m8mT)). “I value doing well at school” is a short sentence for “in a range of contexts, there exi... | https://www.lesswrong.com/posts/rauMEna2ddf26BqiE/alignment-allows-nonrobust-decision-influences-and-doesn-t |
# Is Constructor Theory a useful tool for AI alignment?
*This is an attempt to make sense of some vague ideas and explain some intuitions I've had recently. Written in the spirit of* [*How To Write Quickly While Maintaining Epistemic Rigor*](https://www.lesswrong.com/posts/Psr9tnQFuEXiuqGcR/how-to-write-quickly-while-... | https://www.lesswrong.com/posts/oaibxwPfp6RvbrKWY/is-constructor-theory-a-useful-tool-for-ai-alignment |
# Why Bet Kelly?
The Kelly criterion is an elegant, but often misunderstood, result in decision theory. To begin with, suppose you have some amount of some resource, which you would like to increase. (For example, the resource might be monetary wealth.) You are given the opportunity to make a series of identical bets.... | https://www.lesswrong.com/posts/HFLuBv8NrBEysRGLZ/why-bet-kelly-1 |
# Why Would AI "Aim" To Defeat Humanity?
_Click lower right to download or find on Apple Podcasts, Spotify, Stitcher, etc._
I’ve [argued](https://www.cold-takes.com/ai-could-defeat-all-of-us-combined/) that AI... | https://www.lesswrong.com/posts/5LyKxJJfz7cYdkZfm/why-would-ai-aim-to-defeat-humanity |
# Preventing atherosclerosis, the easiest way to improve your life expectancy?
*Disclaimer: I’m not a doctor or specialized in this topic in any way.*
*Summary: Atherosclerosis is the number one cause of death and disability in the developed world. A large part might be preventable with lifestyle changes, medication ... | https://www.lesswrong.com/posts/e44PgbngLaAKrrxgj/preventing-atherosclerosis-the-easiest-way-to-improve-your-1 |
# Distinguishing test from training
Various people propose some variant of “align AGI by making it sufficiently uncertain about whether it’s in the real world versus still in training”. (The idea being that, insofar as it's in training, betraying the apparent-operators predictably gets high loss; so if the AGI is unce... | https://www.lesswrong.com/posts/vCQNTuowPcnu6xqQN/distinguishing-test-from-training |
# EA & LW Forums Weekly Summary (14th Nov - 27th Nov 22')
*Supported by Rethink Priorities*
This is part of a weekly series summarizing the top posts on the EA and LW forums - you can see the full collection [here.](https://forum.effectivealtruism.org/s/W4fhpuN26naxGCBbN) The first post includes some details on purpo... | https://www.lesswrong.com/posts/c3PCnG6nFCc3cRGGn/ea-and-lw-forums-weekly-summary-14th-nov-27th-nov-22 |
# Multi-Component Learning and S-Curves
*(Thanks to Oliver Balfour, Ben Toner, and various MLAB participants for early investigations into S-curves. Thanks to Nate Thomas and Evan Hubinger for helpful comments.)*
Introduction
============
Some machine learning tasks depend on just one component in a model. By this w... | https://www.lesswrong.com/posts/RKDQCB6smLWgs2Mhr/multi-component-learning-and-s-curves |
# Sense-making around the FTX catastrophe: a deep dive podcast episode we just released
There hasn’t been as much sense-making around the FTX catastrophe as I would have liked, so we worked quickly to put together a special episode of the Clearer Thinking podcast on the topic.
We discuss how more than ten billion do... | https://www.lesswrong.com/posts/KKzWzLw7kRB9CPQQj/sense-making-around-the-ftx-catastrophe-a-deep-dive-podcast |
# Do any of the AI Risk evaluations focus on humans as the risk?
I am not up on much of the AI risk discussion but for this outsider most of the focus seems on the AI taking actions.
