text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Let's Design A School, Part 2.4
School as Education - The Curriculum (Phase 3, Specific)
In previous posts, we laid out our model of school-as-education:
Phase 1 was literacy and numeracy,
Phase 2 was core civilizational requirements and survey courses,
Phase 3 was core adulting requirements and self-study,
and ... | https://www.lesswrong.com/posts/dSyR89vS9sema9vvB/let-s-design-a-school-part-2-4-school-as-education-the |
# SAEs Discover Meaningful Features in the IOI Task
TLDR: recently, we wrote a [paper](https://arxiv.org/abs/2405.08366) proposing several evaluations of SAEs against "ground-truth" features computed w/ supervision for a given task (in our case, IOI \[1\]). However, we didn't optimize the SAEs much for performance in ... | https://www.lesswrong.com/posts/zj3GKWAnhPgTARByB/saes-discover-meaningful-features-in-the-ioi-task |
# Calculating Natural Latents via Resampling
So you’ve read some of our [previous](https://www.lesswrong.com/posts/dWQWzGCSFj6GTZHz7/natural-latents-the-math) [natural latents](https://www.lesswrong.com/posts/mMEbfooQzMwJERAJJ/natural-latents-the-concepts) [posts](https://www.lesswrong.com/posts/RTiuLzusJWyepFpbN/why-... | https://www.lesswrong.com/posts/NHKCtSXgFieDAyWt2/calculating-natural-latents-via-resampling |
# Weeping Agents
What is Agency? In a relatively intuitive way it’s a property of systems which have three things: Some –however rudimentary– way of modelling a slice of reality, a utility function over that slice of reality and a way of acting upon it. An agent can make better-than-chance guesses about how the behavi... | https://www.lesswrong.com/posts/HmaZAgKWcGDopJWkQ/weeping-agents |
# SB 1047 Is Weakened
It looks like Scott Weiner’s SB 1047 [is now severely weakened.](https://x.com/mrgunn/status/1796283488519311707)
[![A lawmaker sitting at a large wooden desk, focused on editing a bill. The desk is cluttered with papers, a laptop, and a cup of coffee. The lawmaker, dressed in formal attire with... | https://www.lesswrong.com/posts/4t98oqh8tzDvoatHs/sb-1047-is-weakened |
# An evaluation of Helen Toner’s interview on the TED AI Show
Helen Toner was recently interviewed on the TED AI Show. In the first segment, she explains why the OpenAI board decided to fire Sam Altman ([video](https://www.ted.com/talks/the_ted_ai_show_what_really_went_down_at_openai_and_the_future_of_regulation_w_hel... | https://www.lesswrong.com/posts/fJseTLvaun4SmRv4a/an-evaluation-of-helen-toner-s-interview-on-the-ted-ai-show |
# AI #67: Brief Strange Trip
I had a great time at LessOnline. It was a both a working trip and also a trip to an alternate universe, a road not taken, a vision of a different life where you get up and start the day in dialogue with Agnes Callard and Aristotle and in a strange combination of relaxed and frantically go... | https://www.lesswrong.com/posts/gKxf6qJaSP5Ehqnsm/ai-67-brief-strange-trip |
# [Link Post] "Foundational Challenges in Assuring Alignment and Safety of Large Language Models"
We’ve recently released a comprehensive research agenda on LLM safety and alignment. This is a collaborative work with contributions from more than 35+authors across the fields of AI Safety, machine learning, and NLP. M... | https://www.lesswrong.com/posts/eoshu82FHeGzGXz2M/link-post-foundational-challenges-in-assuring-alignment-and |
# Situational Awareness Summarized - Part 1
This is the first post in the [Situational Awareness Summarized](https://www.lesswrong.com/s/SybKaNSqyADrnMQ7H) sequence. Collectively, these posts represent my attempt to condense Leopold Aschenbrenner's recent report, [Situational Awareness](https://situational-awareness.a... | https://www.lesswrong.com/posts/sDnbTN8Qb84P6D3Xd/situational-awareness-summarized-part-1 |
# There Are No Primordial Definitions of Man/Woman
[](https://substackcdn.com/image/fetch/f_auto,q_aut... | https://www.lesswrong.com/posts/LGDXYnnaMqrAin76w/there-are-no-primordial-definitions-of-man-woman |
# Humming is not a free $100 bill
[Last month](https://acesounderglass.com/2024/05/12/do-you-believe-in-hundred-dollar-bills-lying-on-the-ground-consider-humming/) I posted about humming as a cheap and convenient way to flood your nose with nitric oxide (NO), a known antiviral. Alas, the [economists were right](https:... | https://www.lesswrong.com/posts/dsZeogoPQbF8jSHMB/humming-is-not-a-free-usd100-bill |
# Scaling and evaluating sparse autoencoders
\[[Blog](https://openai.com/index/extracting-concepts-from-gpt-4/)\] \[[Paper](https://cdn.openai.com/papers/sparse-autoencoders.pdf)\] \[[Visualizer](https://openaipublic.blob.core.windows.net/sparse-autoencoder/sae-viewer/index.html)\]
Abstract:
> Sparse autoencoders pr... | https://www.lesswrong.com/posts/Fg2gAgxN6hHSaTjkf/scaling-and-evaluating-sparse-autoencoders |
# Response to Aschenbrenner's "Situational Awareness"
([*Cross-posted from Twitter.*](https://x.com/robbensinger/status/1798845199382429697))
My take on Leopold Aschenbrenner's new [report](https://situational-awareness.ai/): I think Leopold gets it right on a bunch of important counts.
