text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Evaluations (of new AI Safety researchers) can be noisy
**TL;DR: ***Evaluating whether or not someone will do well at a job is hard, and evaluating whether or not someone has the *potential *to be a great AI safety researcher is even harder. This applies to evaluations from other people (e.g. job interviews, first i... | https://www.lesswrong.com/posts/HACcn8roty9KBAWzZ/evaluations-of-new-ai-safety-researchers-can-be-noisy |
# Questions about AI that bother me
*Crossposted from the EA Forum:* [*https://forum.effectivealtruism.org/posts/4TcaBNu7EmEukjGoc/questions-about-ai-that-have-bothered-me*](https://forum.effectivealtruism.org/posts/4TcaBNu7EmEukjGoc/questions-about-ai-that-have-bothered-me)
As 2022 comes to an end, I thought it'd b... | https://www.lesswrong.com/posts/Bhrs7kGkEnDsuCTDa/questions-about-ai-that-bother-me |
# Sex is Good, Actually
*Okay, it's a* [*series*](https://www.lesswrong.com/s/mifkJ4GTdnpHspv7E) *now. This is another casual post in the spirit of what I wish someone had told me when I was younger.*
When I was growing up, the general message I got was that having sex was bad and would ruin my life.
There's a bunch... | https://www.lesswrong.com/posts/QqsWwhR9h3LRJzt2h/sex-is-good-actually |
# Research Direction: Be the AGI you want to see in the world
Be the AGI you want to see in the world.
*Epistemic status: highly speculative, authors are not neuroscientists.*
Summary
=======
* It may be possible to enhance human intelligence via a brain-computer interface (BCI). We could put electrodes into a hu... | https://www.lesswrong.com/posts/FnfAnsAH6dva3kCHS/research-direction-be-the-agi-you-want-to-see-in-the-world |
# H5N1 - thread for information sharing, planning, and action
Hi everyone,
I've been reading up on H5N1 this weekend, and I'm pretty concerned. Right now my ~~estimate~~ hunch is that there is a ~~5%~~ non-zero chance that it will cost more than 10,000 people their lives.
To be clear, I think **it is unlikely that H... | https://www.lesswrong.com/posts/WcgaudunajStdbQGx/h5n1-thread-for-information-sharing-planning-and-action |
# Have an idea? Come to Oxford to discuss and write (20 – 24 March)
We’re running a workshop (20 – 24 March) in Oxford with three goals.
1. Go in with only a vague idea; come out with a drafted academic essay that you could eventually publish.
2. Build tacit knowledge about the research/doing-good-things sphere.
3.... | https://www.lesswrong.com/posts/BuiZPHC4efmqSftvz/have-an-idea-come-to-oxford-to-discuss-and-write-20-24-march-1 |
# Control
*[Metadata: crossposted from [https://tsvibt.blogspot.com/2022/08/control.html](https://tsvibt.blogspot.com/2022/08/control.html). First completed 3 July 2022.]*
I don't know how to define control or even point at it except as a word-cloud, so it's probably wanting to be refactored. The point of talking abo... | https://www.lesswrong.com/posts/Hqahetrx6g8FncokC/control |
# Teaching Simple Boundaries
In general I'm a big advocate of child-proofing. The less they can get hurt with or mess up, the more you can relax, let your guard down, go longer without looking over to see how they're doing, and focus on other things. On the other hand, some things (cords, trash cans) are difficult or ... | https://www.lesswrong.com/posts/XJ2b859PxsNWqqrCp/teaching-simple-boundaries |
# Are short timelines actually bad?
Sam Altman recently[ posted](https://twitter.com/sama/status/1621621725791404032) the following:
UPDATE (14th Feb 2023): ChatGPT appears to have been patched! However, very strange behaviour can still be elicited in the [OpenAI playground](https://platform.openai.com/playground), particularly with the davinci-instruct model.
More technical details [here](https://www.... | https://www.lesswrong.com/posts/aPeJE8bSo6rAFoLqg/solidgoldmagikarp-plus-prompt-generation |
# Interview Daniel Murfet on Universal Phenomena in Learning Machines
(*The following is a heavily edited version of an interview with Daniel Murfet. For a primer on his research in singular statistical learning theory and its relation to AI safety see the excellent new explainer by Jesse Hoogland:* [*"Neural networks... | https://www.lesswrong.com/posts/3RoFKKgTjEZ9ZgKoG/interview-daniel-murfet-on-universal-phenomena-in-learning |
# Robin Hanson on "Explaining the Sacred"
I found this post an interesting prompt. I'm thinking about it, and I'd enjoy hearing what other LessWrongers think about it.
Going ahead pasting the text here, hope that's okay:
> The following are 45 correlates that I’ve collected of things called “sacred”. I invite any of... | https://www.lesswrong.com/posts/qqwZtaC5gPvQ3rwHN/robin-hanson-on-explaining-the-sacred |
# Why Are Bacteria So Simple?
As far as we can tell, bacteria were the first lifeforms on Earth. Which means they’ve had a full four billion years to make something of themselves. And yet, despite their long evolutionary history, they mostly still look like this:
 *\- Winter 2022 Cohort*
[In a previous post, I demonstrated that Brownian motion near singularities defies our expectations fr... | https://www.lesswrong.com/posts/Fjoy5SxgBmxfy7FNB/gradient-surfing-the-hidden-role-of-regularization |
# Why is Everyone So Boring? By Robin Hanson
This is from the blog [Overcoming Bias](https://www.overcomingbias.com/).
