text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# rabbit (a new AI company) and Large Action Model (LAM)
Just sharing it here, curious of what the LW community would think of [rabbit](https://www.rabbit.tech/)'s [neuro-symbolic](https://www.rabbit.tech/) approach and their new device "R1"?
*.*
Like anyone with a passing interest in ~~Kerbal Space Program~~ physics and spaceflight... | https://www.lesswrong.com/posts/zLnCs96b4JLwJBCTA/an-actually-intuitive-explanation-of-the-oberth-effect-1 |
# The Aspiring Rationalist Congregation
1. *Meta Note: This post has been languishing in a Google doc for many months as I've procrastinated on cleaning it up to be more coherent and polished. So... I'm posting it as is, with very little cleanup, in the hopes that it's valuable in the current state. I'm sure there ar... | https://www.lesswrong.com/posts/uLF56EHBrMvTXuiyS/the-aspiring-rationalist-congregation-9 |
# The Perceptron Controversy
Connectionism died in the 60s from technical limits to scaling, then resurrected in the 80s after backprop allowed scaling. The Minsky–Papert anti-scaling hypothesis explained, psychoanalyzed, and buried.
I wrote it as if it's a companion post to Gwern's [The Scaling Hypothesis](https://g... | https://www.lesswrong.com/posts/wyF8Acenm8c8fLaq7/the-perceptron-controversy |
# Universal Love Integration Test: Hitler
*I'm still not satisfied with this post, but thought I'd ship it since I refer to the concept a fair amount. I write this more as "someone who feels some kernel of univeral-love-shaped thing", but, like, i dunno man i'm not a love expert.*
**tl;dr**
I think "love" means "To ... | https://www.lesswrong.com/posts/cnwgMibdt8ub6JfTF/universal-love-integration-test-hitler |
# Trying to align humans with inclusive genetic fitness
*Epistemic status: I think this post points to some important ideas, I think the specific proposals might have flaws and there are likely better ideas. If you’re interested, I would be interested in other proposals, or converting standard alignment proposals into... | https://www.lesswrong.com/posts/oAyCLEDNpNFKjRF9H/trying-to-align-humans-with-inclusive-genetic-fitness |
# Prediction markets are consistently underconfident. Why?
look at this calibration s-curve from [calibration.city](https://calibration.city)
. Podcast version [here](https://www.buzzsprout.com/2034731/14292300), or search for "Joe Carlsmith Audio" on your podcast app.*
*This essay is part of a series I'm calling "Otherness and control in
th... | https://www.lesswrong.com/posts/cnv5g7jCWLw9LYKxa/an-even-deeper-atheism-3 |
# A Benchmark for Decision Theories
One of the cottage industries of our community is the construction of new [decision theories](https://www.lesswrong.com/tag/decision-theory). I think our community is also in agreement about how to compare decision theories: the rational way to make decisions is the one that [best e... | https://www.lesswrong.com/posts/R3pWxhWEDdr3YHZYs/a-benchmark-for-decision-theories |
# Apply to the 2024 PIBBSS Summer Research Fellowship
**TLDR**: We're hosting a 3-month, fully-funded fellowship to do AI safety research drawing on inspiration from fields like evolutionary biology, neuroscience, dynamical systems theory, and more. Past fellows have been mentored by John Wentworth, Davidad, Abram Dem... | https://www.lesswrong.com/posts/cQDruLEwuPenmmYbb/apply-to-the-2024-pibbss-summer-research-fellowship |
# introduction to solid oxide electrolytes
## physics
Batteries, fuel cells, and electrolyzers all involve:
- an electrolyte which allows passage of some ions with a single charge state (such as Li+ or Na+)
- material with variable charge state on both sides of the electrolyte (such as Li metal and Fe+3 in LFP batter... | https://www.lesswrong.com/posts/dZLtYgpkqzFAiq4iP/introduction-to-solid-oxide-electrolytes |
# Decent plan prize announcement (1 paragraph, $1k)
Edit Jan 20: [Winner & highlights](https://www.lesswrong.com/posts/fEuZdHbCofHKgjmCQ/decent-plan-prize-winner-and-highlights)
Say I'm about to do a real big training run on playing video games, predicting text, predicting physics, writing code that works, etc etc. S... | https://www.lesswrong.com/posts/GjeprNFWyNPW8oseD/decent-plan-prize-announcement-1-paragraph-usd1k |
# A Chinese Room Containing a Stack of Stochastic Parrots
Most people are familiar with John Searle's Chinese Room argument (["Minds, Brains, and Programs](https://web-archive.southampton.ac.uk/cogprints.org/7150/1/10.1.1.83.5248.pdf)" (1980)):
> Suppose that I'm locked in a room and given a large batch of Chinese wr... | https://www.lesswrong.com/posts/H2G23BxNtuwSpHACL/a-chinese-room-containing-a-stack-of-stochastic-parrots |
# What good is G-factor if you're dumped in the woods? A field report from a camp counselor.
I had a surprising experience with a 10 year old child "Carl" a few years back. He had all the stereotypical signals of a gifted kid that can be drilled into anyone by a dedicated parent- 1500 chess elo, constantly pestered me... | https://www.lesswrong.com/posts/pK3eKhBwBiLffqtrk/what-good-is-g-factor-if-you-re-dumped-in-the-woods-a-field |
# Land Reclamation is in the 9th Circle of Stagnation Hell
Land reclamation is a process where swamps, wetlands, or coastal waters are drained and filled to create more dry land.