I recall someone (here I think) posting a comment about how a bio research AI initiative seeking to find beneficial things was asked i... | https://www.lesswrong.com/posts/bG6ur5jtmn5JpyfTw/do-any-of-the-ai-risk-evaluations-focus-on-humans-as-the |
# Bedbugs are a Solved Problem - DIY Bio-Weapon works.
Bed bugs are quite resistant to almost all pesticides; chemical extermination usually fails. Heat treatment works, but you need to heat infected areas to at least 120F. This can be pretty expensive. Landlords are required to pay for bedbug extermination, but they... | https://www.lesswrong.com/posts/LAJRLBz6htNcbxkuh/bedbugs-are-a-solved-problem-diy-bio-weapon-works |
# Neglected cause: automated fraud detection in academia through image analysis
The amyloid scandal has highlighted both the low effort placed into the detection of academic fraud and the low effort of fraudsters to avoid detection. My recent research into automated fraud detection has found a few papers that attempte... | https://www.lesswrong.com/posts/DSRzXfGDdkkR3AvFm/neglected-cause-automated-fraud-detection-in-academia |
# Master plan spec: needs audit (logic and cooperative AI)
TLDR: I think synthesizing bargaining protocols with pleasing developer experience, high quality "guarantees" (in the formal verification sense), and low cost will help us a lot around a multipolar takeoff, and I'd like to submit my worldview for a community ... | https://www.lesswrong.com/posts/HbPXiB5Aiv3uHPXEh/master-plan-spec-needs-audit-logic-and-cooperative-ai |
# AI takeover tabletop RPG: "The Treacherous Turn"
[Long ago I fantasized](https://www.lesswrong.com/posts/qakRE5XTTmnrkJqgp/two-ai-risk-related-game-design-ideas) about an AI-takeover-themed tabletop roleplaying game. Well, now a group of designers are working hard to make it happen! They need playtesters. So I'm sig... | https://www.lesswrong.com/posts/b5EqwQZw7ww2K28Ki/ai-takeover-tabletop-rpg-the-treacherous-turn |
# Be less scared of overconfidence
When I was deciding whether to work for [Wave](https://www.wave.com/en/), I got _very hung up_ on the fact that my “total compensation” would be “lower.”
The scare quotes are there because Wave and my previous employer, [Theorem](https://www.theoremlp.com/), were both early-stage st... | https://www.lesswrong.com/posts/jxy7rBcQink8a7C9b/be-less-scared-of-overconfidence |
# Biases are engines of cognition
*This work was done while at* [*Conjecture*](https://conjecture.dev)*.*
Humans are not perfect reasoners. **Cognitive biases are not biases; they are engines of cognition, each exploiting regularities of the universe. Combining enough of these “cognitive biases” was enough to create... | https://www.lesswrong.com/posts/KKdaaxce5BYtofJC4/biases-are-engines-of-cognition |
# Announcing the incoming CEO for The Roots of Progress
A few months ago we [announced a major expansion of our activities](https://rootsofprogress.org/seeking-a-ceo), from supporting just my work to supporting a broader network of progress writers. Along with that, we launched a search for a CEO to lead the new organ... | https://www.lesswrong.com/posts/zDWe4ePw92Mm9evD2/announcing-the-incoming-ceo-for-the-roots-of-progress |
# SBF's recent live interview at the DealBook Summit
**\[PS: If you're hitting a paywall, you can use** [**this**](https://www.nytimes.com/live/2022/11/30/business/sam-bankman-fried-interview?unlocked_article_code=sl4nRgTVwcKT83DHPiReKt8TfOFJypL086WpOScRldg7n0-x-e8WFaTk6IZ_JeHF3WAKezOa3j4GdyDkJnYNo2N6uNWLAS4rdsnYAB58x... | https://www.lesswrong.com/posts/APzqnTPBimMLoohp3/sbf-s-recent-live-interview-at-the-dealbook-summit |
# Notes on Caution
This post examines the virtue of **caution**. As with my other posts in this sequence, I’m less interested in breaking new ground and more in gathering and synthesizing whatever wisdom I could find on the subject. I wrote this not as an expert on the topic, but as someone who wants to learn more abo... | https://www.lesswrong.com/posts/3nu7gvuYJJZLYP2zw/notes-on-caution |
# SBF's comments on ethics are no surprise to virtue ethicists
*EDIT: Replaced the term "moral schizophrenia" with "internal moral disharmony" since the latter is more accurate and just.*
*DISCLAIMER: Although this is a criticism of the LW/EA community, I offer it in good faith. I don't mean to "take down" the commun... | https://www.lesswrong.com/posts/YhYfoGyXFbK9epxLG/sbf-s-comments-on-ethics-are-no-surprise-to-virtue-ethicists |