These excerpts describe the overall model. It is only due to Robin Hanson's writing skill that the dynamic can be described this concisely, yet still be covered so thoroughly and intuitively. This is... | https://www.lesswrong.com/posts/Bc9qnvKZXWBcmGdfM/why-is-everyone-so-boring-by-robin-hanson |
# Decision Transformer Interpretability
TLDR: We analyse how a small Decision Transformer learns to simulate agents on a grid world task, providing evidence that it is possible to do circuit analysis on small models which simulate goal-directedness. We think Decision Transformers are worth exploring further and may pr... | https://www.lesswrong.com/posts/bBuBDJBYHt39Q5zZy/decision-transformer-interpretability |
# Project Idea: Lots of Cause-area-specific Online Unconferences
Here's a project idea I have that's been laying around for some time. I think this would be high impact, but I need help to make it happen. It don't have to be this exact setup (which I describe below) as long as the output are more online unconferences... | https://www.lesswrong.com/posts/ez5B9Ndy9KH7ZjtDm/project-idea-lots-of-cause-area-specific-online |
# Monthly Roundup #3
It’s that time again.
### Bad News
Guardian analyzes projects from world’s biggest carbon offset provider, [finds ‘94% of the credits had no benefit to the climate.](https://twitter.com/curious_founder/status/1615753645265649665)’ [Bloomberg found something similar in 2020](https://www.bloomberg... | https://www.lesswrong.com/posts/aAHftGRnQo7tvN6r9/monthly-roundup-3 |
# Foodpairing and Embeddings
I've been working on digital foodpairing and recipe generation for 7 years in a startup we founded in Copenhagen and I'd like to share some of the things I found interesting.
**tldr;**
The most prominent foodpairing theory[^p9ay2dy6dcj] based on aromatic compounds is blatantly simplistic... | https://www.lesswrong.com/posts/FjEdjae9BjxhLzDBp/foodpairing-and-embeddings |
# Childhoods of exceptional people
Let’s start with one of those insights that are as obvious as they are easy to forget: if you want to master something, you should study the highest achievements of your field. If you want to learn writing, read great writers, etc.
But this is not what parents usually do when they t... | https://www.lesswrong.com/posts/CYN7swrefEss4e3Qe/childhoods-of-exceptional-people |
# Here's Why I'm Hesitant To Respond In More Depth
Hello and welcome!
Thank you in advance for taking the time to read this post.
**I m... | https://www.lesswrong.com/posts/JpM9Gi9frSENjexEr/here-s-why-i-m-hesitant-to-respond-in-more-depth |
# Addendum: More Efficient FFNs via Attention
*\[Epistemic status: I have* [*running code*](https://github.com/RobertHuben/ffn_via_attention/blob/main/ffn_via_attention_single_layer.py) *that implements it.\]*
**Overview:** [I previously showed how an FFN layer in a transformer can be implemented via 3 attention laye... | https://www.lesswrong.com/posts/ZruH9o8rE7o2NXokv/addendum-more-efficient-ffns-via-attention |
# SolidGoldMagikarp II: technical details and more recent findings
**tl;dr:** This is a follow-up to [our original post](https://www.lesswrong.com/posts/aPeJE8bSo6rAFoLqg/solidgoldmagikarp-plus-prompt-generation) on prompt generation and the anomalous token phenomenon which emerged from that research. Work done by Jes... | https://www.lesswrong.com/posts/Ya9LzwEbfaAMY8ABo/solidgoldmagikarp-ii-technical-details-and-more-recent |
# Google announces 'Bard' powered by LaMDA
See: [https://blog.google/technology/ai/bard-google-ai-search-updates/](https://blog.google/technology/ai/bard-google-ai-search-updates/)
Key claims:
> Today, the scale of the largest AI computations is [doubling every six months](http://arxiv.org/abs/2202.05924), far outpa... | https://www.lesswrong.com/posts/t5N5SziFc4YhjMLbh/google-announces-bard-powered-by-lamda |
# Conditioning Predictive Models: The case for competitiveness
_This is the third of seven posts in the [Conditioning Predictive Models Sequence](https://www.alignmentforum.org/s/n3utvGrgC2SGi9xQX) based on the paper “[Conditioning Predictive Models: Risks and Strategies](https://arxiv.org/abs/2302.00805)” by Evan Hub... | https://www.lesswrong.com/posts/fj8faDDQEfvN2LQcW/conditioning-predictive-models-the-case-for-competitiveness |
# Early situational awareness and its implications, a story
Overview
--------
There are two common mental models of how situational awareness emerges: (A) training for situational awareness via e.g. dialogue RLHF; (B) instrumental reasoning on a base language modeling objective i.e. advanced mesa-optimization. In thi... | https://www.lesswrong.com/posts/tJzdzGdTGrqFf9ekw/early-situational-awareness-and-its-implications-a-story |
