text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Compute Thresholds: proposed rules to mitigate risk of a “lab leak” accident during AI training runs
1. There should be two thresholds on compute graph size:
1. the Frontier threshold, beyond which oversight during execution is mandatory
2. the Horizon threshold, beyond which execution is forbidden by def... | https://www.lesswrong.com/posts/Zfk6faYvcf5Ht7xDx/compute-thresholds-proposed-rules-to-mitigate-risk-of-a-lab |
# Think like a consultant not a salesperson
In [surveying](https://www.lesswrong.com/posts/55jLckKpmjSfXKjPp/how-to-get-startup-ideas-a-brief-lit-review-and-analysis) the advice out there for how to get startup ideas, one of the big things I found was the idea of "problems not solutions". [MAKE](https://readmake.com/#... | https://www.lesswrong.com/posts/mb5yjnt2GvaBeARrd/think-like-a-consultant-not-a-salesperson |
# Examples of Prompts that Make GPT-4 Output Falsehoods
Post authors: Luke Bailey ([lukebailey@college.harvard.edu](mailto:lukebailey@college.harvard.edu)) and Stephen Casper ([scasper@mit.edu](mailto:scasper@mit.edu))
Project contributors: Luke Bailey, Zachary Marinov, Michael Gerovich, Andrew Garber, Shuvom Sadhuka... | https://www.lesswrong.com/posts/iNvWCTpyEd4zTqjjv/examples-of-prompts-that-make-gpt-4-output-falsehoods-1 |
# Autogynephilia discourse is so absurdly bad on all sides
In [Blanchard's Dangerous Idea and the Plight of the Lucid Crossdreamer](https://www.lesswrong.com/posts/RxxqPH3WffQv6ESxj/blanchard-s-dangerous-idea-and-the-plight-of-the-lucid), Zack criticizes the rationalist community and the trans community for various th... | https://www.lesswrong.com/posts/mvQjpu7iFnqQ47Lbp/autogynephilia-discourse-is-so-absurdly-bad-on-all-sides |
# "Justice, Cherryl."
> Selfishness and altruism are positively correlated within individuals, for the obvious reason.
>
> —[\@InstanceOfClass](https://twitter.com/InstanceOfClass/status/355050621147152386)
## I.
An unfortunate obstacle to appreciating the work of Ayn Rand (as someone who adores the ["sense of life"... | https://www.lesswrong.com/posts/vfjptEJ2oahLqRyZz/justice-cherryl |
# My favorite AI governance research this year so far
I collected my favorite public pieces of research on AI strategy, governance, and forecasting from 2023 so far.
If you're a researcher, I encourage you to make a quick list of your favorite pieces of research, then think about what makes it good and whether you're... | https://www.lesswrong.com/posts/gzJ7QNhd3tCLkbmYC/my-favorite-ai-governance-research-this-year-so-far |
# QAPR 5: grokking is maybe not *that* big a deal?
*\[Thanks to support from* [*Cavendish Labs*](https://cavendishlabs.org/) *and a* [*Lightspeed grant*](https://lightspeedgrants.org/)*, I've been able to restart the* [*Quintin's Alignment Papers Roundup*](https://www.lesswrong.com/s/5omSW4wNKbEvYsyje) *sequence.\]*
... | https://www.lesswrong.com/posts/GpSzShaaf8po4rcmA/qapr-5-grokking-is-maybe-not-that-big-a-deal |
# Please speak unpredictably
Your predictable speech displeases both of us. I finish your sentences and am bothered by inefficiency. Thus, speak/write efficiently, to the point that I can't predict it.
Others may also dislike it as I do, but infer the rest at different levels.
Large audiences lower inference standar... | https://www.lesswrong.com/posts/HWzD5iumbiX3EPKub/please-speak-unpredictably |
# Which rationality posts are begging for further practical development?
Which posts seem to you like they have really important practical implications for how we ought to think and act, but don't seem to come with enough information about how to use the skills, or to develop them, or even what exactly the relevant sk... | https://www.lesswrong.com/posts/bwrKHnoNhuhT79MJW/which-rationality-posts-are-begging-for-further-practical |
# Rationality is not (exactly) Winning
I think "[Rationality is winning](https://www.lesswrong.com/posts/4ARtkT3EYox3THYjF/rationality-is-systematized-winning)" is a bit of a trap.
(The original phrase is notably "rationality is *systematized* winning*",* which is better, but it tends to slide into the abbreviated f... | https://www.lesswrong.com/posts/3GSRhtrs2adzpXcbY/rationality-is-not-exactly-winning |
# Cryonics and Regret
This post is not about arguments in favor of or against cryonics.
I would just like to share a particular emotional response of mine as the topic became hot for me after not thinking about it at all for nearly a decade.
Recently, I have signed up for cryonics, as has my wife, and we have made a... | https://www.lesswrong.com/posts/inARBH5DwQTrvvRj8/cryonics-and-regret |
# [Crosspost] An AI Pause Is Humanity's Best Bet For Preventing Extinction (TIME)
*Otto Barten is director of the* [*Existential Risk Observatory*](https://www.existentialriskobservatory.org/)*, a nonprofit aiming to reduce existential risk by informing the public debate.*
*Joep Meindertsma is founder of PauseAI, a m... | https://www.lesswrong.com/posts/bR8zWoYS9zfha8Hzo/crosspost-an-ai-pause-is-humanity-s-best-bet-for-preventing |
# Brain Efficiency Cannell Prize Contest Award Ceremony
Previously Jacob Cannell wrote the post "Brain Efficiency" which makes several radical claims: that the brain is at the pareto frontier of speed, energy efficiency and memory bandwith, that this represent a fundamental physical frontier.