Despite being complex and technologically intensive, land reclamation is quite old and was common in the past. The [reclamation of the Dutc... | https://www.lesswrong.com/posts/w9CAJnopAr649B2mF/land-reclamation-is-in-the-9th-circle-of-stagnation-hell |
# Concrete examples of doing agentic things?
I've been thinking about agency recently. **I want to be more agentic!**
Sometime last year, I read [Cultivating And Destroying Agency](https://www.lesswrong.com/posts/rXX6vd2RrKHfePPfE/cultivating-and-destroying-agency) and [Seven ways to become unstoppably agentic](https... | https://www.lesswrong.com/posts/umJmRfcJndY3Gsr36/concrete-examples-of-doing-agentic-things |
# Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
I'm not going to add a bunch of commentary here on top of what we've already put out, since we've put a lot of effort into the paper itself, and I'd mostly just recommend reading it directly, especially since there are a lot of subtle resul... | https://www.lesswrong.com/posts/ZAsJv7xijKTfZkMtr/sleeper-agents-training-deceptive-llms-that-persist-through |
# Throughput vs. Latency
Hi Ruby! Ready to kick things off when you are. I can start by sharing a toy model about productivity inspired by queuing theory, which seems to have been [floating](https://www.lesswrong.com/posts/yLLkWMDbC9ZNKbjDG/slack?commentId=Znws6RRjBdjxZ4kMv) [around](https://www.lesswrong.com/posts/YE... | https://www.lesswrong.com/posts/kvLXmqC3rYJz5S7dc/throughput-vs-latency |
# Commonwealth Fusion Systems is the Same Scale as OpenAI
Artificial intelligence is not the only exciting emerging technology. Another one that I am personally familiar with is fusion.
It seems interesting to compare companies working on emerging technologies to get a feel for how surprising developments in these co... | https://www.lesswrong.com/posts/9qHtQDvfvwmWWDcdQ/commonwealth-fusion-systems-is-the-same-scale-as-openai |
# D&D.Sci(-fi): Colonizing the SuperHyperSphere
*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.*
It had all seemed so promising at first. Colonizing a newly-discovered planet with t... | https://www.lesswrong.com/posts/T3iG4MQ76988JBfkq/d-and-d-sci-fi-colonizing-the-superhypersphere |
# Introducing Alignment Stress-Testing at Anthropic
Following on from our recent paper, “[Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training](https://www.alignmentforum.org/posts/ZAsJv7xijKTfZkMtr/sleeper-agents-training-deceptive-llms-that-persist-through)”, I’m very excited to announce that... | https://www.lesswrong.com/posts/EPDSdXr8YbsDkgsDG/introducing-alignment-stress-testing-at-anthropic |
# Eliminating Cookie Banners is Hard
Application-monitoring company Sentry recently [wrote](https://blog.sentry.io/we-removed-advertising-cookies-heres-what-happened/) about their experience removing cookies on their site, which allowed them to drop their [cookie banner](https://www.jefftk.com/p/why-so-many-cookie-ban... | https://www.lesswrong.com/posts/yMEHopXYe25mHi4cA/eliminating-cookie-banners-is-hard |