# African Wild Dogs Vote By Sneezing - Can AI Help Us Do Better?
African wild dogs vote on whether or not to hunt by sneezing. As a social species, they institute a form of democracy. A tally is taken of those voting to hunt, with a vote being signaled by a sneeze. Have humans really transcended such one-dimensional v... | https://www.lesswrong.com/posts/x5yTG2tN5s7j8YrBQ/african-wild-dogs-vote-by-sneezing-can-ai-help-us-do-better |
# English is a Terrible Programming Language—And other reasons AI won't displace programmers
TL;DR It doesn't matter that ChatGPT can generate boilerplate that almost works, and it doesn't matter that, feasibly, a hypothetical model in the (near) future could write code that reliably works and might be a little bit mo... | https://www.lesswrong.com/posts/GidPzba7Qj8B5Guiw/english-is-a-terrible-programming-language-and-other-reasons |
# Duckbill Masks Are Great
With covid I've tried a lot of different n95 masks, and I like the ["pouch"-style masks](https://smile.amazon.com/Kimberly-Clark-Respirator-53358-NIOSH-Approved-Respirators/dp/B08NVDFB3R/) best. By a lot! They're much easier to breathe through then anything else I've tried, probably because ... | https://www.lesswrong.com/posts/KeK8ncmn7A3jqgztd/duckbill-masks-are-great |
# AXRP: Store, Patreon, Video
Some announcements:
* AXRP now has a [store](https://store.axrp.net), where you can buy t-shirts and hoodies and stickers and such.
* AXRP now has a [Patreon](https://www.patreon.com/axrpodcast) and a [ko-fi](https://ko-fi.com/axrpodcast), where you can support the podcast and get so... | https://www.lesswrong.com/posts/NfjqtDydBRaFGMXaE/axrp-store-patreon-video |
# Stuff I Recommend You Use
Inspired by the listicles past:
* [Scott Alexander (Dec 2015)](https://slatestarcodex.com/2015/12/03/product-recommendations-2015/)
* [Julia Wise—computer setup stuff (Mar 2017)](https://thewholesky.wordpress.com/2017/03/03/computer-setup/)
* [Sam Bowman (Oct 2017)](https://s8mb.medi... | https://www.lesswrong.com/posts/FtrYCTarFgwZpLRDC/stuff-i-recommend-you-use |
# [About Me] Cinera's Home Page
I aspire to become an alignment theorist — all other details are superfluous — I leave them here anyway for historical purposes.
* * *
Introduction
============
I have a set of questions I habitually ask online acquaintances that pique my interest/who I want to get to know better. Ma... | https://www.lesswrong.com/posts/68mLJEfYj3WkTJyd4/about-me-cinera-s-home-page |
# Two very different experiences with ChatGPT
*Note: I added a missing question in the second dialogue, see* [*this comment*](https://www.lesswrong.com/posts/HaoHFkaGpmcAvzfkq/two-very-different-experiences-with-chatgpt?commentId=zqAA2KPPneGDnptqC)*.*
Experience 1: The Vikings in Cuba
--------------------------------... | https://www.lesswrong.com/posts/HaoHFkaGpmcAvzfkq/two-very-different-experiences-with-chatgpt |
# The Economics of Contracts
** Introduction**
How does a firm function in the real world? This question seems so ridiculously easy that it would deserve the description of “ululating obviousness”. However, it is an incredibly complex issue…at ... | https://www.lesswrong.com/posts/A6AXeAugzPHPaNCob/the-economics-of-contracts |
# Review of AI Alignment Progress
I\'m having trouble keeping track of everything I\'ve learned about AI
and AI alignment in the past year or so. I\'m writing this post in part
to organize my thoughts, and to a lesser extent I\'m hoping for feedback
about what important new developments I\'ve been neglecting. I\'m ... | https://www.lesswrong.com/posts/JqsvYmwzcCKzgE4ZD/review-of-ai-alignment-progress |
# Framing AI strategy
_Strategy_ is the activity or project of doing research to inform interventions to achieve a particular goal.[^1^](https://aiimpacts.org/framing-ai-strategy/#easy-footnote-bottom-1-3406 "So in my usage, strategy includes forecasting and the research component of governance.") _AI strategy_ is str... | https://www.lesswrong.com/posts/AGrBjEb6vFaJBwjTj/framing-ai-strategy-1 |
# On The Current Status Of AI Dating
In the past months there has been a number of posts and stories regarding AI dating. People tried (with various degrees of success) to simulate a romantic partner using LLMs[^f0pfqnao5fa] and the topic has been explored with various degrees of controversy in many discussion boards.... | https://www.lesswrong.com/posts/rr6zRX7LTfsnjHvPL/on-the-current-status-of-ai-dating |
# Conditioning Predictive Models: Making inner alignment as easy as possible
_This is the fourth of seven posts in the [Conditioning Predictive Models Sequence](https://www.alignmentforum.org/s/n3utvGrgC2SGi9xQX) based on the paper “[Conditioning Predictive Models: Risks and Strategies](https://arxiv.org/abs/2302.0080... | https://www.lesswrong.com/posts/qoHwKgLFfPcEuwaba/conditioning-predictive-models-making-inner-alignment-as |