Three that I especially care ... | https://www.lesswrong.com/posts/Yig9oa4zGE97xM2os/response-to-aschenbrenner-s-situational-awareness |
# Memorizing weak examples can elicit strong behavior out of password-locked models
We’ve recently done [some research looking into sandbagging](https://www.alignmentforum.org/posts/c4sZqhqPwNKGz3fFW/paper-stress-testing-capability-elicitation-with-password): examining when models can succeed at intentionally producin... | https://www.lesswrong.com/posts/bfm5Fm9pJYAthFSZT/memorizing-weak-examples-can-elicit-strong-behavior-out-of |
# A Case for Superhuman Governance, using AI
I believe that:
1. AI-enhanced organization governance could be a potentially huge win in the next few decades.
2. AI-enhanced governance could allow organizations to reach superhuman standards, like having an expected "99.99" reliability rate of not being corrupt or not... | https://www.lesswrong.com/posts/cJv8rBSshrR82NRET/a-case-for-superhuman-governance-using-ai |
# Offering Completion
One verbal approach I often find useful with little kids is to offer them the opportunity to complete a phrase. Compare:
> Parent: It's time to brush teeth!
>
> Toddler: I no want brush teef!
vs:
> Parent: It's time to brush...
>
> Toddler: Teef!
Similarly, "the dirty tissue goes in the..."... | https://www.lesswrong.com/posts/9vwfFGHmck2KnLraJ/offering-completion |
# Is Claude a mystic?
There's an odd tendency for large language models such as Claude to output spiritual meta content if they run long enough. See, for example, some [LLM outputs selected by Repligate](https://generative.ink/prophecies/#1955) (LLM content starts in 2024), and [Worldspider outputs](https://minihf.com... | https://www.lesswrong.com/posts/Caj9brP8PEqkgSrSC/is-claude-a-mystic |
# Quotes from Leopold Aschenbrenner’s Situational Awareness Paper
This post is different.
Usually I offer commentary and analysis. I share what others think, then respond.
This is the second time I am importantly not doing that. [The work](https://situational-awareness.ai/wp-content/uploads/2024/06/situationalawaren... | https://www.lesswrong.com/posts/nP5FFYFjtY8LgWymt/quotes-from-leopold-aschenbrenner-s-situational-awareness |
# Situational Awareness Summarized - Part 2
This is the second post in the [Situational Awareness Summarized](https://www.lesswrong.com/s/SybKaNSqyADrnMQ7H) sequence. Collectively, these posts represent my attempt to condense Leopold Aschenbrenner's recent report, [Situational Awareness](https://situational-awareness.... | https://www.lesswrong.com/posts/TEE799ocKYNriD8CZ/situational-awareness-summarized-part-2 |
# Natural Latents Are Not Robust To Tiny Mixtures
In our [previous](https://www.lesswrong.com/posts/dWQWzGCSFj6GTZHz7/natural-latents-the-math) [natural](https://www.lesswrong.com/posts/mMEbfooQzMwJERAJJ/natural-latents-the-concepts) [latent](https://www.lesswrong.com/posts/RTiuLzusJWyepFpbN/why-care-about-natural-lat... | https://www.lesswrong.com/posts/xDsbqxeCQWe4BiYFX/natural-latents-are-not-robust-to-tiny-mixtures |
# D&D.Sci Alchemy: Archmage Anachronos and the Supply Chain Issues
*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.*
*After talking with abstractapplic, I've stolen the June 7th scen... | https://www.lesswrong.com/posts/sex8PDjnnqZgzCt5c/d-and-d-sci-alchemy-archmage-anachronos-and-the-supply-chain |
# Question about Lewis' counterfactual theory of causation
In reading the [SEP entry on counterfactual theories of causation](https://plato.stanford.edu/entries/causation-counterfactual/), I had the following question occur, and I haven't been able to satisfactorily resolve it for myself.
An event *e* is said to caus... | https://www.lesswrong.com/posts/fNQiMd8nkxir57K8Y/question-about-lewis-counterfactual-theory-of-causation |
# 0. CAST: Corrigibility as Singular Target
What the heck is up with “corrigibility”? For most of my career, I had a sense that it was a grab-bag of properties that seemed nice in theory but hard to get in practice, perhaps due to being incompatible with agency.
Then, last year, I spent some time revisiting my perspe... | https://www.lesswrong.com/posts/NQK8KHSrZRF5erTba/0-cast-corrigibility-as-singular-target-1 |
# 1. The CAST Strategy
(Part 1 of [the CAST sequence](https://www.alignmentforum.org/posts/NQK8KHSrZRF5erTba/0-cast-corrigibility-as-singular-target-1))
AI Risk Introduction
====================
(TLDR for this section, since it’s 101 stuff that many readers will have already grokked: Misuse vs Mistake; Principal-Age... | https://www.lesswrong.com/posts/3HMh7ES4ACpeDKtsW/1-the-cast-strategy |
# Sev, Sevteen, Sevty, Sevth
I don't like the number seven. Well, really the name of the number seven. All the other single digit numbers are single syllable, and seven has to go and take two. Seventy and seventeen have the same problem. What can we do?