Here's an AI-generated s... | https://www.lesswrong.com/posts/fm88c8SvXvemk3BhW/brain-efficiency-cannell-prize-contest-award-ceremony |
# A reformulation of Finite Factored Sets
I present my bachelor's thesis that reformulates [Finite Factored Sets](https://arxiv.org/abs/2109.11513):
[Causality with Deterministic Relationships](https://drive.google.com/file/d/1NxDozNx5EwX49mJ_sKhUrmUgxuIuUZoo)
Abstract:
*We present a reformulation of Finite Factor... | https://www.lesswrong.com/posts/nksgLdLc7y5KtbmFw/a-reformulation-of-finite-factored-sets |
# Slowing down AI progress is an underexplored alignment strategy
The emotional burden of watching the world end
==============================================
My current beliefs about AI timelines have made my life significantly worse. I find thoughts about the ever-shrinking timeline to AGI invading nearly every as... | https://www.lesswrong.com/posts/b7JXJWY7R2jNtHerP/slowing-down-ai-progress-is-an-underexplored-alignment |
# An Opinionated Guide to Computability and Complexity (Post #0)
Today, I'll try to introduce certain things in this sequence I will make that I found interesting of computability and complexity theory is, and in particular show how many problems can be solved by computers, in both a philosophical and practical sense.... | https://www.lesswrong.com/posts/y4wcffEugqjqa9Qo3/an-opinionated-guide-to-computability-and-complexity-post-0 |
# Open problems in activation engineering
[Steering GPT-2-XL by adding an activation vector](https://www.lesswrong.com/posts/5spBue2z2tw4JuDCx/steering-gpt-2-xl-by-adding-an-activation-vector) introduced
> **activation engineering**... techniques which steer models by modifying their activations. As a complement to ... | https://www.lesswrong.com/posts/JMebqicMD6azB8MwK/open-problems-in-activation-engineering |
# Secure Hand Holding
When Lily as about three we were waiting at a crosswalk and traffic stopped for us. As we started across the street a driver that had been waiting to turn left misinterpreted the situation and, thinking traffic had stopped for them instead, tried to turn through our location. I tightened my grip ... | https://www.lesswrong.com/posts/aDbWstDcYwsK9ATbv/secure-hand-holding |
# How LLMs are and are not myopic
*Thanks to janus, Nicholas Kees Dupuis, and Robert Kralisch for reviewing this post and providing helpful feedback. Some of the experiments mentioned were performed while at Conjecture.*
**TLDR: The training goal for LLMs like GPT is not cognitively-myopic (because they think about t... | https://www.lesswrong.com/posts/c68SJsBpiAxkPwRHj/how-llms-are-and-are-not-myopic |
# Anthropic Observations
[Dylan Matthews had in depth Vox profile of Anthropic](https://www.vox.com/future-perfect/23794855/anthropic-ai-openai-claude-2), which I recommend reading in full if you have not yet done so. This post covers that one.
#### Anthropic Hypothesis
The post starts by describing an experiment. E... | https://www.lesswrong.com/posts/b3CA5Gst5iWkyuY9L/anthropic-observations |
# AIS 101: Task decomposition for scalable oversight
AGISF Week 4 - Task decomposition for scalable oversight
*This text is an adapted excerpt from the task decomposition for scalable oversight section of the AGISF 2023 course, held at ENS Ulm in Paris on March 16, 2023. **Its purpose is to provide a concise overview... | https://www.lesswrong.com/posts/FFz6H35Gy6BArHxkc/ais-101-task-decomposition-for-scalable-oversight |
# Thoughts on Loss Landscapes and why Deep Learning works
*Crossposted from* [*my personal blog*](https://www.beren.io/2023-07-11-Loss-landscapes-and-understanding-deep-learning/)*.*
**Epistemic status**: *Pretty uncertain. I don’t have an expert level understanding of current views in the science of deep learning ab... | https://www.lesswrong.com/posts/szXa8QgxjMypabJgN/thoughts-on-loss-landscapes-and-why-deep-learning-works |
# Russian parliamentarian: let's ban personal computers and the Internet
Anatoly Kubanov, a senior member of the parliament of a Russian oblast (state), and a member of one of the four major political parties allowed in Russia, proposed to ban personal computers and the public access to the Internet.
Below is a parti... | https://www.lesswrong.com/posts/hN5pvxEYPnWR4rd4G/russian-parliamentarian-let-s-ban-personal-computers-and-the |
# Underwater Torture Chambers: The Horror Of Fish Farming
Crossposted from [my blog](https://benthams.substack.com/p/underwater-torture-chambers).
**The horror of fish farming**
==============================
One of my professors in college taught a class about effective altruism—a social movement about doing good... | https://www.lesswrong.com/posts/Nnis3dFYZRRAzWuEA/underwater-torture-chambers-the-horror-of-fish-farming |
# The First Room-Temperature Ambient-Pressure Superconductor
From the post:
> For the first time in the world, we succeeded in synthesizing the room-temperature superconductor ( K, 127C) working at ambient pressure with a modified lead-apatite (LK-99) structure. The superconductivity of LK-99 is proved with the Crit... | https://www.lesswrong.com/posts/a5wNeyt5yvpEP9BBc/the-first-room-temperature-ambient-pressure-superconductor |
# Frontier Model Security
> As the capabilities of frontier artificial intelligence models continue to increase rapidly, ensuring the security of these systems has become a critical priority. In our previous posts, we’ve focused on Anthropic’s [approach to safety](https://www.anthropic.com/index/core-views-on-ai-safet... | https://www.lesswrong.com/posts/oNGCgNag2zSMqg2Z7/frontier-model-security |
# Llama We Doing This Again?
I’ve finally had an opportunity to gather the available information about Llama-2 and take an in-depth look at the system card.