# Why do so many think deception in AI is important?
I see a lot of energy and interest being devoted toward detecting deception in AIs, trying to make AIs less deceptive, making AIs honest, etc. But I keep trying to figure out why so many think this is very important. For less-than-human intelligence, deceptive tacti... | https://www.lesswrong.com/posts/CYu6ZB6fFjGh2bAik/why-do-so-many-think-deception-in-ai-is-important |
# Takeaways from the NeurIPS 2023 Trojan Detection Competition
This link summarizes our [research takeaways](https://confirmlabs.org/posts/TDC2023) as participants (and winners in one of the four tracks) of the NeurIPS 2023 Trojan Detection Competition, a competition about red-teaming LLMs and reverse-engineering plan... | https://www.lesswrong.com/posts/fTKFJy2vAz2krfmK8/takeaways-from-the-neurips-2023-trojan-detection-competition |
# AI #47: Meet the New Year
\[NOTE: I forgot to post this to WP/LW/RSS on Thursday, so posting it now. Sorry about that.\]
Will be very different from the old year by the time we are done. This year, it seems like various continuations of the old one. Sometimes I look back on the week, and I wonder how so much happen... | https://www.lesswrong.com/posts/iygs57bHJ36AvzpMh/ai-47-meet-the-new-year |
# Some additional SAE thoughts
*Thanks to Lee Sharkey for feedback on the first section and Lee Sharkey, Jake Mendel, Kaarel Hänni and others at the LISA office for conversations around this post.*
This post is a collection of a few small experiments and thoughts from the last couple of months that never really turne... | https://www.lesswrong.com/posts/fqgn56tS5AgjmDpnX/some-additional-sae-thoughts |
# D&D.Sci Hypersphere Analysis Part 1: Datafields & Preliminary Analysis
This is a post (hopefully eventually a short series of posts) detailing my analysis of the [recently released D&D.Sci scenario](https://www.lesswrong.com/posts/T3iG4MQ76988JBfkq/d-and-d-sci-fi-colonizing-the-superhypersphere) by abstractapplic. ... | https://www.lesswrong.com/posts/rAnS5jCQ5r87eMvus/d-and-d-sci-hypersphere-analysis-part-1-datafields-and |
# Case Studies in Reverse-Engineering Sparse Autoencoder Features by Using MLP Linearization
*Epistemic status: preliminary/exploratory.*
*Work performed as a part of Neel Nanda's MATS 5.0 (Winter 2023-2024) Research Sprint.*
TL;DR: We develop a method for understanding how sparse autoencoder features in transformer... | https://www.lesswrong.com/posts/93nKtsDL6YY5fRbQv/case-studies-in-reverse-engineering-sparse-autoencoder |
# Vote With Your Face
Starting with our July 2nd dance [BIDA](https://www.bidadance.org/) has been running half our dances as [mask-optional](https://blog.bidadance.org/2023/06/some-mask-optional-dances.html). Some people only want to attend if they know everyone on the floor will be wearing a [high-filtration mask](h... | https://www.lesswrong.com/posts/EitMbGASqTAuyWPJA/vote-with-your-face |
# Against most, but not all, AI risk analogies
I personally dislike most AI risk analogies that I've seen people use. While I think analogies can be helpful for explaining concepts to people and illustrating mental pictures, I think they are frequently misused, and often harmful. At the root of the problem is that ana... | https://www.lesswrong.com/posts/SnfjK9ALrzFJB8x7B/against-most-but-not-all-ai-risk-analogies |
# Notice When People Are Directionally Correct
I started watching [Peter Zeihan](https://www.youtube.com/@ZeihanonGeopolitics) videos last year.
He shares a lot of interesting information, although he seems to have a very strong bias towards doom and gloom.
One thing in particular stood out to me as completely absur... | https://www.lesswrong.com/posts/TgKGpTvXQmPem9kcF/notice-when-people-are-directionally-correct |
# The Leeroy Jenkins principle: How faulty AI could guarantee "warning shots"
*Confidence level: This is all speculation about tech that doesn’t exist yet. Everything is highly uncertain. All “simulations” within are illustrative models and should not be taken as serious predictions.*
When talking about AI, existenti... | https://www.lesswrong.com/posts/LZQm7osgjh5tBbEfF/the-leeroy-jenkins-principle-how-faulty-ai-could-guarantee |
# List of projects that seem impactful for AI Governance
(*Cross-posted from* [*EA Forum*](https://forum.effectivealtruism.org/posts/MWLyQxbAuP2Je3iCK/list-of-projects-that-seem-impactful-for-ai-governance)*.)*
Goals of this post
==================
The main goal of this post is to clarify what projects can be done a... | https://www.lesswrong.com/posts/MPMsnpHkHoRQxarnG/list-of-projects-that-seem-impactful-for-ai-governance |
# Gender Exploration
The rationalist community has been discussing whether ‘AGP males’ should try hormones or not. Eneaz Brodsky says [Transitioning Is Harmful To Most AGP Males](https://deathisbad.substack.com/p/transitioning-is-harmful-to-most). Ozy has a [thoughtful, but paywalled, reply](https://thingofthings.subs... | https://www.lesswrong.com/posts/KznQLLpDprpwqcAKD/gender-exploration |
# D&D.Sci Hypersphere Analysis Part 2: Nonlinear Effects & Interactions
Following on from [Part 1](https://www.lesswrong.com/posts/rAnS5jCQ5r87eMvus/d-and-d-sci-hypersphere-analysis-part-1-datafields-and), again don't read this if you're trying to avoid spoilers!
Nonlinear Effects
=================
The correlation a... | https://www.lesswrong.com/posts/F5nkk9dtypicyJY5M/d-and-d-sci-hypersphere-analysis-part-2-nonlinear-effects |
# Detachment vs attachment [AI risk and mental health]
> "What? Why would I choose cosmic energy over Katara?"
>
> *Avatar the Last Airbender*
"You idiot" said the monk from atop his mountain, "there's all this fresh air up here and the view is breathtaking and you're down there dredging *mud* ." The mud-dredger onl... | https://www.lesswrong.com/posts/6xrJxWbJZPSzTZrby/detachment-vs-attachment-ai-risk-and-mental-health |
# Even if we lose, we win
*Epistemic status: Updating on* [*this comment*](https://www.lesswrong.com/posts/NARNMfMxLwgWFK2xK/even-if-we-lose-we-win?commentId=j8HCxZkNjcartKvgh) *and taking into account uncertainty about my own values, my credence in this post is around 50%.*
TLDR: Even in worlds where we create an un... | https://www.lesswrong.com/posts/NARNMfMxLwgWFK2xK/even-if-we-lose-we-win |
# AI doing philosophy = AI generating hands?
I've been playing around with Stable Diffusion recently, and an analogy occurred to me between today's AI's notoriously bad generation of hands and future AI's potentially bad reasoning about philosophy.