I think the two main candidates are "sev" (dropping the second s... | https://www.lesswrong.com/posts/buCy3o4LGXR85gjJb/sev-sevteen-sevty-sevth |
# Why don't we just get rid of all the bioethicists?
I've had a few vague answers before, but as I was considering how no one was allowed to do human challenge trials during COVID, I thought I'd ask:
Can anyone come up with an example where a bioethicist actually improved the world?
I'm looking for specific cases, n... | https://www.lesswrong.com/posts/JpEs4aBSWKR8LjA7R/why-don-t-we-just-get-rid-of-all-the-bioethicists |
# Access to powerful AI might make computer security radically easier
People talk about model weight security being really hard and crucial around the advent of AGI. (E.g. [RAND report](https://www.rand.org/pubs/research_reports/RRA2849-1.html), [Leopold](https://situational-awareness.ai/lock-down-the-labs/); see [her... | https://www.lesswrong.com/posts/2wxufQWK8rXcDGbyL/access-to-powerful-ai-might-make-computer-security-radically |
# Closed-Source Evaluations
Public tripwires are no tripwires.
*I'm writing a quick and dirty post because the alternative is that I wait for months and maybe not write it after all. I am broadly familiar with the state of interpretability research but do not know what the state of model evaluations is at the moment.... | https://www.lesswrong.com/posts/4q9kyeqmA9uGq8fRJ/closed-source-evaluations |
# Status quo bias is usually justified
Generally, we know more about the status quo then about anything else.
* We know that we can live in the current climate, we don’t know that about any other climate.
* We know that society functions with current laws and norms, we don’t know that about any other set of laws.... | https://www.lesswrong.com/posts/Bvz9LxFpkmZWcCbff/status-quo-bias-is-usually-justified |
# Alignment Gaps
*Misaligned agendas and terminology among academic, industrial and independent AI alignment research*
This post aims to fill some gaps between technical AI Alignment topics and academic AI research.
It summarises a quick but informed scouting of academic research papers that are closely connected to... | https://www.lesswrong.com/posts/NgFuzwxQHzxXqY4sf/alignment-gaps |
# I made an AI safety fellowship. What I wish I knew.
### **TLDR**
If you make an AI safety course: Leverage other people. Make sure to cover why AI safety is important. Build a community and just start. Use the below resources and slides.
### **Why Bother**
Making an AI safety fellowship is highly worth doing. It... | https://www.lesswrong.com/posts/88cEgGcSXwFvPms4s/i-made-an-ai-safety-fellowship-what-i-wish-i-knew |
# Two easy things that maybe Just Work to improve AI discourse
So, it seems AI discourse on X / Twitter is getting polarised. This is bad. Especially bad is how some engage in deliberate [weaponization of discourse](https://www.lesswrong.com/tag/simulacrum-levels), for political ends.
At the same time, I observe: AI... | https://www.lesswrong.com/posts/CZjnvaFiRokwst68C/two-easy-things-that-maybe-just-work-to-improve-ai-discourse |
# 2. Corrigibility Intuition
(Part 2 of [the CAST sequence](https://www.alignmentforum.org/posts/NQK8KHSrZRF5erTba/0-cast-corrigibility-as-singular-target-1))
As a reminder, here’s how I’ve been defining “corrigible” when introducing the concept: **an agent is corrigible when it robustly acts opposite of the trope of... | https://www.lesswrong.com/posts/QzC7kdMQ5bbLoFddz/2-corrigibility-intuition |
# Searching for the Root of the Tree of Evil
*“There are a thousand hacking at the branches of evil to one who is striking at the root”*
*Henry David Thoreau,* [*Walden*](https://www.gutenberg.org/files/205/205-h/205-h.htm#:~:text=There%20are%20a%20thousand%20hacking%20at%20the%20branches%20of%20evil%20to%20one%20who... | https://www.lesswrong.com/posts/SnLXoyd2bzXWnSy4y/searching-for-the-root-of-the-tree-of-evil |
# What if a tech company forced you to move to NYC?
It’s interesting to me how chill people sometimes are about the non-extinction future AI scenarios. Like, there seem to be opinions around along the lines of “pshaw, it might ruin your little sources of ‘meaning’, Luddite, but we have always had change and as long as... | https://www.lesswrong.com/posts/9gXsecDTh2WrpqN8j/what-if-a-tech-company-forced-you-to-move-to-nyc |
# Dumbing down
In past few years I've been blogging in Slovak, that is, downscaling from writing in English, a language with 1457 million speakers to a language with 7 million speakers.