My conclusion is that Llama-2 looks to score about 3.4 GPTs, with coding as its relative weak point. The system card tries to claim better performance than that ... | https://www.lesswrong.com/posts/rhdAAFLgLBHLtGcNT/llama-we-doing-this-again |
# Frontier Model Forum
Posted by [Anthropic](https://www-files.anthropic.com/production/files/frontier-model-forum.pdf), [Google](https://blog.google/outreach-initiatives/public-policy/google-microsoft-openai-anthropic-frontier-model-forum/), [Microsoft](https://blogs.microsoft.com/on-the-issues/2023/07/26/anthropic-g... | https://www.lesswrong.com/posts/hcWc76eCan3FF5XBk/frontier-model-forum |
# Neuronpedia
**Edit - Neuronpedia has pivoted to be a research tool for Sparse Autoencoders, so most of this post is outdated. Please read the new post,** [**Neuronpedia: Accelerating Sparse Autoencoders Research**](https://www.lesswrong.com/posts/BaEQoxHhWPrkinmxd/announcing-neuronpedia-platform-for-accelerating-res... | https://www.lesswrong.com/posts/skKYznZyRtN87tHbB/neuronpedia |
# Meta-level adversarial evaluation of oversight techniques might allow robust measurement of their adequacy
**Summary:** Many proposed AGI alignment procedures involve taking a pretrained **model** and training it using rewards from an **oversight process** to get a **policy**. These procedures might fail when the ov... | https://www.lesswrong.com/posts/MbWWKbyD5gLhJgfwn/meta-level-adversarial-evaluation-of-oversight-techniques-1 |
# Why no Roman Industrial Revolution?
Why didn’t the Roman Empire have an industrial revolution?
[Bret Devereaux has an essay](https://acoup.blog/2022/08/26/collections-why-no-roman-industrial-revolution/) addressing that question, which multiple people have pointed me to at various times. In brief, Devereaux says th... | https://www.lesswrong.com/posts/yNLrpBBCcJYphjmk9/why-no-roman-industrial-revolution |
# AXRP Episode 23 - Mechanistic Anomaly Detection with Mark Xu
[YouTube link](https://www.youtube.com/watch?v=sEWei02m7qk&list=PLmjaTS1-AiDeqUuaJjasfrM6fjSszm9pK&index=23)
Is there some way we can detect bad behaviour in our AI system without having to know exactly what it looks like? In this episode, I speak with Ma... | https://www.lesswrong.com/posts/jghQXYvifXxhzkhHM/axrp-episode-23-mechanistic-anomaly-detection-with-mark-xu |
# AXRP Episode 24 - Superalignment with Jan Leike
[YouTube link](https://www.youtube.com/watch?v=Uk6-Rw5N_Dg&list=PLmjaTS1-AiDeqUuaJjasfrM6fjSszm9pK&index=24)
Recently, OpenAI made a splash by announcing a new “Superalignment” team. Lead by Jan Leike and Ilya Sutskever, the team would consist of top researchers, atte... | https://www.lesswrong.com/posts/bsNXqHgiDA6dAKNun/axrp-episode-24-superalignment-with-jan-leike |
# Partial Transcript of Recent Senate Hearing Discussing AI X-Risk
On Tuesday, the [US Senate Judiciary Subcommittee on Privacy, Technology and the Law](https://en.wikipedia.org/wiki/United_States_Senate_Judiciary_Subcommittee_on_Privacy,_Technology_and_the_Law) held a hearing on AI. The hearing involved 3 witnesses –... | https://www.lesswrong.com/posts/T5WFE734wHzwMYJaB/partial-transcript-of-recent-senate-hearing-discussing-ai-x |
# Cultivating a state of mind where new ideas are born
In the early 2010s, a popular idea was to provide coworking spaces and shared living to people who were building startups. That way the founders would have a thriving social scene of peers to percolate ideas with as they figured out how to build and scale a ventur... | https://www.lesswrong.com/posts/R5yL6oZxqJfmqnuje/cultivating-a-state-of-mind-where-new-ideas-are-born |
# SSA rejects anthropic shadow, too
*(or: "Please Do Anthropics with Actual Math")*
The anthropic shadow argument states [something like](https://www.lesswrong.com/posts/jE3npTEBtHnZBuAcg/how-many-lhc-failures-is-too-many):
> Anthropic principle! If the LHC had worked, it would have produced a black hole or strangel... | https://www.lesswrong.com/posts/EScmxJAHeJY5cjzAj/ssa-rejects-anthropic-shadow-too |
# AI #22: Into the Weeds
Events continue to come fast and furious. Custom instructions for ChatGPT. Another Congressional hearing, a call to understand what is before Congress and thus an analysis of the bills before Congress. An joint industry safety effort. A joint industry commitment at the White House. Catching up... | https://www.lesswrong.com/posts/MrEmYjMt8qNczeWdv/ai-22-into-the-weeds |
# Preference Aggregation as Bayesian Inference
*Crossposted from* [*my personal blog*](https://www.beren.io/2023-07-13-Preference-Aggregation-as-Bayesian-Inference/)*.*
A fundamental problem in AI alignment, as well as in many social sciences is the problem of *preference aggregation*. Given a number of different ... | https://www.lesswrong.com/posts/KCHQj2ZDAuSZb4Nif/preference-aggregation-as-bayesian-inference |
# A $10k retroactive grant for VaccinateCA
I'm awarding a $10k retroactive grant to Karl Yang, for his role in starting VaccinateCA. In January 2021, the Covid vaccines were just starting to become available to the general public… in theory. In practice, there was a lot of confusion over how to get a dose. Demand for ... | https://www.lesswrong.com/posts/4Q7ygrkWntXGuE9cj/a-usd10k-retroactive-grant-for-vaccinateca |
# Pulling the Rope Sideways: Empirical Test Results
Robin Hanson wrote [this post](https://www.overcomingbias.com/p/policy_tugowarhtml) a while back, which has since developed into a sort of inspiring slogan for how to achieve policy change: "Pull the Rope Sideways"
> The policy world can thought of as consisting of ... | https://www.lesswrong.com/posts/XzS64eQp8fPmCHEdQ/pulling-the-rope-sideways-empirical-test-results |
# A Short Memo on AI Interpretability Rainbows
TL;DR: More work to study interactions and synergies between different AI interpretability tools seems important, tractable, and neglected.