In case you aren't already familiar, currently available image gene... | https://www.lesswrong.com/posts/G4ARgcnFogpqorQgb/ai-doing-philosophy-ai-generating-hands-1 |
# How to Promote More Productive Dialogue Outside of LessWrong
I’m relatively new to lesswrong.com, but it seems to be a site where there’s a strong tendency for people to actually be respectful to each other in the comments, even when they disagree. How do we promote more of that in the world in general, where people... | https://www.lesswrong.com/posts/7FuF2p8TGFPyrX99q/how-to-promote-more-productive-dialogue-outside-of-lesswrong |
# The case for training frontier AIs on Sumerian-only corpus
> *Let your every day be full of joy, love the child that holds... | https://www.lesswrong.com/posts/PkqGxkm8XRASJ35bF/the-case-for-training-frontier-ais-on-sumerian-only-corpus-1 |
# Three Types of Constraints in the Space of Agents
*\[Epistemic status: a new perspective on an old thing that may or may not turn out to be useful.\]*
**TDLR*****:** *What sorts of forces and/or constraints shape and structure the space of possible agents? What sort of agents are possible? What sort of agents are l... | https://www.lesswrong.com/posts/2Z3wWP8DBYKJmzuTc/three-types-of-constraints-in-the-space-of-agents |
# What does it look like for AI to significantly improve human coordination, before superintelligence?
I'm currently thinking through the tech tree of developments in human coordination that would increase the probability that the development of systems that are superintelligent (in some to all domains), goes well.
... | https://www.lesswrong.com/posts/knkrxdw3KdFEygAhX/what-does-it-look-like-for-ai-to-significantly-improve-human |
# Review of Alignment Plan Critiques- December AI-Plans Critique-a-Thon Results
These are the Critiques from the AI-Plans.com December Critique-a-Thon of 2023.
Prizes will be awarded to contestants based on collective judge feedback.
Critiques were anonymized when reviewed by the judges.
Judges:
\- [@So8... | https://www.lesswrong.com/posts/LvJdqAfXkAXB2EbM2/review-of-alignment-plan-critiques-december-ai-plans |
# Sparse MLP Distillation
This is a research report about my attempt to extract interpretable features from a transformer MLP by distilling it into a larger student MLP, while encouraging sparsity by applying an L1 penalty to the activations, as depicted in Figure 1. I investigate the features learned by the distilled... | https://www.lesswrong.com/posts/MXabwqMwo3rkGqEW8/sparse-mlp-distillation |
# Investigating Bias Representations in LLMs via Activation Steering
*Produced as part of the *[*SPAR program*](https://sparai.notion.site/Supervised-Program-for-Alignment-Research-SPAR-4da6be132e974823961abfdd0c218536) *(fall 2023) under the mentorship of Nina Rimsky.*
**Introduction**
Given recent advances in the ... | https://www.lesswrong.com/posts/cLfsabkCPtieJ5LoK/investigating-bias-representations-in-llms-via-activation |
# Introducing REBUS: A Robust Evaluation Benchmark of Understanding Symbols
*This is a summary of* [*https://arxiv.org/abs/2401.05604*](https://arxiv.org/abs/2401.05604)*.*
When Google announced Gemini ... | https://www.lesswrong.com/posts/ccouWaRKyxZSexm7X/introducing-rebus-a-robust-evaluation-benchmark-of |
# Goals selected from learned knowledge: an alternative to RL alignment
Summary:
Alignment work on network-based AGI focuses on reinforcement learning. There is an alternative approach that avoids some, but not all, of the difficulties of RL alignment. Instead of trying to *build* an adequate representation of t... | https://www.lesswrong.com/posts/DfJCTp4MxmTFnYvgF/goals-selected-from-learned-knowledge-an-alternative-to-rl |
# Sparse Autoencoders Work on Attention Layer Outputs
*This post is the result of a 2 week research sprint project during the training phase of Neel Nanda’s MATS stream. *
Executive Summary
-----------------
* We replicate [Anthropic's MLP Sparse Autoencoder (SAE) paper](https://transformer-circuits.pub/2023/monos... | https://www.lesswrong.com/posts/DtdzGwFh9dCfsekZZ/sparse-autoencoders-work-on-attention-layer-outputs |
# The impossible problem of due process
*I wrote this entire post in February of 2023, during the fallout from the TIME article. I didn't post it at the time for multiple reasons:*
* *because I had no desire to get involved in all that nonsense*
* *because I was horribly burned out from my own community conflict ... | https://www.lesswrong.com/posts/sJEcNgqnSL2n35QWR/the-impossible-problem-of-due-process |
# Dealing with Awkwardness
This post introduces my model of awkwardness and the methods I’ve used to overcome it.[^75f56qrz2l]
Awkwardness used to be a big struggle in my life - limiting my agency through excessive risk aversion. I can vividly recall a conversation where I was visibly shaking and stammering throughou... | https://www.lesswrong.com/posts/APbb2SSZzySK4pbhD/dealing-with-awkwardness |