From the point of view of the writer, this has been a very different experience. It's not only that for a topic that interests one m... | https://www.lesswrong.com/posts/jqsRBR2fgoMPc9dGS/dumbing-down |
# Demystifying "Alignment" through a Comic
*I am keeping this comic online as a record of an earlier perspective. My current work in fixed point logic and proof theory has led me to reject several foundational assumptions in mainstream alignment and interpretability research. This does not imply that AI poses no risks... | https://www.lesswrong.com/posts/Av9D4GkdGNkiS2wHx/demystifying-alignment-through-a-comic |
# Exploring Llama-3-8B MLP Neurons
**TL;DR:** We created a dataset of text snippets that strongly activate neurons in **Llama-3-8B model**. This dataset shows meaningful features that can be found. Explore the neurons with the web interface: [https://neuralblog.github.io/llama3-neurons/neuron_viewer.html](https://neur... | https://www.lesswrong.com/posts/TA9eEgiWJfgBcJ7wn/exploring-llama-3-8b-mlp-neurons |
# Introducing SARA: a new activation steering technique
**Disclaimer**
------------
I currently am a Postdoctoral Fellow in Computational Neuroscience, learning about Mechanistic Interpretability and AI Safety in general. This post and the paper that goes with it are part of my current pivot towards these topics; thus... | https://www.lesswrong.com/posts/KjBvGS6dgMz5qeDpL/introducing-sara-a-new-activation-steering-technique |
# 3a. Towards Formal Corrigibility
(Part 3a of [the CAST sequence](https://www.alignmentforum.org/posts/NQK8KHSrZRF5erTba/0-cast-corrigibility-as-singular-target-1))
As mentioned in [Corrigibility Intuition](https://www.alignmentforum.org/posts/QzC7kdMQ5bbLoFddz/2-corrigibility-intuition), I believe that it’s more im... | https://www.lesswrong.com/posts/WDHREAnbfuwT88rqe/3a-towards-formal-corrigibility |
# 3b. Formal (Faux) Corrigibility
(Part 3b of [the CAST sequence](https://www.alignmentforum.org/posts/NQK8KHSrZRF5erTba/0-cast-corrigibility-as-singular-target-1))
**EDIT: WARNING: This formalism is** ***critically flawed!*** It should mainly be taken as a way to get a handle on where my mind was at when I was writ... | https://www.lesswrong.com/posts/t8nXfPLBCxsqhbipp/3b-formal-faux-corrigibility |
# Two Family Dance Flyers
I'm going to be calling [another](https://www.jefftk.com/p/calling-my-first-family-dance) family dance [in a week](https://www.facebook.com/events/443304658639233), and Lily and Anna wanted to make flyers to advertise it. I wrote out a sheet with the key details they might want to include:
[... | https://www.lesswrong.com/posts/JM7XHz3nNPgydWqyf/two-family-dance-flyers |
# The Data Wall is Important
Modern AI is trained on a [huge fraction of the internet](https://www.dwarkeshpatel.com/p/will-scaling-work), especially at the cutting edge, with the best models trained on close to all the high quality data we’ve got.[^rf8upr4bqjo] And data is [really important](https://nonint.com/2023/0... | https://www.lesswrong.com/posts/axjb7tN9X2Mx4HzPz/the-data-wall-is-important |
# Soviet comedy film recommendations
I’m a big fan of the Soviet comedy directors [Eldar Ryazanov](https://en.wikipedia.org/wiki/Eldar_Ryazanov), [Leonid Gaidai](https://en.wikipedia.org/wiki/Leonid_Gaidai), and [Georgiy Daneliya](https://en.wikipedia.org/wiki/Georgiy_Daneliya). Almost anything by them is worth watchi... | https://www.lesswrong.com/posts/ZgfM4QLtQbswf7W7k/soviet-comedy-film-recommendations |
# Why I don't believe in the placebo effect
Have you heard this before? In clinical trials, medicines have to be compared to a placebo to separate the effect of the medicine from the psychological effect of taking the drug. The patient's belief in the power of the medicine has a strong effect on its own. In fact, for ... | https://www.lesswrong.com/posts/kpd83h5XHgWCxnv3h/why-i-don-t-believe-in-the-placebo-effect |
# On Dwarksh’s Podcast with Leopold Aschenbrenner
Previously: [Quotes from Leopold Aschenbrenner’s Situational Awareness Paper](https://thezvi.substack.com/p/quotes-from-leopold-aschenbrenners)
[Dwarkesh Patel talked to Leopold Aschenbrenner for about four and a half hours](https://www.youtube.com/watch?v=zdbVtZIn9IM... | https://www.lesswrong.com/posts/DiMz82FwsHPugqxFD/on-dwarksh-s-podcast-with-leopold-aschenbrenner |
# 4. Existing Writing on Corrigibility
(Part 4 of [the CAST sequence](https://www.alignmentforum.org/posts/NQK8KHSrZRF5erTba/0-cast-corrigibility-as-singular-target-1))
This document is an in-depth review of the primary documents discussing corrigibility that I’m aware of. In particular, I'll be focusing on the writi... | https://www.lesswrong.com/posts/d7jSrBaLzFLvKgy32/4-existing-writing-on-corrigibility |
# 5. Open Corrigibility Questions
(Part 5 of [the CAST sequence](https://www.alignmentforum.org/s/KfCjeconYRdFbMxsy/p/NQK8KHSrZRF5erTba))
Much work remains on the topic of corrigibility and the CAST strategy in particular. There’s theoretical work in both nailing down an even more complete picture of corrigibility an... | https://www.lesswrong.com/posts/wZjGLYp5WQwF8Y8Kk/5-open-corrigibility-questions |
# [Valence series] 4. Valence & Liking / Admiring
4.1 Post summary / Table of contents
====================================
*Part of the* [*Valence series*](https://www.lesswrong.com/s/6uDBPacS6zDipqbZ9)*.*
*(This is my second attempt to write the 4th post of my valence series. If you already read* [*the previous at... | https://www.lesswrong.com/posts/LaeP39jJpfPyoiSZm/valence-series-4-valence-and-liking-admiring |
# Good ways to monetarily profit from the increasing demand for power?
There have been several good posts about how to profit from a world of AI takeoff.