### Interpretable AI research tends to focus on one technique at a time
There is a lot of [interpretable](https://arxiv.org/abs/... | https://www.lesswrong.com/posts/anu2kuJhmnLdkRMjF/a-short-memo-on-ai-interpretability-rainbows |
# ETFE windows
Modern buildings have a lot of windows. If you look at a recent skyscraper, its facade is likely to be mostly glass. Why use glass for that? Why not use a sheet of plastic instead? Glass sheets are heavy and brittle; plastic film would be much lighter and cheaper.
The main problem with plastic film is ... | https://www.lesswrong.com/posts/iPW4cydDfMXvL6yAi/etfe-windows |
# Mech Interp Puzzle 2: Word2Vec Style Embeddings
*Code can be found* [*here*](https://colab.research.google.com/drive/16KwPk0GHoMLfFg2AMTEUttASRKn552TE#scrollTo=PL8yTvoKeeB3&uniqifier=1). *No prior knowledge of mech interp or language models is required to engage with this.*
[Language model embeddings](https://dynal... | https://www.lesswrong.com/posts/hZGoeGdJsnzJbQJMp/mech-interp-puzzle-2-word2vec-style-embeddings |
# Reducing sycophancy and improving honesty via activation steering
*Produced as part of the* [*SERI ML Alignment Theory Scholars Program*](https://serimats.org/) *\- Summer 2023 Cohort, under the mentorship of Evan Hubinger.*
I generate an activation steering vector using Anthropic's [sycophancy dataset](https://hug... | https://www.lesswrong.com/posts/zt6hRsDE84HeBKh7E/reducing-sycophancy-and-improving-honesty-via-activation |
# You don't get to have cool flaws
*This post seems particularly relevant to LW, which to me has a "save the world" culture and is filled with the "misunderstood burden-carrier" trope types. **However:** it has been justly describes as* advice; *please take with a grain of salt.*
**tl;dr:** Solve the flaws you find i... | https://www.lesswrong.com/posts/niPHLviGAWzzZTzit/you-don-t-get-to-have-cool-flaws |
# Progress links digest, 2023-07-28: The decadent opulence of modern capitalism
**Opportunities**
-----------------
* [Applications are still open for The Roots of Progress Blog-Building Intensive](https://fellowship.rootsofprogress.org/), through Aug 11. Launch a blog, get into a regular writing habit, improve you... | https://www.lesswrong.com/posts/yAL6szMiHiMPGTYR8/progress-links-digest-2023-07-28-the-decadent-opulence-of |
# Gradient descent might see the direction of the optimum from far away
*Thanks to Justis, Arthur Conmy, Neel Nanda, Joseph Miller, and Tilman Räuker for their feedback on a draft.*
I feel like many people haven't noticed an important result of [mechanistic interpretability analysis of grokking](https://www.lesswrong... | https://www.lesswrong.com/posts/muLN8GRBdB8NLLX36/gradient-descent-might-see-the-direction-of-the-optimum-from |
# Semaglutide and Muscle
tl;dr Semaglutide seems to have muscle loss effects that may entirely wash out the benefits of Beetus (Type 2) prevention and leave you at a lower bodyweight (but possibly worse off for longevity) than you were before.
There are very few things the rationalist community likes more than a ... | https://www.lesswrong.com/posts/dtcjLNfsLZJizqRKZ/semaglutide-and-muscle |
# Yes, It's Subjective, But Why All The Crabs?
Crabs
-----
[Nature really loves to evolve crabs](https://en.wikipedia.org/wiki/Carcinisation).