# Why are people unkeen to immortality that would come from technological advancements and/or AI?
Nearly everyone with whom I have talked about the mindblowing possibilities of a friendly ASI -people that hadn't heard about it before-, I have seen the same reaction: the person is skeptical and rejects the idea of immo... | https://www.lesswrong.com/posts/rhKXPpS8o4mHuRZTR/why-are-people-unkeen-to-immortality-that-would-come-from |
# Why wasn't preservation with the goal of potential future revival started earlier in history?
*Cross-posted from my blog,* [*Neurobiology Notes*](https://neurobiology.substack.com/p/why-wasnt-preservation-with-the-goal)*. *
John Hunter (1728–1793) did not have an especially promising start to his academic life. He ... | https://www.lesswrong.com/posts/mCu8hnycFkdiBMvD7/why-wasn-t-preservation-with-the-goal-of-potential-future |
# Managing catastrophic misuse without robust AIs
Many people worry about catastrophic misuse of future AIs with highly dangerous capabilities. For instance, powerful AIs might substantially lower the bar to building bioweapons or allow for massively scaling up cybercrime.
How could an AI lab serving AIs to customers... | https://www.lesswrong.com/posts/KENtuXySHJgxsH2Qk/managing-catastrophic-misuse-without-robust-ais |
# How polysemantic can one neuron be? Investigating features in TinyStories.
Summary
=======
I use a pre-trained Sparse Autoencoder (SAE) to examine some features in a 33M-parameter, 2-layer TinyStories language model. I look at how frequently SAE features occur in the dataset and examine how the features are distrib... | https://www.lesswrong.com/posts/LBaTQGD5KyXcbTtKt/how-polysemantic-can-one-neuron-be-investigating-features-in |
# Being nicer than Clippy
*(Cross-posted from [my website](https://joecarlsmith.com/2024/01/16/being-nicer-than-clippy). Podcast version [here](https://www.buzzsprout.com/2034731/14319858), or search "Joe Carlsmith Audio" on your podcast app.*
*This essay is part of a series I'm calling "Otherness and control in
the ... | https://www.lesswrong.com/posts/rdTgtHn3neGzkyCrL/being-nicer-than-clippy |
# Medical Roundup #1
Saving up medical and health related stories from several months allowed for much better organizing of them, so I am happy I split these off. I will still post anything more urgent on a faster basis. There’s lots of things here that are fascinating and potentially very important, but I’ve had to p... | https://www.lesswrong.com/posts/kFDk4Q9QhqrDE68qp/medical-roundup-1 |
# D&D.Sci Hypersphere Analysis Part 3: Beat it with Linear Algebra
As before, this is composed entirely of spoilers for abstractapplic's [D&D.Sci scenario](https://www.lesswrong.com/posts/T3iG4MQ76988JBfkq/d-and-d-sci-fi-colonizing-the-superhypersphere), don't read if you want to solve on your own!
[Last time](https:... | https://www.lesswrong.com/posts/gs9yhwB3LW2CM8rvS/d-and-d-sci-hypersphere-analysis-part-3-beat-it-with-linear |
# Coalescer Models
So we're here today to discuss this post on an improvement to diffuser networks: [https://www.lesswrong.com/posts/6Cdusp5xzrHEoHz9n/faster-latent-diffusion](https://www.lesswrong.com/posts/6Cdusp5xzrHEoHz9n/faster-latent-diffusion)
Just as some general background for readers, there are various kind... | https://www.lesswrong.com/posts/XtDALNuFFr3Zv7Zai/coalescer-models |
# Vote in the LessWrong review! (LW 2022 Review voting phase)
*Click* [*here*](http://lesswrong.com/reviewVoting) *to begin voting. Click* [*here*](https://www.lesswrong.com/posts/B6CxEApaatATzown6/the-lesswrong-2022-review) *for a general overview of the 2022 review.*
The final phase of the 2022 review is upon us! W... | https://www.lesswrong.com/posts/EXcnsGuQJC9khBEkL/vote-in-the-lesswrong-review-lw-2022-review-voting-phase |
# An Introduction To The Mandelbrot Set That Doesn't Mention Complex Numbers
Note: This post assumes you've heard of the [Mandelbrot set](https://en.wikipedia.org/wiki/Mandelbrot_set) before, and you want to know more about it, but that you find [imaginary](https://en.wikipedia.org/wiki/Imaginary_number) and [complex]... | https://www.lesswrong.com/posts/CMiDbAyroRG7kLJNW/an-introduction-to-the-mandelbrot-set-that-doesn-t-mention |
# On Anthropic’s Sleeper Agents Paper
[The recent paper from Anthropic](https://arxiv.org/pdf/2401.05566.pdf) is getting unusually high praise, much of it I think deserved.