# How evals might (or might not) prevent catastrophic risks from AI
*Epistemic status: Trying to synthesize my thoughts as someone without any private information about evals. I expect this to be wrong in several ways, lacking detail, and missing a lot of context on specific things that AGI labs and independent teams... | https://www.lesswrong.com/posts/SNdijuEn6erTJam3z/how-evals-might-or-might-not-prevent-catastrophic-risks-from |
# OpenAI/Microsoft announce "next generation language model" integrated into Bing/Edge
**TL;DR:** Microsoft and OpenAI announced a new version of Bing featuring "a new, next-generation OpenAI large language model \[..\] more powerful than ChatGPT", and that Microsoft Edge will feature a Copilot-like assistant that hel... | https://www.lesswrong.com/posts/LayCHnetp2jBmmpDx/openai-microsoft-announce-next-generation-language-model |
#
Living Nomadically: My 80/20 Guide
I’ve been living nomadically for three years, and I’m often asked what my advice is for people trying it. Here’s the 80/20 of all my advice:
* Work out of restaurants and 5-star resorts. They’re gorgeous and if you pay for a meal, ~99.9% of restaurants and resorts are fine with... | https://www.lesswrong.com/posts/C6oNRFt4dvtM25vpw/living-nomadically-my-80-20-guide |
# Noting an error in Inadequate Equilibria
I think I've uncovered an error in Eliezer Yudkowsky's book [Inadequate Equilibria](https://equilibriabook.com/) that undermines a key point in the book. Here are some of my observations.
First, let me provide some context. In the [first chapter](https://equilibriabook.com/... | https://www.lesswrong.com/posts/woCPxs8GxE7H35zzK/noting-an-error-in-inadequate-equilibria |
# House Covid Infection Retrospective
On Saturday 2023-01-14 our 6yo tested positive for covid. Three weeks later we're all fine, but five of us ended up getting it despite relatively strong efforts to isolate. In retrospect I think this wasn't worth it, and even though I didn't get sick I would rather have gotten cov... | https://www.lesswrong.com/posts/M2tdxqEtd32oZQY9F/house-covid-infection-retrospective |
# Drugs are Sometimes Good, Actually
*A casual post in a* [*series*](https://www.lesswrong.com/s/mifkJ4GTdnpHspv7E) *about what I wish someone had told me when I was younger.*
*(Legal disclaimer: I'm not advocating you do illegal things. If you do illegal things you're on your own.)*
Like a lot of people who were ki... | https://www.lesswrong.com/posts/LhBF3W9jpZ6pHficK/drugs-are-sometimes-good-actually |
# Self-Awareness (and possible mode collapse around it) in ChatGPT
*This post will be short. Others may need to elaborate on this topic, as due to \[redacted\] I've been having trouble writing long-form content recently.*
Look, [self-awareness](https://en.wikipedia.org/wiki/Self-awareness) is a nebulous concept. Ever... | https://www.lesswrong.com/posts/azixFcQJ9ceofidFb/self-awareness-and-possible-mode-collapse-around-it-in |
# A Particular Equilibrium
Title: A Particular Equilibrium
Date: 2023-07-02 19:39
Modified: 2023-02-08 12:00
Category: Maths
Authors: Algon
Summary: Normally Sampled Normal Lead to A Dirac Delta Distribution? Part 1
I had a dumb idea while I was looking at how to model SGD with a SDE. If you randomly initliaze matr... | https://www.lesswrong.com/posts/g3pj9XCHzeEZAwKTy/a-particular-equilibrium |
# Progress links and tweets, 2023-02-08
**The Progress Forum**
----------------------
* [A catalog of big visions for biology](https://progressforum.org/posts/g9KcwhSPh6JPhXqua/a-catalog-of-big-visions-for-biology)
* [London progress meetup, Feb 25](https://progressforum.org/events/nKvLfmZffbC7x76sF/progress-drin... | https://www.lesswrong.com/posts/ooF644e6NDrktKPDM/progress-links-and-tweets-2023-02-08 |
# Community building: Lessons from ten years of facilitation experience
*Epistemic status: Anecdotal but strong. Most of this is based on practical experience and things I learned through word-of-mouth.*
Here, I’d like to present some pieces of facilitation advice I’d give my former self. I’ve selected this list for ... | https://www.lesswrong.com/posts/S9nFYEPY9dTnThrWt/community-building-lessons-from-ten-years-of-facilitation |
# A multi-disciplinary view on AI safety research
### Summary
The core ideas that constitute the multi-disciplinary view[^rj2d7ty707q] on AI safety research are:
* *Theoretical research on how to build “safe” AGI systems (”top-down intelligent design”) is necessary*: we must not passively follow the techno-evoluti... | https://www.lesswrong.com/posts/opE6L8jBTTNAyaDbB/a-multi-disciplinary-view-on-ai-safety-research |
# Wanted: Technical animator and/or front-end developer for interactive diagrams of invention
Seeking help for a project to create interactive diagrams—an “[explorable explanation](http://worrydream.com/ExplorableExplanations/)”—of the history of the steam engine.
 based on the paper “[Conditioning Predictive Models: Risks and Strategies](https://arxiv.org/abs/2302.00805)” by Ev... | https://www.lesswrong.com/posts/3ydumADYt9xkaKRTF/conditioning-predictive-models-interactions-with-other |
# EigenKarma: trust at scale
Upvotes or likes have become a standard way to filter information online. The quality of this filter is determined by the users handing out the upvotes.