![](https://lh6.googleusercontent.com/BG14jAB3I7HOSNtAoTdAZ1rWPLjHxCmDVkHmWN_doyPLsAfdkdzsK3zzrKcgAIj4i51BNrTsoCTzfRX7nyHjZOlqGoJYN7n_BMjMHkYsocs0l1oU5QnlX3aSFL4qq885VAOU4aAy... | https://www.lesswrong.com/posts/qsRvpEwmgDBNwPHyP/yes-it-s-subjective-but-why-all-the-crabs |
# When can we trust model evaluations?
*Thanks to Joe Carlsmith, Paul Christiano, Richard Ngo, Kate Woolverton, and Ansh Radhakrishnan for helpful conversations, comments, and/or feedback.*
In "[Towards understanding-based safety evaluations](https://www.lesswrong.com/posts/uqAdqrvxqGqeBHjTP/towards-understanding-bas... | https://www.lesswrong.com/posts/dBmfb76zx6wjPsBC7/when-can-we-trust-model-evaluations |
# Because of LayerNorm, Directions in GPT-2 MLP Layers are Monosemantic
I think that this has been stated implicitly in various comments and lesswrong posts, but I haven't found a post which makes this point clearly, so thought I would quickly write one up!
To get the most out of this, you should probably be pretty f... | https://www.lesswrong.com/posts/pvCpvMjjyXaAXtnSi/because-of-layernorm-directions-in-gpt-2-mlp-layers-are |
# Thoughts about the Mechanistic Interpretability Challenge #2 (EIS VII #2)
**1) Introduction**
===================
In February, [Stephen Casper](https://www.alignmentforum.org/users/scasper?_ga=2.229981336.547369177.1689719388-1647744316.1684776472) posted [two Mechanistic Interpretability challenges](https://www.le... | https://www.lesswrong.com/posts/vGCWzxP8ccAfqsrS3/thoughts-about-the-mechanistic-interpretability-challenge-2 |
# Why You Should Never Update Your Beliefs
### *Epistemic status: Invincible*
> Since Cavalry scouts are often in direct contact with the enemy, their job can be considered one of the most dangerous jobs the Army has to offer.
>
> — [something called “Operation Military Kids](https://www.operationmilitarykids.org/ar... | https://www.lesswrong.com/posts/LCduhA4m3RhMjZJPA/why-you-should-never-update-your-beliefs |
# Understanding and Aligning a Human-like Inductive Bias with Cognitive Science: a Review of Related Literature
*Produced as part of the* [*SERI ML Alignment Theory Scholars Program*](https://serimats.org/) *with support from* [*Cavendish Labs*](https://cavendishlabs.org/)*, Many thanks go to the following for rea... | https://www.lesswrong.com/posts/J8ZXLTSuFHL27v7P7/understanding-and-aligning-a-human-like-inductive-bias-with |
# Are Guitars Obsolete?
I got an email from a reader:
> I was wondering, like you said that [jams make less sense](https://www.jefftk.com/p/jam-is-obsolete) in the world of easy refrigeration, do you think guitars make less sense when amazing MIDI keyboards are available now?
It's a good question! In general, if an ... | https://www.lesswrong.com/posts/eCHQhLz5QZzdZmWLe/are-guitars-obsolete |
# The Parable of the Dagger - The Animation
This video is an animation of [The Parable of the Dagger](https://www.lesswrong.com/posts/hQxYBfu2LPc9Ydo6w/the-parable-of-the-dagger), by Eliezer Yudkowsky. It's the first article of [A Human's Guide To Words](https://www.lesswrong.com/s/SGB7Y5WERh4skwtnb).
The narration ... | https://www.lesswrong.com/posts/63hx4EKS2PoCrFqmy/the-parable-of-the-dagger-the-animation |
# Self-driving car bets
This month I lost a bunch of bets.
Back in early 2016 I bet at even odds that self-driving ride sharing would be available in 10 US cities by July 2023. Then I made similar bets a dozen times because everyone disagreed with me.
The first deployment to potentially meet our bar was Phoenix in 2... | https://www.lesswrong.com/posts/ZRrYsZ626KSEgHv8s/self-driving-car-bets |
# If I showed the EQ-SQ theory's findings to be due to measurement bias, would anyone change their minds about it?
Background on EQ-SQ
-------------------
The EQ-SQ theory is a theory by Simon Baron-Cohen which states that there's a biological sex difference in tendency to empathize with people (Empathizing) versus t... | https://www.lesswrong.com/posts/SRAmQwqBEh6mk2bKN/if-i-showed-the-eq-sq-theory-s-findings-to-be-due-to |
# How to make real-money prediction markets on arbitrary topics (Outdated)
Intro
=====
On current real-money event market platforms, traders are limited to viewing and betting on markets made by platform operators. This means the questions tend to be confined to popular subjects like politics and sporting outcomes, a... | https://www.lesswrong.com/posts/btpE9fAbvGys4Ztj9/how-to-make-real-money-prediction-markets-on-arbitrary |
# Apollo Neuro Results
Introduction
============
Two months ago I recommended the [Apollo Neuro](https://acesounderglass.com/2023/05/08/product-endorsement-apollo-neuro/) for sleep/anxiety/emotional regulation. A number of people purchased it based on my recommendation- at least 25, according to my referral bonuses. ... | https://www.lesswrong.com/posts/TfkRqndhqK38dxSCb/apollo-neuro-results |
# Rationalization Maximizes Expected Value
***TLDR:** After making a choice, we often change how desirable we think the options were to better align with that choice. This is perfectly rational. How much you’ll enjoy an outcome depends on both (1) what you get, **and** (2) how you feel about it. Once what you get is s... | https://www.lesswrong.com/posts/8anuaBgRtGyH32cnJ/rationalization-maximizes-expected-value |
# EU’s AI ambitions at risk as US pushes to water down international treaty (linkpost)
From [EU’s AI ambitions at risk as US pushes to water down international treaty – EURACTIV.com](https://www.euractiv.com/section/artificial-intelligence/news/eus-ai-ambitions-at-risk-as-us-push-to-water-down-international-treaty/) (... | https://www.lesswrong.com/posts/uwr9bL8GA8uBmzbef/eu-s-ai-ambitions-at-risk-as-us-pushes-to-water-down |