The title is: **Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training.**
[Scott Alexander also covers this](https://www.a... | https://www.lesswrong.com/posts/Sf5CBSo44kmgFdyGM/on-anthropic-s-sleeper-agents-paper |
# AlphaGeometry: An Olympiad-level AI system for geometry
\[Published today by DeepMind\]
> **Our AI system surpasses the state-of-the-art approach for geometry problems, advancing AI reasoning in mathematics**
>
> Reflecting the Olympic spirit of ancient Greece, [the International Mathematical Olympiad](https://www... | https://www.lesswrong.com/posts/GGLpjugLQv6TupQgS/alphageometry-an-olympiad-level-ai-system-for-geometry |
# What do people colloquially mean by deep breathing? Slow, large, or diaphragmatic?
Some version of the idea of "relax, take a deep breath" is common
across cultures. But what does the "deep" in "deep breath" mean to
people? I've come across three plausible interpretations, and I'm
curious what people colloquially m... | https://www.lesswrong.com/posts/nwyYLfLZjYESuqRaZ/what-do-people-colloquially-mean-by-deep-breathing-slow |
# Four visions of Transformative AI success
Tl;dr
=====
When people work towards making a good future in regards to Transformative AI (TAI), what’s the vision of the future that they have in mind and are working towards?
I’ll propose four (caricatured) answers that different people seem to give:
* (Vision 1) “Hel... | https://www.lesswrong.com/posts/3aicJ8w4N9YDKBJbi/four-visions-of-transformative-ai-success |
# In Strategic Time, Open-Source Games Are Loopy
For many [situations we'd like to analyze](https://www.lesswrong.com/posts/puutBJLWbg2sXpFbu/newcomblike-problems-are-the-norm), information flow is more important than chronological order. When game theorists say that two players choose their actions [simultaneously](h... | https://www.lesswrong.com/posts/DzfBgxoAC3kPir4PX/in-strategic-time-open-source-games-are-loopy |
# Some heuristics I use for deciding how much I trust scientific results
I've done nothing to test these heuristics and have no empirical evidence for how well they work for forecasting replications or anything else. I’m going to write them anyway. The heuristics I’m listing are roughly in order of how important I thi... | https://www.lesswrong.com/posts/8DLeEvvDvCpB4HHqa/some-heuristics-i-use-for-deciding-how-much-i-trust |
# D&D.Sci Hypersphere Analysis Part 4: Fine-tuning and Wrapup
As before, this is extreme spoilers for [this D&D.Sci scenario](https://www.lesswrong.com/posts/T3iG4MQ76988JBfkq/d-and-d-sci-fi-colonizing-the-superhypersphere), if you want to solve on your own try that first!
[ Last time](https://www.lesswrong.com/posts... | https://www.lesswrong.com/posts/tFYec8tBHQbLZ72ES/d-and-d-sci-hypersphere-analysis-part-4-fine-tuning-and |
# Does literacy remove your ability to be a bard as good as Homer?
*Epistemic status: probably we did lose the ability to memorize long songs due to improper practice, but it may be possible to enjoy the benefits of literacy and epic memory simultaneously.*
*Thanks to Niels uit de Bos for better links and editing, an... | https://www.lesswrong.com/posts/DqtQtc7sB6Hz5mcSt/does-literacy-remove-your-ability-to-be-a-bard-as-good-as |
# What evidence is there for (or against) theories about the extent to which effective altruist interests motivated the ouster of Sam Altman last year?
There was of course initially much speculation that effective altruism was solely responsible. That was a common accusation from some among the effective accelerationi... | https://www.lesswrong.com/posts/JzhqkoFqnjJksRb8C/what-evidence-is-there-for-or-against-theories-about-the |
# AI #48: Exponentials in Geometry
The biggest event of the week was the Sleeper Agents paper from Anthropic. I expect that to inform our thoughts for a while to come, and to lay foundation for additional work. We also had the first third of the IMO solved at almost gold metal level by DeepMind, discovering that math ... | https://www.lesswrong.com/posts/WRGmBE3h4WjA5EC5a/ai-48-exponentials-in-geometry |
# Good job opportunities for helping with the most important century
_Yes, this is my first post in almost a year. I’m no longer prioritizing this blog, but I will still occasionally post something._
I [wrote](https://www.cold-takes.com/call-to-vigilance/) ~2 years ago that it was hard to point to concrete opportunit... | https://www.lesswrong.com/posts/7NsGdAcCceiFhMKKh/good-job-opportunities-for-helping-with-the-most-important |
# More Usable Recipes
Instead of a recipe book I keep a [web page](https://www.jefftk.com/recipes/). Whenever I make something and think I'll make it again, I add it. It's public since I generally make things public by default, though it's terse since it's intended as a reference for someone (me!) who already knows wh... | https://www.lesswrong.com/posts/SCpwLpD4dWpsEfGkT/more-usable-recipes |
# On the abolition of man
*(Cross-posted from [my website](https://joecarlsmith.com/2024/01/18/on-the-abolition-of-man). Podcast version [here](https://joecarlsmithaudio.buzzsprout.com/2034731/14333032-on-the-abolition-of-man), or search for "Joe Carlsmith Audio" on your podcast app.*