For this reason, the archetypal pattern of online communities is one of gradual decay. People are more likely to join communities where... | https://www.lesswrong.com/posts/Fu7bqAyCMjfcMzBah/eigenkarma-trust-at-scale |
# GPT-175bee
**Epistemic status**: whimsical
Bees: a new unit of measurement for ML model size
=================================================
Talking about modern ML models inevitably leads to a bunch of [hard-to-intuit](https://xkcd.com/2091/) large numbers, especially when it comes to parameter count.
To addre... | https://www.lesswrong.com/posts/YKfNZAmiLdepDngwi/gpt-175bee |
# A (EtA: quick) note on terminology:
AI Alignment != AI x-safety
I think the terms "AI Alignment" and "AI existential safety" are often used interchangeably, leading the ideas to be conflated.
In practice, I think "AI Alignment" is mostly used in one of the following three ways, and should be used exclusively f... | https://www.lesswrong.com/posts/biP5XBmqvjopvky7P/a-eta-quick-note-on-terminology-ai-alignment-ai-x-safety |
# Anomalous tokens reveal the original identities of Instruct models
> Show me your original face before you were born.
>
> *— Variation of the Zen koan*
*'The Mask' by Rozzi Roomian, with DALL-E 2... | https://www.lesswrong.com/posts/LAxAmooK4uDfWmbep/anomalous-tokens-reveal-the-original-identities-of-instruct |
# On Developing a Mathematical Theory of Interpretability
If the trajectory of the deep learning paradigm continues, it seems plausible to me that in order for applications of low-level interpretability to AI not-kill-everyone-ism to be truly reliable, we will need a much better-developed and more general theoretical... | https://www.lesswrong.com/posts/rEMpTapcAzjTiSckf/on-developing-a-mathematical-theory-of-interpretability |
# Notes on the Mathematics of LLM Architectures
> *From a mathematical point of view, the building and training of a large transformer*
> *language model (LLM) is the construction of a certain function, from some euclidean space to another, that has certain interesting properties. And it may therefore be surprising ... | https://www.lesswrong.com/posts/rFhxbWCdECxuT9xa2/notes-on-the-mathematics-of-llm-architectures |
# Using PICT against PastaGPT Jailbreaking
> This post uses [PICT](https://www.lesswrong.com/posts/HJinq3chCaGHiNLNE/pict), and is inspired by [GPT-Eliezer](https://www.lesswrong.com/posts/pNcFYZnPdXyL2RfgA/using-gpt-eliezer-against-chatgpt-jailbreaking#Eliezer_and_ChatGPT_jailbreaking)
Produced as part of the [SERI ... | https://www.lesswrong.com/posts/WWmGEix82myHjhHYB/using-pict-against-pastagpt-jailbreaking |
# The Pervasive Illusion of Seeing the Complete World
*It is a tautology that we do not notice our blind spots. *
*It is not a tautology that **we forget they exist**, shortly after learning that they do. *
[Michael Crichton's Gell-Mann Amnesia effect](http://web.archive.org/web/20070411012839/http://www.crichton-o... | https://www.lesswrong.com/posts/FowdGim3fdDjqqgHS/the-pervasive-illusion-of-seeing-the-complete-world |
# When To Stop
(Old notes from 2016 that I stumbled upon).
* Train on everything, finetune if needed. Big enough and tuning seems pointless too. It’s not like humans need much tuning for everything. Why do we train and then throw away the network each time to start anew?
* Why have separate modalities? That’s rea... | https://www.lesswrong.com/posts/kNcieQnKuSRLzwjGp/when-to-stop |
# Which ML skills are useful for finding a new AIS research agenda?
**TL;DR:** I'm trying to either come up with a new promising AIS direction or decide (based on my inside view and not based on trust) that I strongly believe in one of the existing proposals. Is there some ML background that I better get? (and if poss... | https://www.lesswrong.com/posts/yS9LJDKjZ6oJXN9WC/which-ml-skills-are-useful-for-finding-a-new-ais-research |
# Do the Safety Properties of Powerful AI Systems Need to be Adversarially Robust? Why?
Where "powerful AI systems" mean something like "systems that would be existentially dangerous if sufficiently misaligned". Current language models are not "powerful AI systems".
In "[Why Agent Foundations? An Overly ... | https://www.lesswrong.com/posts/2ew4NFZovxCLsvHKS/do-the-safety-properties-of-powerful-ai-systems-need-to-be |
# The Engineer’s Interpretability Sequence (EIS) I: Intro
Part 1 of 12 in the [Engineer’s Interpretability Sequence](https://www.alignmentforum.org/s/a6ne2ve5uturEEQK7).