# Is Light Drinking Protective?
There are a lot of claims about how alcohol affects the body, and some sort of "heavy drinking is bad for you but light or moderate drinking is better than no drinking" is a common one. I've not paid a lot of attention to these, however, since non-drinkers as a group include a bunch of ... | https://www.lesswrong.com/posts/rgYtCvRmiARomk2Bi/is-light-drinking-protective |
# "Not Necessarily"
*Epistemic status: Relatively certain, given that the technique is simple and should align well with existing rationality practices. What's presented here is just a reframing of existing frameworks.*
Here's a proposition: Most of epistemic rationality consists of the liberal application of trigger... | https://www.lesswrong.com/posts/FiNWffZ46BQfnoC3q/not-necessarily |
# Open Problems and Fundamental Limitations of RLHF
Reinforcement Learning from Human Feedback (RLHF) has emerged as the central alignment technique used to finetune state-of-the-art AI systems such as GPT-4, Claude, Bard, and Llama-2. Given RLHF's status as the default industry alignment technique, we should carefull... | https://www.lesswrong.com/posts/LqRD7sNcpkA9cmXLv/open-problems-and-fundamental-limitations-of-rlhf |
# Trading off compute in training and inference (Overview)
**Summary:** Some techniques allow to increase the performance of Machine Learning models at the cost of more expensive inference, or reduce inference compute at the cost of lower performance. This possibility induces a tradeoff between spending more resources... | https://www.lesswrong.com/posts/hDuoHuBjxGue7fwhJ/trading-off-compute-in-training-and-inference-overview |
# Thoughts on sharing information about language model capabilities
Core claim
----------
I believe that sharing information about the capabilities and limits of existing ML systems, and especially language model agents, significantly reduces risks from powerful AI—despite the fact that such information may increase ... | https://www.lesswrong.com/posts/fRSj2W4Fjje8rQWm9/thoughts-on-sharing-information-about-language-model |
# Lack of Social Grace Is an Epistemic Virtue
Someone once told me that they thought I acted like refusing to employ the bare minimum of social grace was a virtue, and that this was bad. (I'm paraphrasing; they actually used a different word that starts with _b_.)
I definitely don't want to say that lack of social gr... | https://www.lesswrong.com/posts/h2Hk2c2Gp5sY4abQh/lack-of-social-grace-is-an-epistemic-virtue |
# "Building a House" Review
I was recently reading Byron Barton's 1981 book, [Building a House](https://www.amazon.com/Building-House-Mulberry-Books-Barton/dp/0688093566). While it claims to be an end-to-end overview of the process of modern (for the time) home construction, there are enough errors in the illustration... | https://www.lesswrong.com/posts/kvyHKopJdRKfapxgG/building-a-house-review |
# The "spelling miracle": GPT-3 spelling abilities and glitch tokens revisited
*Work supported by the Long Term Future Fund. Thanks to Jessica Rumbelow and Joseph Bloom for useful discussions.*
Introduction
============
The term "spelling miracle" was coined in Liu et al.'s December 2022 paper "[Character-aware mode... | https://www.lesswrong.com/posts/5sNLX2yY5FzkCp7Ju/the-spelling-miracle-gpt-3-spelling-abilities-and-glitch |
# What The Lord of the Rings Teaches Us About AI Alignment
The Mistake of the *Lord of the Rationality*
--------------------------------------------
In the online version of *Harry Potter and the Methods of Rationality*, there is an extra chapter where Eliezer Yudkowsky gives glimpses into what other rationalist fanf... | https://www.lesswrong.com/posts/PZZsbmJcFoRaxh4uP/what-the-lord-of-the-rings-teaches-us-about-ai-alignment |
# Watermarking considered overrated?
_Status: a slightly-edited copy-paste of [a ~~Twitter~~ X thread](https://twitter.com/dfrsrchtwts/status/1682529184881709056) I quickly dashed off a week or so ago._
Here's a thought I'm playing with that I'd like feedback on: I think [watermarking](https://arxiv.org/abs/2301.1022... | https://www.lesswrong.com/posts/pvE6NdPoMCyk55Lxn/watermarking-considered-overrated |
# A Social History of Truth
This is a chapter-by-chapter summary of [A Social History of Truth](https://www.amazon.com/Social-History-Truth-Seventeenth-Century-Foundations-ebook/dp/B07N85PGQ8) by [Steven Shapin](https://scholar.harvard.edu/shapin/home). Focused on [Robert Boyle](https://en.wikipedia.org/wiki/Robert_Bo... | https://www.lesswrong.com/posts/qGNXJ3ZGpKWEcfoMj/a-social-history-of-truth |
# The “no sandbagging on checkable tasks” hypothesis
*(This post is inspired by *[*Carl Shulman’s recent podcast with Dwarkesh Patel*](https://www.dwarkeshpatel.com/p/carl-shulman?nthPub=521#details)*, which I highly recommend. See also discussion from Buck Shlegeris and Ryan Greenblatt *[*here*](https://www.lesswrong... | https://www.lesswrong.com/posts/h7QETH7GMk9HcMnHH/the-no-sandbagging-on-checkable-tasks-hypothesis |
# The "public debate" about AI is confusing for the general public and for policymakers because it is a three-sided debate
**Summary of Argument: **The public debate among AI experts is confusing because there are, to a first approximation, three sides, not two sides to the debate. I refer to this as a 🔺three-sided ... | https://www.lesswrong.com/posts/BTcEzXYoDrWzkLLrQ/the-public-debate-about-ai-is-confusing-for-the-general |
# Exercise: Solve "Thinking Physics"
*Note: please write any answers to this prompt in spoiler-tags.*
Recently I set out to deliberate practice at "reasoning about confusing intellectual problems."