*This essay is part of a series ... | https://www.lesswrong.com/posts/oJRMDAvWk8diDjNsG/on-the-abolition-of-man |
# The True Story of How GPT-2 Became Maximally Lewd
*This video recounts an incident that occurred at OpenAI in which flipping a single minus sign led the RLHF process to make GPT-2 only output sexually explicit continuations. *
*The incident is described in OpenAI's paper "*[*Fine-Tuning Language Models from Human P... | https://www.lesswrong.com/posts/xD3wymX24BpqezBpw/the-true-story-of-how-gpt-2-became-maximally-lewd |
# Toward A Mathematical Framework for Computation in Superposition
*Author order randomized. Authors contributed roughly equally — see attribution section for details.*
*Update as of July 2024: we have collaborated with* [*@LawrenceC*](https://www.alignmentforum.org/users/lawrencec?mention=user) *to expand section 1 ... | https://www.lesswrong.com/posts/2roZtSr5TGmLjXMnT/toward-a-mathematical-framework-for-computation-in |
# Against Nonlinear (Thing Of Things)
[Against Nonlinear](https://thingofthings.substack.com/p/against-nonlinear) is a followup that Ozy wrote to [Practically A Book Review: Appendix to "Nonlinear's Evidence: Debunking False and Misleading Claims"](https://thingofthings.substack.com/p/practically-a-book-review-appendi... | https://www.lesswrong.com/posts/oZ7Nqtw8dtjC2ZTCZ/against-nonlinear-thing-of-things |
# The Underreaction to OpenAI
TL; DR
------
Seeing OpenAI's impressive technology should make media commentators think it more likely that AGI is possible and existentially dangerous. Some do, but overall this shift in views does not seem consistent enough; the default reaction still seems to be to view the whole AI ... | https://www.lesswrong.com/posts/3ii8GeiYN5hn2QQgH/the-underreaction-to-openai |
# Manifund: 2023 in Review
Manifund is a new funding org that experiments with systems and software to support awesome projects. In 2023, we built a website ([manifund.org](http://manifund.org/)) and donor ecosystem supporting three main programs: impact certificates, regranting, and an open call for applications. We ... | https://www.lesswrong.com/posts/cxjRjs5BWw5pBHZiA/manifund-2023-in-review |
# Are there high-quality surveys available detailing the rates of polyamory among Americans age 18-45 in metropolitan areas in the United States?
In recent years, there has been more public attention paid to seemingly increasing rates of less common identity characteristics or alternative lifestyle choices. This has p... | https://www.lesswrong.com/posts/isvZjeMQoh84hcFaa/are-there-high-quality-surveys-available-detailing-the-rates |
# Logical Line-Of-Sight Makes Games Sequential or Loopy
In the [last post](https://www.lesswrong.com/posts/DzfBgxoAC3kPir4PX/in-strategic-time-open-source-games-are-loopy), we talked about strategic time and the strategic time loops studied in [open-source game theory](https://www.lesswrong.com/tag/open-source-game-th... | https://www.lesswrong.com/posts/6nWhntKPtCFm8LFo6/logical-line-of-sight-makes-games-sequential-or-loopy |
# What rationality failure modes are there?
How do people fail to improve their rationality? How do they accidentally harm themselves in the process? I'm thinking of writing a post "How not to improve your rationality" or "A nuanced guide to reading the sequences" that preempts common mistakes, and I'd appreciate hear... | https://www.lesswrong.com/posts/4Jqq9obEGCjmvQgjd/what-rationality-failure-modes-are-there |
# I Want XMP But I Know Why I Can't Have It
When writing text that will be displayed to users as HTML I write HTML. This seems like it would be the normal way to do it, though it's unusual these days. Either you draft in a [fancy content editor](https://en.wikipedia.org/wiki/WYSIWYG) or you write [Markdown](https://en... | https://www.lesswrong.com/posts/cZGRn5yzPMqDyTcNJ/i-want-xmp-but-i-know-why-i-can-t-have-it |
# Estimating efficiency improvements in LLM pre-training
TL/DR:
------
GPT-3 was trained on 10,000 NVIDIA V100’s for 14.8 days. Based on my - very uncertain - estimates, OpenAI should now be able train a comparably capable model on an equivalent cluster in under an hour. This implies a drop in hardware requirements o... | https://www.lesswrong.com/posts/tJAD2LG9uweeEfjwq/estimating-efficiency-improvements-in-llm-pre-training |
# There is way too much serendipity
*Crossposted from* [*substack*](https://malmesbury.substack.com/p/there-is-way-too-much-serendipity)*.*
As we all know, sugar is sweet and so are the [$30B in yearly revenue](https://www.statista.com/statistics/1201648/revenue-artificial-sweetener-market-worldwide/) from the artifi... | https://www.lesswrong.com/posts/oA23zoEjPnzqfHiCt/there-is-way-too-much-serendipity |
# On "Geeks, MOPs, and Sociopaths"
Hey, alkjash! I'm excited to talk about some of David Chapman's work with you. Full disclosure, I'm a big fan of Chapman's in general and also a creator within the meta/post-rationality scene with him (to use some jargon to be introduced very shortly).