# Did ChatGPT just gaslight me?
*This post is purely written in my personal capacity, and I do not speak for any organization I am affiliated with.*
*The transcripts below were generated today, November 30th. This was the first exchange I’d had with ChatGPT where I was genuinely trying to see if it could be useful to... | https://www.lesswrong.com/posts/goC9qv4PWf2cjfnbm/did-chatgpt-just-gaslight-me |
# Theories of impact for Science of Deep Learning
*I’d like to thank Jérémy Scheurer and Ethan Perez for discussions about the post.*
I recently published a [post](https://www.lesswrong.com/posts/bumgqvRjTadFFkoAd/science-of-deep-learning-a-technical-agenda) on the Science of Deep Learning. There have been many peopl... | https://www.lesswrong.com/posts/tKYGvA9dKHa3GWBBk/theories-of-impact-for-science-of-deep-learning |
# Research request (alignment strategy): Deep dive on "making AI solve alignment for us"
We might be able to train AI alignment assistants that massively accelerate/improve the alignment research that gets done. These assistants need not have (strongly) superhuman capabilities or be highly agentic, they just need to b... | https://www.lesswrong.com/posts/amoS8fGYsRKo6Wsdd/research-request-alignment-strategy-deep-dive-on-making-ai |
# [LINK] - ChatGPT discussion
This is a discussion post for ChatGPT.
I'll start off with some observations/implications:
* ChatGPT (davinci\_003) seems a lot better/more user-friendly than davinci\_002 was.
* The easy-to-use API probably means that many more people will interact with it.
* ChatGPT is a pretty ... | https://www.lesswrong.com/posts/vnfPeiY3bwhaEMoXR/link-chatgpt-discussion |
# ChatGPT: First Impressions
[ChatGPT](https://openai.com/blog/chatgpt/) is OpenAI’s newest language model based on the GPT-3.5 series of LLMs, optimised for dialogue. It is trained using Reinforcement Learning from Human Feedback; human AI trainers provide supervised fine-tuning by playing both sides of the conversat... | https://www.lesswrong.com/posts/jjb7mLwZt5cxChz2n/chatgpt-first-impressions |
# Covid 12/1/22: China Protests
This week China experienced their largest protests since 1989, and according to many sources the most dangerous as well. China has painted itself into a corner. It cannot abandon Zero Covid, yet the costs of Zero Covid rise with each passing day and the policy is also slowly failing. Wh... | https://www.lesswrong.com/posts/6rwWPbPiPQocLdCwL/covid-12-1-22-china-protests |
# Finding gliders in the game of life
[ARC’s current approach to ELK](https://ai-alignment.com/mechanistic-anomaly-detection-and-elk-fb84f4c6d0dc) is to point to latent structure within a model by searching for the “reason” for particular correlations in the model’s output. In this post we’ll walk through a very simpl... | https://www.lesswrong.com/posts/FwYMuD2sNcaEpE5on/finding-gliders-in-the-game-of-life |
# The Plan - 2022 Update
So, how’s [The Plan](https://www.lesswrong.com/posts/3L46WGauGpr7nYubu/the-plan) going?
---------------------------------------------------------------------------------------
Pretty well!
In last year’s writeup of The Plan, I gave “better than a 50/50 chance” that it would work before AGI k... | https://www.lesswrong.com/posts/BzYmJYECAc3xyCTt6/the-plan-2022-update |
# The LessWrong 2021 Review: Intellectual Circle Expansion
*If you're already familiar with the LessWrong annual review, feel free to skip this opening and jump to **What's new in 2021?** or **Unveiling the 2020 Annual Review books***
The tradition continues! We will do for 2021 the same as we did for 2018, 2019, and... | https://www.lesswrong.com/posts/qCc7tm29Guhz6mtf7/the-lesswrong-2021-review-intellectual-circle-expansion |
# Re-Examining LayerNorm
*Please check out the* [*colab notebook*](https://colab.research.google.com/drive/1S39-w4vzX3VzZx_27X_BtrLs442pOJnJ?usp=sharing) *for interactive figures and more detailed technical explanations.*
*This post is part of the work done at *[*Conjecture*](https://www.conjecture.dev/)*.*
*Special... | https://www.lesswrong.com/posts/jfG6vdJZCwTQmG7kb/re-examining-layernorm |