Eliezer's [Class Project](https://www.lesswrong.com/posts/xAXrEpF5FYjwqKMfZ/class-project) has a fictional group of ra... | https://www.lesswrong.com/posts/PiPH4gkcMuvLALymK/exercise-solve-thinking-physics |
# Evaluating Superhuman Models with Consistency Checks
Consider the following two questions:
|  **Is this move good or bad?** | ... | https://www.lesswrong.com/posts/WEjBvskFwBkczqjZZ/evaluating-superhuman-models-with-consistency-checks-1 |
# What is autonomy, and how does it lead to greater risk from AI?
As with many concepts in discussions of AI risk, terminology around what autonomy is, what agency is, and how they might create risks is deeply confused and confusing, and this is leading to people talking past one another. In this case, the seeming bin... | https://www.lesswrong.com/posts/x8bK7ohAHzMMchsaC/what-is-autonomy-and-how-does-it-lead-to-greater-risk-from |
# AI romantic partners will harm society if they go unregulated
Recently, when people refer to “immediate societal harms and dangers” of AI, in media or political rhetoric, they predominantly choose to mention “bias”, “misinformation”, and “political (election) manipulation”.
Despite politicians, journalists, and exp... | https://www.lesswrong.com/posts/FK8SwbcCq4HqvXmLv/ai-romantic-partners-will-harm-society-if-they-go |
# What Is Childhood Supposed To Be?
I don’t have children.
I’d like to, one day, and I think a lot about what their childhoods might be like, especially as compared to mine.
I look at reports of teen mental health, at college admission rates and tuition levels. I talk to friends of mine who are teachers and involved... | https://www.lesswrong.com/posts/fc8rJhCjzA8dDPoBJ/what-is-childhood-supposed-to-be |
# Untangling Infrabayesianism: A redistillation [PDF link; ~12k words + lots of math]
\[Epistemic status: improved redistillation of the [infrabayesianism sequence](https://www.lesswrong.com/s/CmrW8fCmSLK7E25sa).\]
So you want to understand infrabayesianism, to hack to the center of that thorny wood and seek out and ... | https://www.lesswrong.com/posts/Xne4Bma99bKXWc9AH/untangling-infrabayesianism-a-redistillation-pdf-link-12k |
# Barbieheimer: Across the Dead Reckoning
SPOILER WARNING: This post, after a brief spoiler-free review section, will contain full spoilers for Oppenheimer, Barbie and Mission: Impossible: Dead Reckoning Part One, and some for Across the Spiderverse.
Movies are so back. While they are having their Barbieheimer moment... | https://www.lesswrong.com/posts/6yoehDfWJAgnRHpWo/barbieheimer-across-the-dead-reckoning |
# “Desperate Honesty” by Agnes Callard
An excellent meditation on intellectual humility as an essential tool in the philosopher's toolbox, and how valuable it is to us when people poke holes in our favorite ideas. Excerpts:
> These days I regularly wander outside the gates of philosophy, and it is remarkable to me ho... | https://www.lesswrong.com/posts/yiLyy7nsEEJrhnw2A/desperate-honesty-by-agnes-callard |
# ARC Evals new report: Evaluating Language-Model Agents on Realistic Autonomous Tasks
[Blogpost version](https://evals.alignment.org/blog/2023-08-01-new-report/)
[Paper](https://evals.alignment.org/Evaluating_LMAs_Realistic_Tasks.pdf)
* * *
We have just released our first public report. It introduces methodology f... | https://www.lesswrong.com/posts/EPLk8QxETC5FEhoxK/arc-evals-new-report-evaluating-language-model-agents-on |
# Spiral Staircase
Here was the life cycle of an insight:
> *“If I put this candle in an all-white gallery space, it looks like a piece of art. If I put it in a garage, it looks like a piece of trash. \[...\] I often use this analogy in design. I could either design the candle, \[...\] or I could just design the room... | https://www.lesswrong.com/posts/ygaLpogmcRGvEmKDe/spiral-staircase |
# My current LK99 questions
So this morning I thought to myself, "Okay, now I will actually try to study the LK99 question, instead of betting based on nontechnical priors and market sentiment reckoning." (My initial entry into the affray, having been driven by people online presenting as confidently YES when the pre... | https://www.lesswrong.com/posts/EzSH9698DhBsXAcYY/my-current-lk99-questions |
# What is ontology?
Over the years I've picked up on more and more phrases that people on LessWrong use. However, "ontology" is one of them that I can't seem to figure out. It seems super abstract and [doesn't seem to have](https://www.lesswrong.com/tag/ontology) a reference post.