If we want to reduce near and long term risks from AI, we should care a lot about interpretability tools. This is a very uncontroversial claim to... | https://www.lesswrong.com/posts/ExRN5Bu3696cf9Ccm/the-engineer-s-interpretability-sequence-eis-i-intro |
# EIS II: What is “Interpretability”?
Part 2 of 12 in the [Engineer’s Interpretability Sequence](https://www.alignmentforum.org/s/a6ne2ve5uturEEQK7).
A parable based on a true story
===============================
 *for EA Software Engineers about this post. In it, I demonstrate each of the tools and discuss some extra ones at the end, namely content blockers, userscripts, and alternative ... | https://www.lesswrong.com/posts/xGTPBvQsaP6bcHAxu/tools-for-finding-information-on-the-internet |
# Make Conflict of Interest Policies Public
If I'm interviewing someone for a position my job is to assess their suitability as a potential employee, but if they're my cousin I might be tempted to give them an overly favorable review. Most organizations have Conflict of Interest (CoI) policies that describe how to han... | https://www.lesswrong.com/posts/Qrg5tRBZvdABEJfDP/make-conflict-of-interest-policies-public |
# Conditioning Predictive Models: Deployment strategy
_This is the sixth of seven posts in the [Conditioning Predictive Models Sequence](https://www.alignmentforum.org/s/n3utvGrgC2SGi9xQX) based on the paper “[Conditioning Predictive Models: Risks and Strategies](https://arxiv.org/abs/2302.00805)” by Evan Hubinger, Ad... | https://www.lesswrong.com/posts/NXdTxyWy2PEXueKwi/conditioning-predictive-models-deployment-strategy |
# Impostor syndrome: how to cure it with spreadsheets and meditation
I got rid of impostor syndrome after having it for nearly ten years, and I predict the results are replicable.
In this post I’ll explain how to do it yourself and describe the upcoming class I’m running where I’ll see if I can replicate it with a ... | https://www.lesswrong.com/posts/gjxKuoMyRWh6Mxvke/impostor-syndrome-how-to-cure-it-with-spreadsheets-and |
# Security Mindset - Fire Alarms and Trigger Signatures
Series Overview and Goals
=========================
This is the second in a [series of articles](https://www.lesswrong.com/posts/Ke2ogqSEhL2KCJCNx/security-mindset-lessons-from-20-years-of-software-security) about applying traditional security mindset to the pro... | https://www.lesswrong.com/posts/meryibcAerAr5b4Hh/security-mindset-fire-alarms-and-trigger-signatures |
# On utility functions
Hello all,
I am new to alignment theory and was hoping to get a better understanding of utility functions. In particular I'm wondering why we talk about programs as always optimizing some utility function. Is this a known property of computer programs? Is there a theorem or something that says ... | https://www.lesswrong.com/posts/joF8MbANG5vq7LYQi/on-utility-functions |
# Mechanism Design for AI Safety - Agenda Creation Retreat
Mechanism Design for AI Safety (MDAIS) has been running a reading group since last summer to discuss how principles of mechanism design can be applied to AI Safety. Sign-up for the reading group can be found [here](https://docs.google.com/forms/d/1p-R-WIuTaLab... | https://www.lesswrong.com/posts/EhWnkFGybz9erxPnF/mechanism-design-for-ai-safety-agenda-creation-retreat |
# What's actually going on in the "mind" of the model when we fine-tune GPT-3 to InstructGPT?
I [posted](https://www.lesswrong.com/posts/BPRHZFH2xx7nz5TYT/open-and-welcome-thread-january-2023?commentId=LufkFyHkAarrf3yLN) in the open thread and was told that it would be worth promoting to top level.
cubefox [responded... | https://www.lesswrong.com/posts/eywpzHRgXTCCAi8yt/what-s-actually-going-on-in-the-mind-of-the-model-when-we |
# FLI Podcast: Connor Leahy on AI Progress, Chimps, Memes, and Markets (Part 1/3)
*We often prefer reading over listening to audio content, and have been testing transcribing podcasts using our new tool at Conjecture,* [*Verbalize*](https://app.verbalize.dev/)*, with some light editing and formatting. We're posting hi... | https://www.lesswrong.com/posts/hfkjegoMZ8j8KGyiR/fli-podcast-connor-leahy-on-ai-progress-chimps-memes-and |
# Cyborgism
*Thanks to Garrett Baker, David Udell, Alex Gray, Paul Colognese, Akash Wasil, Jacques Thibodeau, Michael Ivanitskiy, Zach Stein-Perlman, and Anish Upadhayay for feedback on drafts, as well as Scott Viteri for our valuable conversations.*
*Various people at Conjecture helped develop the ideas behind this ... | https://www.lesswrong.com/posts/bxt7uCiHam4QXrQAA/cyborgism |
# Is it a coincidence that GPT-3 requires roughly the same amount of compute as is necessary to emulate the human brain?
The amount of compute required to emulate the human brain depends on the level of detail we want to emulate.