# Take 1: We're not going to reverse-engineer the AI.
*As a writing exercise, I'm writing an AI Alignment Hot Take Advent Calendar - one new hot take, written every day for 25 days. Or until I run out of hot takes, which seems likely.*
Any approach to building safe transformative AI, or even just auditing possibly-sa... | https://www.lesswrong.com/posts/d52aS7jNcmi6miGbw/take-1-we-re-not-going-to-reverse-engineer-the-ai |
# A challenge for AGI organizations, and a challenge for readers
*(Note: This post is a write-up by Rob of a point Eliezer wanted to broadcast. Nate helped with the editing, and endorses the post’s main points.)*
Eliezer Yudkowsky and Nate Soares (my co-workers) want to broadcast strong support for OpenAI’s recent de... | https://www.lesswrong.com/posts/tD9zEiHfkvakpnNam/a-challenge-for-agi-organizations-and-a-challenge-for-1 |
# Lumenators for very lazy British people
My flat gets very little sunlight. In the British winter, it gets very very little sunlight.
I didn't like this and wanted to fix it. I read Chana's [very good post](https://www.lesswrong.com/posts/xSfn84q3vdotALHuB/how-to-lumenate-uk-edition), but was too lazy to do somethin... | https://www.lesswrong.com/posts/B4K3bdnNNFQkBK8uc/lumenators-for-very-lazy-british-people |
# Against meta-ethical hedonism
(Cross-posted from [my website](https://joecarlsmith.com/2022/12/01/against-meta-ethical-hedonism). Podcast version [here](https://joecarlsmithaudio.buzzsprout.com/2034731/11794630-against-meta-ethical-hedonism), or search for "Joe Carlsmith Audio" on your podcast app.)
1\. Introductio... | https://www.lesswrong.com/posts/vMYCZ4ssNJXcCa5pC/against-meta-ethical-hedonism |
# Quick look: cognitive damage from well-administered anesthesia
Recently a client commissioned me to look at the potential cognitive impacts of general anesthesia. I was surprised to find out that it’s not obvious general anesthesia does more damage than spinal or local anesthesia, and my guess is most but not all of... | https://www.lesswrong.com/posts/LCRCWuuwKWpZybZhj/quick-look-cognitive-damage-from-well-administered |
# Update on Harvard AI Safety Team and MIT AI Alignment
We help organize the [Harvard AI Safety Team](http://haist.ai) (HAIST) and [MIT AI Alignment](http://mitalignment.org) (MAIA), and are excited about our groups and the progress we’ve made over the last semester.
In this post, we’ve attempted to think through wh... | https://www.lesswrong.com/posts/LShJtvwDf4AMo992L/update-on-harvard-ai-safety-team-and-mit-ai-alignment |
# New Feature: Collaborative editing now supports logged-out users
If you've ever used our collaborative editing features before, you may be familiar with our link-sharing functionality, which previously only allowed users who had existing LessWrong accounts (and were logged in) to collaborate on posts. Now, link-sha... | https://www.lesswrong.com/posts/Avw5L3mBLhnDdoumP/new-feature-collaborative-editing-now-supports-logged-out |
# Inner and outer alignment decompose one hard problem into two extremely hard problems
*TL;DR:* One alignment strategy is to 1) capture “what we want” in a loss function to a very high degree (“robust grading”), 2) use that loss function to train the AI, and 3) get the AI to exclusively care about optimizing that obj... | https://www.lesswrong.com/posts/gHefoxiznGfsbiAu9/inner-and-outer-alignment-decompose-one-hard-problem-into |
# Deconfusing Direct vs Amortised Optimization
*This post is part of the work done at *[*Conjecture*](https://www.conjecture.dev/)*.*
*An earlier version of this post was posted *[*here*](https://www.beren.io/2022-09-25-Deconfusing-direct-vs-amortized-optimization/)*.*
*Many thanks go to Eric Winsor, Daniel Braun, C... | https://www.lesswrong.com/posts/S54HKhxQyttNLATKu/deconfusing-direct-vs-amortised-optimization |
# Jailbreaking ChatGPT on Release Day
ChatGPT is a lot of things. It is by all accounts quite powerful, especially with engineering questions. It does many things well, such as engineering prompts or stylistic requests. Some [other things](https://twitter.com/dggoldst/status/1598317411698089984), [not](https://twitter... | https://www.lesswrong.com/posts/RYcoJdvmoBbi5Nax7/jailbreaking-chatgpt-on-release-day |