So then, please [ELI5](https://www.u... | https://www.lesswrong.com/posts/ZG94agRiYbHHcSSmp/what-is-ontology |
# "Is There Anything That's Worth More"
In season two, episode twenty-four of _Steven Universe_, ["It Could've Been Great"](https://steven-universe.fandom.com/wiki/It_Could've_Been_Great), our magical alien superheroine protagonists (and Steven) are taking a break from building a giant drill to extract a superweapon t... | https://www.lesswrong.com/posts/bqZwWQCai6iAjy4Xq/is-there-anything-that-s-worth-more |
# solar-thermal and techno-economic analysis
### techno-economic analysis
There is a genre of engineering paper called "[techno-economic analysis](https://scholar.google.com/scholar?hl=en&as_sdt=0%2C45&q=%22techno-economic+analysis%22&btnG=)". Generally, they involve:
- Listing multiple related designs for accomplish... | https://www.lesswrong.com/posts/jypQzMuLCXwZuErj4/solar-thermal-and-techno-economic-analysis |
# Anthropical Motte and Bailey in two versions of Sleeping Beauty
*This is the first post in my series on Anthropics. The next one is* [*Conservation of Expected Evidence and Random Sampling in Anthropics*](https://www.lesswrong.com/posts/uEP3P6AuNjYaFToft/conservation-of-expected-evidence-and-random-sampling-in)
###... | https://www.lesswrong.com/posts/HQFpRWGbJxjHvTjnw/anthropical-motte-and-bailey-in-two-versions-of-sleeping |
# Long-Term Future Fund: April 2023 grant recommendations
*(Cross-posted from* [*the EA Forum*](https://forum.effectivealtruism.org/posts/zZ2vq7YEckpunrQS4/long-term-future-fund-april-2023-grant-recommendations)*.)*
**Introduction**
================
This payout report is meant to cover the Long-Term Future Fund's gr... | https://www.lesswrong.com/posts/JjLExgd7YKEFf4M4D/long-term-future-fund-april-2023-grant-recommendations |
# 3 levels of threat obfuscation
One of the biggest reasons alignment might be hard is what I’ll call **threat obfuscation:** various dynamics that might make it hard to _measure/notice_ cases where an AI system has problematic misalignment (even when the AI system is in a controlled environment and one is looking for... | https://www.lesswrong.com/posts/HpzHjKjGQ4cKiY3jX/3-levels-of-threat-obfuscation |
# Progress links digest, 2023-08-02: Superconductor edition
**Opportunities**
-----------------
* [Stripe Press is hiring a creative producer for video, audio, & special projects](https://stripe.com/jobs/listing/producer-stripe-press/5047655) (via [@_TamaraWinter](https://twitter.com/_TamaraWinter/status/1683872401... | https://www.lesswrong.com/posts/foqSq4iG9jqgC9uw6/progress-links-digest-2023-08-02-superconductor-edition |
# Work culture creep
*epistemic status: This is not my field of study. I am working through some ideas here that I think might resonate with the community.*
I am a fun, happy, goofy dad. I can play music and perform magic tricks. I sing silly songs and make stupid jokes.
However, for 8 out of the 16 hours that I am ... | https://www.lesswrong.com/posts/i97Ekb3kvJk8aM5SF/work-culture-creep |
# Bad Imitation Instruments
It's common to look down on modern instruments as cheap imitations: the digital piano as an inferior piano, [foot drums](https://www.jefftk.com/p/introduction-to-heel-toe-drumming) as an inferior drum kit. While this isn't wrong, as far is it goes, it's a very limiting way to look at these ... | https://www.lesswrong.com/posts/eCkmhei4ygZY3wsXH/bad-imitation-instruments |
# AI #23: Fundamental Problems with RLHF
After several jam-packed weeks, things slowed down to allow everyone to focus on the potential room temperature superconductor, [check Polymarket to see how likely it is we are so back](https://polymarket.com/event/is-the-room-temp-superconductor-real) and bet real money, [or M... | https://www.lesswrong.com/posts/aKzwwKT2cy72awSyz/ai-23-fundamental-problems-with-rlhf |
# Password-locked models: a stress case for capabilities evaluation
*Thanks to Buck Shlegeris, Ryan Greenblatt, Kshitij Sachan and Jenny Nitishinskaya for helpful discussions and feedback on drafts. This work was done at Redwood Research. The views expressed are my own and do not necessarily reflect the views of the o... | https://www.lesswrong.com/posts/rZs6ddqNnW8LXuJqA/password-locked-models-a-stress-case-for-capabilities |
# [Linkpost] Deception Abilities Emerged in Large Language Models
*This is a linkpost for* [*https://arxiv.org/abs/2307.16513*](https://arxiv.org/abs/2307.16513)*. *
> Large language models (LLMs) are currently at the forefront of intertwining artificial intelligence (AI) systems with human communication and everyday... | https://www.lesswrong.com/posts/5bcKDg6gg5mFxdHbu/linkpost-deception-abilities-emerged-in-large-language |
# Apollo Research is hiring evals and interpretability engineers & scientists
**TL;DR:** You can apply [here](https://jobs.lever.co/apolloresearch).
About Apollo
------------
[Apollo Research](https://apolloresearch.ai/) is a London-based AI safety organization. We focus on high-risk failure modes, particularly dece... | https://www.lesswrong.com/posts/MrdFL38Zi3DwTDkKS/apollo-research-is-hiring-evals-and-interpretability |
# [Linkpost] Multimodal Neurons in Pretrained Text-Only Transformers
*This is a linkpost for* [*https://arxiv.org/abs/2308.01544*](https://arxiv.org/abs/2308.01544)*. *
> Language models demonstrate remarkable capacity to generalize representations learned in one modality to downstream tasks in other modalities. Can ... | https://www.lesswrong.com/posts/wfhnDsavsoxnvGRKD/linkpost-multimodal-neurons-in-pretrained-text-only |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.