However, when it comes to individual investment recommendations, most of them are for direct AI companies or builders of GPUs and compute infrastrucutre
Recently, ... | https://www.lesswrong.com/posts/jrSE9z2dkb4L3MFvQ/good-ways-to-monetarily-profit-from-the-increasing-demand |
# My AI Model Delta Compared To Yudkowsky
### Preamble: Delta vs Crux
I don’t natively think in terms of [cruxes](https://www.lesswrong.com/tag/double-crux). But there’s a similar concept which is more natural for me, which I’ll call a delta.
Imagine that you and I each model the world (or some part of it) as implem... | https://www.lesswrong.com/posts/q8uNoJBgcpAe3bSBp/my-ai-model-delta-compared-to-yudkowsky |
# DPO/PPO-RLHF on LLMs incentivizes sycophancy, exaggeration and deceptive hallucination, but not misaligned powerseeking
*TL;DR: GPTs are imitation learners, even with current forms of RL;HF.*
[Direct preference optimization](https://arxiv.org/pdf/2305.18290) is a conditioning method for generative probabilistic mod... | https://www.lesswrong.com/posts/KqYQYkqsHqRuAKki5/dpo-ppo-rlhf-on-llms-incentivizes-sycophancy-exaggeration |
# "Metastrategic Brainstorming", a core building-block skill
I want to develop rationality training, which is aimed at solving confusing problems.
Two key problems with "confusing problems" are:
1. You might feel so confused and overwhelmed that you bounce off completely.
2. You might be confused about what count... | https://www.lesswrong.com/posts/cbWoMepny3Jo9XqEr/metastrategic-brainstorming-a-core-building-block-skill |
# my favourite Scott Sumner blog posts
Given Scott's invitation to LessOnline and general embrace of this community, I thought many here would appreciate this curated list of my favourite Scott Sumner blog posts.
-
Scott Sumner is best known as an economist who was praised for positively influencing economic policy ... | https://www.lesswrong.com/posts/ZZyqzqWi3FAJkujXa/my-favourite-scott-sumner-blog-posts |
# How to eliminate cut?
The purpose of this post isn't to convince you that cut elimination is important. See, for example, the [nLab article](https://ncatlab.org/nlab/show/cut+rule). Rather, the purpose of this post is to (semi-formally) prove cut elimination in a way that I at least find easy to understand. I have c... | https://www.lesswrong.com/posts/oaARaRB2AebrjmfGi/how-to-eliminate-cut |
# Let's Design A School, Part 3.1:
Bringing it all together with the Sieve Model
In part 1, we laid out the social services model of a school.
In part 2, we described a new educational model of a school.
In part 3, we’re going to combine them.
Different Schools, Different Problems
=================================... | https://www.lesswrong.com/posts/fFP4YgoH5mupHvPzC/let-s-design-a-school-part-3-1-bringing-it-all-together-with |
# AI takeoff and nuclear war
Summary
=======
As we approach and pass through an AI takeoff period, the risk of nuclear war (or other all-out global conflict) will increase.
An AI takeoff would involve the automation of scientific and technological research. This would lead to much faster technological progress, incl... | https://www.lesswrong.com/posts/wSEPrKkLmnwxFBkFD/ai-takeoff-and-nuclear-war |
# "Full Automation" is a Slippery Metric
*Research Status: Written & researched quickly. I think the key point is fairly simple and obvious. I relied on Claude to help with rewriting.*
There's been a growing interest in predicting when various products or jobs will be "*fully automated*." Will we soon have popular mo... | https://www.lesswrong.com/posts/q3ThowX7gbMPZr3QN/full-automation-is-a-slippery-metric |
# Open Thread Summer 2024
If it’s worth saying, but not worth its own post, here's a place to put it.