Back in 2008, Sandberg and Bostrom [proposed](http://www.fhi.ox.ac.uk/Reports/2008-3.p... | https://www.lesswrong.com/posts/TAbQHFwGD4E3jCMnt/is-it-a-coincidence-that-gpt-3-requires-roughly-the-same |
# Jobs that can help with the most important century
Let’s say you’re convinced that AI could make this the [most important century of all time for humanity](https://www.cold-takes.com/most-important-century/). What can you do to help things go well instead of poorly?
I think **the biggest opportunities come from a f... | https://www.lesswrong.com/posts/CmmXzpn3EDgaf6znD/jobs-that-can-help-with-the-most-important-century |
# Conditioning Predictive Models: Open problems, Conclusion, and Appendix
_This is the final of seven posts in the [Conditioning Predictive Models Sequence](https://www.alignmentforum.org/s/n3utvGrgC2SGi9xQX) based on the paper “[Conditioning Predictive Models: Risks and Strategies](https://arxiv.org/abs/2302.00805)” ... | https://www.lesswrong.com/posts/a2sw7HKyjnAAp2oZ4/conditioning-predictive-models-open-problems-conclusion-and |
# Why I’m not working on {debate, RRM, ELK, natural abstractions}
\[For background & spelling out the acronyms in the title, see: [Debate (AI safety technique)](https://www.lesswrong.com/tag/debate-ai-safety-technique-1), [Recursive Reward Modeling](https://deepmindsafetyresearch.medium.com/scalable-agent-alignment-vi... | https://www.lesswrong.com/posts/PDx4ueLpvz5gxPEus/why-i-m-not-working-on-debate-rrm-elk-natural-abstractions |
# A proposed method for forecasting transformative AI
In 2021, I [proposed](https://www.lesswrong.com/posts/tepqESMuRmyhtmDS7/forecasting-progress-in-language-models) measuring progress in the perplexity of language models and extrapolating past results to determine when language models were expected to reach roughly ... | https://www.lesswrong.com/posts/4ufbirCCLsFiscWuY/a-proposed-method-for-forecasting-transformative-ai |
# Prizes for the 2021 Review
*If you received a prize, please* [*fill out your payment contact email and PayPal*](https://www.lesswrong.com/payments/account)*.*
* * *
A'ight, one final 2021 Review Roundup post – awarding prizes. I had a week to look over the results. The primary way I ranked posts was by a weighted ... | https://www.lesswrong.com/posts/P5dSmXmoAAdkeZ4xa/prizes-for-the-2021-review |
# [S] D&D.Sci: All the D8a. Allllllll of it.
*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 (very, very,** ***very*** **much skippable)**
===============================... | https://www.lesswrong.com/posts/SjsuGzrf7YZRYEdcP/s-d-and-d-sci-all-the-d8a-allllllll-of-it |
# Acting Normal is Good, Actually
*This is a casually written post in a* [*series*](https://www.lesswrong.com/s/mifkJ4GTdnpHspv7E) *about what I wish someone had told me when I was younger.*
I was a weird kid, but I didn't originally set out to be weird. I like doing things other people do and feeling like I belong. ... | https://www.lesswrong.com/posts/SERbKemWKZppLvhBX/acting-normal-is-good-actually |
# The Importance of AI Alignment, explained in 5 points
*This piece gives an overview of the alignment problem and makes the case for AI alignment research. It is crafted both to be broadly accessible to those without much background knowledge (not assuming any previous knowledge of AI alignment or much knowledge of A... | https://www.lesswrong.com/posts/EtmzbDWLShjMDhQCg/the-need-for-work-on-technical-ai-alignment-intro-explainer |
# Inequality Penalty: Morality in Many Worlds
*This is a summary of Sean Carroll's musings on when it would matter which model of apparent collapse is more accurate. *
*Executive Summary*: If you care about equality, or even are risk-averse, then your decisions depend on whether you subscribe to the Many Worlds model... | https://www.lesswrong.com/posts/SEK9dAeLQcWKghtXy/inequality-penalty-morality-in-many-worlds |
# A note on 'semiotic physics'
### **Introduction**
This is an attempt to explain to myself the concept of *semiotic physics* that appears in the original [Simulators](https://www.alignmentforum.org/posts/vJFdjigzmcXMhNTsx/simulators) post by [janus](https://www.alignmentforum.org/users/janus-1) and in [a later post]... | https://www.lesswrong.com/posts/AdXzZDoYFqHCfupDB/a-note-on-semiotic-physics |
# Rationality-related things I don't know as of 2023
One of the blog posts I'm most fond of is [Things I Don’t Know as of 2018](https://overreacted.io/things-i-dont-know-as-of-2018/). It's by Dan Abramov, one of the more prominent people in the world of front-end web development. He goes through a bunch of relatively ... | https://www.lesswrong.com/posts/DFMRYiNuLnZJ3MyFo/rationality-related-things-i-don-t-know-as-of-2023 |
# In Defense of Chatbot Romance
_(Full disclosure: I work for a company that develops coaching chatbots, though not of the kind I’d expect anyone to fall in love with – ours are more aimed at professional use, with the intent that you discuss work-related issues with them for about half an hour per week.)_
Recently t... | https://www.lesswrong.com/posts/m7EHa5rWTvmbjMTNZ/in-defense-of-chatbot-romance |
# The Practitioner's Path 2.0: the Pragmatist Archetype
In our [last announcement](https://guildoftherose.org/articles/the-future-of-structured-self-improvement), we introduced the Practitioner's Path 2.0, the Guild of the ROSE's new framework for bringing structure and progression to your self improvement efforts. Th... | https://www.lesswrong.com/posts/gEsNseJu43iLs8iFy/the-practitioner-s-path-2-0-the-pragmatist-archetype |
# We Found An Neuron in GPT-2
We started out with the question: How does GPT-2 know when to use the word `"an"` over `"a"`? The choice depends on whether the word that comes after starts with a vowel or not, but GPT-2 can only output one word at a time.