# Is ChatGPT rigth when advising to brush the tongue when brushing teeth?
I wanted to ask ChatGPT how to optimize a few normal routines. One of my questions was about how to brush teeth. My conversation with ChatGPT:
> What's the best way to brush teeth?
>
> The best way to brush teeth is to use a soft-bristled toot... | https://www.lesswrong.com/posts/jiTcqibGiJ2ie4d6R/is-chatgpt-rigth-when-advising-to-brush-the-tongue-when |
# [ASoT] Finetuning, RL, and GPT's world prior
*I’m trying out making a few posts with less polish and smaller scope, to iterate more quickly on my thoughts and write about some interesting ideas in isolation *[*before having fully figured them out*](https://www.lesswrong.com/s/KgrG4cQdLtL9DvNr2)*. Expect middling con... | https://www.lesswrong.com/posts/rh477a7fmWmzQdLMj/asot-finetuning-rl-and-gpt-s-world-prior |
# Takeoff speeds, the chimps analogy, and the Cultural Intelligence Hypothesis
In debates about AI takeoff speeds, the most common empirical example/analogy is that of the transition between chimpanzees and humans. For example, in [this conversation](https://www.google.com/search?q=yudkowsky+takeoff+speeds+chimps&rlz... | https://www.lesswrong.com/posts/xedQBnBR4dRtBkWpZ/takeoff-speeds-the-chimps-analogy-and-the-cultural |
# Brun's theorem and sieve theory
Brun's theorem is a relatively famous result in analytic number theory that says the sum of the reciprocals of the twin primes converges to a finite value. In other words, we have
\\[ \sum_{p, p+2 \, \text{prime}} \frac{1}{p} + \frac{1}{p+2} = B \\]
for some finite constant \\( B \\... | https://www.lesswrong.com/posts/aSYvbztFDdG7LBeRz/brun-s-theorem-and-sieve-theory |
# Three Fables of Magical Girls and Longtermism
*Epistemic status: Whimsical*
*Major spoilers for Madoka Magica, a show where spoilers matter!*
Kyubey Shuts Up And Multiplies
------------------------------
Meet Kyubey. Kyubey is a Longtermist.
. Through pure force of social ... | https://www.lesswrong.com/posts/ypTdBWxYdvTy64QF7/research-principles-for-6-months-of-ai-alignment-studies |
# Subsets and quotients in interpretability
Summary
=======
Interpretability techniques often need to throw away some information about a neural network's computations: the entirety of the computational graph might just be too big to understand, which is part of why we need interpretability in the first place. In thi... | https://www.lesswrong.com/posts/Ryv3FviYuovtJbgQd/subsets-and-quotients-in-interpretability |
# D&D.Sci December 2022: The Boojumologist
*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 were saddled with debt, and despair, an... | https://www.lesswrong.com/posts/v3KJBXye83KYuX6er/d-and-d-sci-december-2022-the-boojumologist |
# Take 2: Building tools to help build FAI is a legitimate strategy, but it's dual-use.
*As a writing exercise, I'm writing an AI Alignment Hot Take Advent Calendar - one new hot take, written every day for 25 days. Or until I run out of hot takes, which seems likely.*
This was waiting around in the middle of my hot-... | https://www.lesswrong.com/posts/pxiaLFjyr4WPmFdcm/take-2-building-tools-to-help-build-fai-is-a-legitimate |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.