You mentioned being superficia... | https://www.lesswrong.com/posts/uiEoxLh8kaDzkayfL/on-geeks-mops-and-sociopaths |
# A quick investigation of AI pro-AI bias
**Question**: Do LLMs exhibit a pro-AI bias in their answers?
**Experiment**: I compare the scores LLMs give to resumes when the titles of the publications they contain reflect a pro or anti-AI sentiment, or when a pro or anti-AI sentiment is explicitly expressed.
**Result**... | https://www.lesswrong.com/posts/WcRwTjRxY3fZTKmaw/a-quick-investigation-of-ai-pro-ai-bias |
# Decent plan prize winner & highlights
Last week I [announced](https://www.lesswrong.com/posts/GjeprNFWyNPW8oseD) a contest to write a 1-paragraph plan for training & deploying nextgen general AI models which prevents them from killing you. Many of the answers were excellent. If you feel there's a big bag of AI safet... | https://www.lesswrong.com/posts/fEuZdHbCofHKgjmCQ/decent-plan-prize-winner-and-highlights |
# Legibility Makes Logical Line-Of-Sight Transitive
In the [last post](https://www.lesswrong.com/posts/6nWhntKPtCFm8LFo6/logical-line-of-sight-makes-games-sequential-or-loopy), we talked about how [logical line-of-sight](https://www.lesswrong.com/tag/acausal-trade) makes games sequential or loopy, in a way that causal... | https://www.lesswrong.com/posts/M7jg8FiXHt5YpcW4D/legibility-makes-logical-line-of-sight-transitive |
# legged robot scaling laws
Fiction [has lots of giant walking robots](https://tvtropes.org/pmwiki/pmwiki.php/Main/HumongousMecha). Those designs are generally considered impractical or impossible, but they've been discussed for [thousands of years](https://en.wikipedia.org/wiki/Talos), there must be something appeali... | https://www.lesswrong.com/posts/vZ7jTb2bxJDAKhno2/legged-robot-scaling-laws |
# Against the Burden of Knowledge
[Several posts back](https://maximumprogress.substack.com/p/something-is-getting-harder-but-its) I wrote about the 2020 paper “[Are Ideas Getting Harder To Find?](https://www.nber.org/system/files/working_papers/w23782/w23782.pdf)” by Nicholas Bloom, Charles Jones, John Van Reenen, an... | https://www.lesswrong.com/posts/5gN7RQSvWtEvDzyhv/against-the-burden-of-knowledge |
# Why Improving Dialogue Feels So Hard
Earlier this week, @sweenesm published [a post with techniques for making dialogue more productive](https://www.lesswrong.com/posts/7FuF2p8TGFPyrX99q/how-to-promote-more-productive-dialogue-outside-of-lesswrong#Other_Thoughts_). He opened it with the question, "_How do we promote... | https://www.lesswrong.com/posts/nTTKFHoHHgTHjyuAW/why-improving-dialogue-feels-so-hard |
# [linkpost] Self-Rewarding Language Models
> Abstract: We posit that to achieve superhuman agents, future models require super- human feedback in order to provide an adequate training signal. Current approaches commonly train reward models from human preferences, which may then be bottlenecked by human performance le... | https://www.lesswrong.com/posts/PqQdZEcynkBWfJt7K/linkpost-self-rewarding-language-models |
# Why have insurance markets succeeded where prediction markets have not?
Insurance is big business and is load-bearing for many industries. It has gained popular acceptance. Prediction markets have not. This is despite clear similarities between the two domains.
One can list similarities:
* *You are trading finan... | https://www.lesswrong.com/posts/kJEruRccFEvspaWNH/why-have-insurance-markets-succeeded-where-prediction |
# You can rack up massive amounts of data quickly by asking questions to all your friends
**tl;dr** I asked 15 friends and acquaintances a question out of the blue because I was curious about something. I collected a massive amount of evidence, and it was fun, and I’m more agentic now than I was yesterday. Do it too... | https://www.lesswrong.com/posts/owtncQS5Z2LdjAEaN/you-can-rack-up-massive-amounts-of-data-quickly-by-asking |
# Book review: Cuisine and Empire
People began cooking our food maybe two million years ago and have not stopped since. Cooking is almost a cultural universal. Bits of raw fruit or leaves or flesh are a rare occasional treat or garnish – we prefer our meals transformed. There are other millennia-old procedures we do t... | https://www.lesswrong.com/posts/CcmQr5RvP2pvnxTG5/book-review-cuisine-and-empire |
# Another Non-Anthropic Paradox: The Unsurprising Rareness of Rare Events
*This is the fifth post in my series on Anthropics. The previous one is* [*Anthropical Paradoxes are Paradoxes of Probability Theory*](https://www.lesswrong.com/posts/8rWP3ZvHe2HkQd53o/anthropical-paradoxes-are-paradoxes-of-probability-theory)*.... | https://www.lesswrong.com/posts/bHzcKEEHL854KsWGM/another-non-anthropic-paradox-the-unsurprising-rareness-of |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.