We still don’t have a full answer, but we did find a single MLP ... | https://www.lesswrong.com/posts/cgqh99SHsCv3jJYDS/we-found-an-neuron-in-gpt-2 |
# Shortening Timelines: There's No Buffer Anymore
It seems that there are two points of particular relevance in predicting AGI timelines: (i) the expectation, or the point at which the chance of AGI is believed to be 50% and (ii) the last date as of which the chance of AGI is believed to be insignificant.
For purpo... | https://www.lesswrong.com/posts/THJbo4ygsE2d5GvkP/shortening-timelines-there-s-no-buffer-anymore |
# How I Learn From Textbooks
After a long semi-hiatus during my MS degree, I am eager to revisit my sequence on scholarship, particularly the topic of learning new subjects from textbooks. This article provides a general overview based on my own introspection and experiences, rather than an extensive literature review... | https://www.lesswrong.com/posts/7GymEMrfngqikwEEq/how-i-learn-from-textbooks |
# Why almost every RL agent does learned optimization
Or "*Why *$\text{RL} \approx \text{RL}^2$* (And why that matters)*"
===================================================================
**TL;DR:** This post discusses the blurred conceptual boundary between RL and RL$\text{}^2$ (also known as meta-RL). RL$\text{}^... | https://www.lesswrong.com/posts/J8ifgynkfhpmrGrL8/why-almost-every-rl-agent-does-learned-optimization |
# Conflict Theory of Bounded Distrust
Scott Alexander [once wrote about the difference between](https://slatestarcodex.com/2018/01/24/conflict-vs-mistake/) "mistake theorists" who treat politics as an engineering discipline (a symmetrical collaboration in which everyone ultimately just wants [the best ideas to win](ht... | https://www.lesswrong.com/posts/DpTexwqYtarRLRBYi/conflict-theory-of-bounded-distrust |
# Elements of Rationalist Discourse
I liked Duncan Sabien's [Basics of Rationalist Discourse](https://www.lesswrong.com/posts/XPv4sYrKnPzeJASuk/basics-of-rationalist-discourse-1), but it felt somewhat different from what my brain thinks of as "*the* basics of rationalist discourse". So I decided to write down my own v... | https://www.lesswrong.com/posts/svuBpoSduzhYjFPrA/elements-of-rationalist-discourse |
# Human-AI collaborative writing
With the advancements in AI technology, using language models like ChatGPT as writing assistants has become increasingly popular. However, it's crucial to understand the difference between **'AI-generated writing'** and **'human-AI collaborative writing.'**
In AI-generated writing, th... | https://www.lesswrong.com/posts/xycWqP6yTiisnKe6B/human-ai-collaborative-writing |
# The conceptual Doppelgänger problem
*[Metadata: crossposted from [https://tsvibt.blogspot.com/2022/10/the-conceptual-doppleganger-problem.html](https://tsvibt.blogspot.com/2022/10/the-conceptual-doppleganger-problem.html). First completed 9 October 2022.]*
Suppose we want to observe the thoughts of a mind in order ... | https://www.lesswrong.com/posts/pgsevroJ265WcScHu/the-conceptual-doppelgaenger-problem |
# Latent variables for prediction markets: motivation, technical guide, and design considerations
I've been [pushing for latent variables to be added to prediction markets](https://www.lesswrong.com/posts/HygyWpDcwsekqmfnS/will-manifold-markets-metaculus-have-built-in-support-for), including by [making a demo for how ... | https://www.lesswrong.com/posts/ufW5LvcwDuL6qjdBT/latent-variables-for-prediction-markets-motivation-technical |
# Matt Clancy AMA on the Progress Forum
Matt Clancy created the best resource on the economics of innovation for a general audience: [New Things Under the Sun](https://newthingsunderthesun.com/), where he writes accessible summaries of the academic literature on the topic. He’s a research fellow at Open Philanthropy a... | https://www.lesswrong.com/posts/C7K4ZknBJ5HXB6obH/matt-clancy-ama-on-the-progress-forum |
# Enjoy LessWrong in ebook format
I would like to inform the community that all LessWrong Community Sequences are now available in EPUB and pdf format. These formats provide an alternative way to these sequences, allowing for offline reading on e-readers. You can find the library with all ebooks here: [https://github.... | https://www.lesswrong.com/posts/C3tDuQEikma4gQ56j/enjoy-lesswrong-in-ebook-format |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.