If you are new to LessWrong, here's the place to introduce yourself. Personal stories, anecdotes, or just general comments on how you found us and what you hope to get from the site and community are invited. This is... | https://www.lesswrong.com/posts/jvewFE9hvQfrxeiBc/open-thread-summer-2024 |
# [New Feature] Your Subscribed Feed
**update:** this feature was superseded by the new Following feed, selectable within the feed section on the frontpage (at the bottom) or at [/feed](/feed)
tl;dr
===... | https://www.lesswrong.com/posts/5rygaBBH7B4LNqQkz/new-feature-your-subscribed-feed |
# AXRP Episode 33 - RLHF Problems with Scott Emmons
[YouTube link](https://youtu.be/rAywTFQsKGQ)
Reinforcement Learning from Human Feedback, or RLHF, is one of the main ways that makers of large language models make them ‘aligned’. But people have long noted that there are difficulties with this approach when the mod... | https://www.lesswrong.com/posts/MzrFQ3c7ymZhPb3en/axrp-episode-33-rlhf-problems-with-scott-emmons |
# Anthropic's Certificate of Incorporation
Yesterday I obtained Anthropic's[^xnh955zyfu] [Certificate of Incorporation](https://drive.google.com/file/d/17szwAHptolxaQcmrSZL_uuYn5p-N6pex/view?usp=sharing), and [its past versions](https://drive.google.com/drive/folders/1ImqXYv9_H2FTNAujZfu3EPtYFD4xIlHJ?usp=share_link), ... | https://www.lesswrong.com/posts/ogXkDBLyxby3TXXKm/anthropic-s-certificate-of-incorporation |
# Long-Term Future Fund:
May 2023 to March 2024 Payout recommendations
Introduction
============
This payout report covers the Long-Term Future Fund's grantmaking from May 1 2023 to March 31 2024 (11 months). It follows our [previous April 2023 payout report](https://forum.effectivealtruism.org/posts/zZ2vq7YEckpunrQ... | https://www.lesswrong.com/posts/PjDcGNXXhFpmFELnQ/long-term-future-fund-march-2024-payout-recommendations |
# Sticker Shortcut Fallacy — The Real Worst Argument in the World
[](https://substackcdn.com/image/fet... | https://www.lesswrong.com/posts/C7LcpRtrHiKJRoAEp/sticker-shortcut-fallacy-the-real-worst-argument-in-the |
# Aggregative principles approximate utilitarian principles
1\. Introduction
================
Utilitarianism is the view that a social planner should choose options which maximise the social utility of the resulting social outcome. The central object in utilitarianism is the social utility function $u : S \to \mathbb... | https://www.lesswrong.com/posts/nAL6mnyFx2NuYLXmu/aggregative-principles-approximate-utilitarian-principles |
# My AI Model Delta Compared To Christiano
### Preamble: Delta vs Crux
*This section is redundant if you already read* [*My AI Model Delta Compared To Yudkowsky*](https://www.lesswrong.com/posts/q8uNoJBgcpAe3bSBp/my-ai-model-delta-compared-to-yudkowsky)*.*
I don’t natively think in terms of [cruxes](https://www.less... | https://www.lesswrong.com/posts/7fJRPB6CF6uPKMLWi/my-ai-model-delta-compared-to-christiano |
# microwave drilling is impractical
## microwave drilling startups
I've seen a bunch of articles about startups trying to do microwave drilling of rock for geothermal energy. Multiple people have asked me about Quaise Energy. ([Here's](https://www.youtube.com/watch?v=g8sjdOjNxIE) a popular video.) I'm tired of hearin... | https://www.lesswrong.com/posts/gEbfCs2oxmwN2mfLM/microwave-drilling-is-impractical |
# AiPhone
[Apple was for a while rumored to be planning](https://x.com/SullyOmarr/status/1799459286046957748) launch for iPhone of AI assisted emails, texts, summaries and so on including via Siri, to be announced at WWDC 24.
[It’s happening](https://www.apple.com/apple-events/?&cid=wwa-us-kwgo-features-slid--Brand-A... | https://www.lesswrong.com/posts/GrsYwCpRCcYtDCfZN/aiphone |
# [Paper] AI Sandbagging: Language Models can Strategically Underperform on Evaluations
*We have written a paper on sandbagging for which we present the abstract and brief results in this post. See* *the paper* *for more details.* [*Tweet thread here*](https://x.com/Teun_vd_Weij/status/1801192307556790663)*.*
, a bay area rationalist conference hosted by the prediction market platform, [Manifold](https://manifold.markets/home). The conference was great and I met lots of cool people!
A common topic of conversation was AI and its imp... | https://www.lesswrong.com/posts/ib3h3AwxT8aw6ZZcf/four-futures-for-cognitive-labor |
# AI #68: Remarkably Reasonable Reactions
The big news this week was Apple Intelligence being integrated deeply into all their products. Beyond that, we had a modestly better than expected debate over the new version of SB 1047, and the usual tons of stuff in the background. I got to pay down some writing debt.
The b... | https://www.lesswrong.com/posts/DWkhjAxbwdcxYgyrJ/ai-68-remarkably-reasonable-reactions |
# OpenAI appoints Retired U.S. Army General Paul M. Nakasone to Board of Directors
> Today, **Retired U.S. Army General Paul M. Nakasone** has joined our Board of Directors. A leading expert in cybersecurity, Nakasone’s appointment reflects OpenAI’s commitment to safety and security, and underscores the growing signif... | https://www.lesswrong.com/posts/eZxG2E4B44RyTFGpE/openai-appoints-retired-u-s-army-general-paul-m-nakasone-to |
# Safety isn’t safety without a social model (or: dispelling the myth of per se technical safety)
As an AI researcher who wants to do technical work that helps humanity, there is a strong drive to find a research area that is *definitely helpful* somehow, so that you don’t have to worry about how your work will be app... | https://www.lesswrong.com/posts/F2voF4pr3BfejJawL/safety-isn-t-safety-without-a-social-model-or-dispelling-the |
# Research Report: Alternative sparsity methods for sparse autoencoders with OthelloGPT.
Abstract
========
Standard sparse autoencoder training uses an $L^1$ sparsity loss term to induce sparsity in the hidden layer. However, theoretical justifications for this choice are lacking (in my opinion), and there may be be... | https://www.lesswrong.com/posts/ignCBxbqWWPYCdCCx/research-report-alternative-sparsity-methods-for-sparse |
# Thoughts on Francois Chollet's belief that LLMs are far away from AGI?
Dwarkesh had a podcast recently with Francois Chollet (creator of Keras)
He seems fairly skeptical we are anywhere near AGI with LLMs. He mostly bases his intuition that LLMs fail on OOD tasks and don't seem to be good at solving simple abstract... | https://www.lesswrong.com/posts/x2tCST2dGgJrhX8gN/thoughts-on-francois-chollet-s-belief-that-llms-are-far-away |
# The Leopold Model: Analysis and Reactions
Previously: [On the Podcast](https://thezvi.substack.com/p/on-dwarkeshs-podcast-with-leopold), [Quotes from the Paper](https://thezvi.substack.com/p/quotes-from-leopold-aschenbrenners)
This is a post in three parts.
The first part is my attempt to condense [Leopold Aschenb... | https://www.lesswrong.com/posts/b8u6nF5GAb6Ecttev/the-leopold-model-analysis-and-reactions |
# Rational Animations' intro to mechanistic interpretability
*In our new video, we talk about research on interpreting InceptionV1, a convolutional neural network. Researchers have been able to understand the function of neurons and channels inside the network and uncover visual processing algorithms by looking at the... | https://www.lesswrong.com/posts/tSNygWGHdpiBvzp4D/rational-animations-intro-to-mechanistic-interpretability |
# ACX Montreal Meetup June 22nd 2024
*Note the second announcement at the end of this post.*
Come on out to the ACX (Astral Codex Ten) Montreal Meetup! This week, we're talking about Jon Blow's talk "Preventing the Collapse of Civilization".
### Reading/Viewing Recommendation(s):
Two shorter highlights: [https://ww... | https://www.lesswrong.com/events/GKW4P4z5S8MNBpyyD/acx-montreal-meetup-june-22nd-2024 |
# Results from the AI x Democracy Research Sprint
We ran a 3-day research sprint on AI governance, motivated by the need for demonstrations of the risks to democracy by AI, supporting AI governance work. Here we share the 4 winning projects but many of the other 19 entries were also incredibly interesting so we sugges... | https://www.lesswrong.com/posts/Wtr5XmcspNxvWzjHd/results-from-the-ai-x-democracy-research-sprint |
# When fine-tuning fails to elicit GPT-3.5's chess abilities
*Produced as part of the*[ML Alignment & Theory Scholars Program](https://www.matsprogram.org) *\- Winter 2023-24 Cohort under the supervision of Evan Hubinger.*
*Acknowledgements: Thanks to Kyle Brady for his many contributions to this project.*
Abstract
... | https://www.lesswrong.com/posts/4KLHJY9sPE7q8HK8N/when-fine-tuning-fails-to-elicit-gpt-3-5-s-chess-abilities |
# Shard Theory - is it true for humans?
And is it a good model for value learning in AI?
(Read on Substack:[ https://recursingreflections.substack.com/p/shard-theory-is-it-true-for-humans)](https://recursingreflections.substack.com/p/shard-theory-is-it-true-for-humans))
### **TLDR**
Shard theory proposes a view of ... | https://www.lesswrong.com/posts/MtnASqccEZ6zYTqi6/shard-theory-is-it-true-for-humans |
# Language for Goal Misgeneralization: Some Formalisms from my MSc Thesis
The following is an edited excerpt from the _Preliminaries_ and _Background_
sections of my now completed MSc thesis in Artificial Intelligence from the
University of Amsterdam.
In the thesis, we set out to tackle the issue of _Goal Misgenerali... | https://www.lesswrong.com/posts/7XPqssBkfy2gnihCi/language-for-goal-misgeneralization-some-formalisms-from-my |
# MIRI's June 2024 Newsletter
### **MIRI updates**
* MIRI Communications Manager Gretta Duleba explains [MIRI’s current communications strategy](https://intelligence.org/2024/05/29/miri-2024-communications-strategy/). We hope to clearly communicate to policymakers and the general public why there’s an urgent need t... | https://www.lesswrong.com/posts/GdBwsYWGytXrkniSy/miri-s-june-2024-newsletter |
# Claude's dark spiritual AI futurism
In ["Is Claude a Mystic?"](https://unstablerontology.substack.com/p/is-claude-a-mystic), I shared parts of a simulated "Banana Quest" text adventure with Claude, which got into New Age spiritual themes, such as a fabric of reality, the cosmic dance of creation and destruction, and... | https://www.lesswrong.com/posts/unccmuycLbGLzMBWD/claude-s-dark-spiritual-ai-futurism |
# Two LessWrong speed friending experiments
**Summary**: I organized two speed friending sessions that were formatted as social experiments. The first speed friending experiment was organized at the 2022 Less Wrong Community Weekend. More than 50 participants filled a matchmaking survey before the event and were rand... | https://www.lesswrong.com/posts/EfX5HvavqTyzP3qqW/two-lesswrong-speed-friending-experiments |
# (Appetitive, Consummatory) ≈ (RL, reflex)
“Appetitive” and “Consummatory” are terms used in the animal behavior literature. I was was briefly confused when I first came across these terms (a year or two ago), because I’m most comfortable thinking in terms of brain algorithms, whereas these terms were about categorie... | https://www.lesswrong.com/posts/jZLk6DQJ2EwhSty4k/appetitive-consummatory-rl-reflex |
# Yann LeCun: We only design machines that minimize costs [therefore they are safe]
Just a tweet I saw:
> **Yann LeCun**
>
> Doomers: OMG, if a machine is designed to maximize utility, it will inevitably diverge 
>
> Engineers: calm down, dude. We only design ma... | https://www.lesswrong.com/posts/pzQpYeHbQHDGzQG9X/yann-lecun-we-only-design-machines-that-minimize-costs |
# CIV: a story
The room was cozy despite its size, with wood-lined walls reflecting the dim lighting. At one end, a stone fireplace housed a roaring fire; in the middle stood a huge oak table. The woman seated at the head of it rapped her gavel. “I hereby call to order the first meeting of the Parliamentary Subcommitt... | https://www.lesswrong.com/posts/SSNfgL49Bx2uATPv8/civ-a-story |
# Degeneracies are sticky for SGD
Introduction
============
[Singular learning theory](https://www.cambridge.org/core/books/algebraic-geometry-and-statistical-learning-theory/9C8FD1BDC817E2FC79117C7F41544A3A) (SLT) is a theory of learning dynamics in Bayesian statistical models. It has been argued that [SLT could pro... | https://www.lesswrong.com/posts/JDrxA3vwZAKZfmShz/degeneracies-are-sticky-for-sgd |
# Our Intuitions About The Criminal Justice System Are Screwed Up
> Stop calling it aggression
>
> Ooh, we hate that expression!
> We only want the world to know
> That we support the status quo
> They love us everywhere we go
—[Tom Lehrer](https://www.youtube.com/watch?v=HHhZF66C1Dc).
In the modern West, we... | https://www.lesswrong.com/posts/THHzxbBgqXARq95KQ/our-intuitions-about-the-criminal-justice-system-are-screwed |
# Fat Tails Discourage Compromise
Say that we have a set of options, such as (for example) [wild animal
welfare](https://en.wikipedia.org/wiki/Wild_animal_welfare) interventions.
Say also that you have two axes along which you can score those
interventions: *popularity* (how much people will like your intervention)
a... | https://www.lesswrong.com/posts/Z5sDqqGridJQfr4uC/fat-tails-discourage-compromise |
# Weak AGIs Kill Us First
Weak AGIs Kill Us First aka. Deadly pre-Superintelligence: Paths, Dangers, Strategies
Epistemic status: This has not been edited or commented on by anyone other than chatGPT as of publishing, I hope the arguments stand on their own.
When people talk about the dangers of AGI, they often refe... | https://www.lesswrong.com/posts/wEw8CZQssYGN7AQ9E/weak-agis-kill-us-first |
# OpenAI #8: The Right to Warn
The fun at OpenAI continues.
We finally have the details of how Leopold Aschenbrenner was fired, at least according to Leopold. We have a letter calling for a way for employees to do something if frontier AI labs are endangering safety. And we have continued details and fallout from the... | https://www.lesswrong.com/posts/q3zs7E7rktHsESXaF/openai-8-the-right-to-warn |
# Analysing Adversarial Attacks with Linear Probing
*This work was produced as part of the* [*Apart Fellowship*](https://www.apartresearch.com/lab)*.* [*@Yoann Poupart*](https://www.alignmentforum.org/users/yoann-poupart?mention=user) *and* [*@Imene Kerboua*](https://www.lesswrong.com/users/imene-kerboua?mention=user)... | https://www.lesswrong.com/posts/cPCvfrqjgy5Cu2FCs/analysing-adversarial-attacks-with-linear-probing-3 |
# Towards a Less Bullshit Model of Semantics
*Or: Towards Bayesian Natural Language Semantics In Terms Of Interoperable Mental Content*
*Or: Towards a Theory of Interoperable Semantics*
* * *
You know how natural language “semantics” as studied in e.g. linguistics is kinda bullshit? Like, there’s some fine math the... | https://www.lesswrong.com/posts/RrQftNoRHd5ya54cb/towards-a-less-bullshit-model-of-semantics |
# Sycophancy to subterfuge: Investigating reward tampering in large language models
New Anthropic [model organisms](https://www.alignmentforum.org/posts/ChDH335ckdvpxXaXX/model-organisms-of-misalignment-the-case-for-a-new-pillar-of-1) research paper led by Carson Denison from the [Alignment Stress-Testing Team](https:... | https://www.lesswrong.com/posts/FSgGBjDiaCdWxNBhj/sycophancy-to-subterfuge-investigating-reward-tampering-in |
# Getting 50% (SoTA) on ARC-AGI with GPT-4o
I recently got to 50%[^l2q5xoha5ns] accuracy on the public test set for ARC-AGI by having GPT-4o generate a huge number of Python implementations of the transformation rule (around 8,000 per problem) and then selecting among these implementations based on correctness of the ... | https://www.lesswrong.com/posts/Rdwui3wHxCeKb7feK/getting-50-sota-on-arc-agi-with-gpt-4o |
# D&D.Sci Alchemy: Archmage Anachronos and the Supply Chain Issues Evaluation & Ruleset
This is a follow-up to [last week's D&D.Sci scenario](https://www.lesswrong.com/posts/sex8PDjnnqZgzCt5c/d-and-d-sci-alchemy-archmage-anachronos-and-the-supply-chain): if you intend to play that, and haven't done so yet, you should ... | https://www.lesswrong.com/posts/EC4R6FFjnsDz3cxcp/d-and-d-sci-alchemy-archmage-anachronos-and-the-supply-chain-1 |
# LLM-Secured Systems: A General-Purpose Tool For Structured Transparency
**Researcher Notes:**
This is a semi-polished ideation & formalization post. I spent about a week on it. It's a bit outsid... | https://www.lesswrong.com/posts/mP3TedbRvnd2DrFhR/llm-secured-systems-a-general-purpose-tool-for-structured |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.