text stringlengths 300 320k | source stringlengths 52 154 |
|---|---|
# Aging and the geroscience hypothesis
Cautionary introduction
=======================
I am a biomedical engineer with an interest in the science of aging. The Handbook of the Biology of Aging, 9th ed, is a great 400-page introductory resource, looking at aging on every level from evolutionary and conceptual, to publ... | https://www.lesswrong.com/posts/RNqguj2PdcodYLnkr/aging-and-the-geroscience-hypothesis |
# A transcript of the TED talk by Eliezer Yudkowsky
*The TED talk is available* [*on YouTube*](https://youtu.be/Yd0yQ9yxSYY) *and the* [*TED website*](https://www.ted.com/talks/eliezer_yudkowsky_will_superintelligent_ai_end_the_world)*. Previously, a live recording was published behind the paywall* [*on the conference... | https://www.lesswrong.com/posts/kXHBz2Z5BBgaBiazf/a-transcript-of-the-ted-talk-by-eliezer-yudkowsky |
# Housing and Transit Roundup #5
Recently Tyler Cowen has doubled and tripled down on the idea that congestion pricing in Manhattan would be inefficient. I respond to that in the section “[A Strange Take on Congestion Pricing in New York City.](https://thezvi.substack.com/i/133207950/a-strange-take-on-congestion-prici... | https://www.lesswrong.com/posts/Wn5pWiFwQMbtMa3qY/housing-and-transit-roundup-5 |
# What is some unnecessarily obscure jargon that people here tend to use?
In particular I'm thinking of computer programming jargon used in unrelated contexts ("My algorithm for choosing a restaurant to eat at...") just because all parties to the conversation happen to be computer programmers. I wonder, if there were ... | https://www.lesswrong.com/posts/bM9rZ7RfQRje4HMZa/what-is-some-unnecessarily-obscure-jargon-that-people-here |
# An Overview of the AI Safety Funding Situation
*Note: this post was updated in January 2025 to reflect all available data from 2024.*
Introduction
============
AI safety is a field concerned with preventing negative outcomes from AI systems and ensuring that AI is beneficial to humanity. The field does research on... | https://www.lesswrong.com/posts/WGpFFJo2uFe5ssgEb/an-overview-of-the-ai-safety-funding-situation |
# Compression of morbidity
Previous: [Aging and the geroscience hypothesis](https://www.lesswrong.com/posts/RNqguj2PdcodYLnkr/aging-and-the-geroscience-hypothesis)
This post distills Chapter 19 of the Handbook of the Biology of Aging:
Crimmins, E. M., Zhang, Y. S., Kim, J. K., & Levine, M. E. (2021). Trends in morbi... | https://www.lesswrong.com/posts/NfC5jdaFgHgkxmYqR/compression-of-morbidity |
# Report on modeling evidential cooperation in large worlds
I have published a [report](https://arxiv.org/pdf/2307.04879.pdf) on modeling [evidential cooperation in large worlds](https://longtermrisk.org/msr). I originally started working on the report during a CEA Summer Research Fellowship back in 2018 and have now ... | https://www.lesswrong.com/posts/mKnbHwRc7mXtENNEm/report-on-modeling-evidential-cooperation-in-large-worlds |
# AISN#14: OpenAI’s ‘Superalignment’ team, Musk’s xAI launches, and developments in military AI use
Welcome to the AI Safety Newsletter by the [Center for AI Safety](https://www.safe.ai/). We discuss developments in AI and AI safety. No technical background required.
Subscribe [here](https://newsletter.safe.ai/subscr... | https://www.lesswrong.com/posts/RzwewrTLNcB7xvktQ/aisn-14-openai-s-superalignment-team-musk-s-xai-launches-and |
# Goal-Direction for Simulated Agents
*tldr: consistent LLM failure suggests possible avenue for alignment and control*
*epistemic status: somewhat hastily written, speculative in places, but lots of graphs of actual model probabilities*
Today in ‘surprisingly simple tasks that even the most powerful large language ... | https://www.lesswrong.com/posts/hmPCPyr6JFLEigHJx/goal-direction-for-simulated-agents |
# A review of Principia Qualia
[Principia Qualia](https://opentheory.net/PrincipiaQualia.pdf), by Michael Johnson, is a long document describing an approach to theorizing about and studying consciousness. I have heard enough positive things about the document to consider it worth reviewing.
I will split the paper... | https://www.lesswrong.com/posts/nurQ3X2oRH4Ms7YKi/a-review-of-principia-qualia |
# Flowchart: How might rogue AIs defeat all humans?
*Epistemic status: I put this together for myself mostly off the top of my head without carefully cross-referencing other sources. I'm hoping to flesh this out a lot more and do a much more thorough research job of it at some point, but in the meantime I've decided t... | https://www.lesswrong.com/posts/kuZyapXSJoDaMhuqK/flowchart-how-might-rogue-ais-defeat-all-humans |
# Towards Developmental Interpretability
Developmental interpretability is a research agenda that has grown out of a meeting of the [Singular Learning Theory (SLT)](https://www.lesswrong.com/tag/singular-learning-theory) and AI alignment communities. To mark the completion of the first [SLT & AI alignment summit](http... | https://www.lesswrong.com/posts/TjaeCWvLZtEDAS5Ex/towards-developmental-interpretability |
# What does the launch of x.ai mean for AI Safety?
[x.ai](https://x.ai/) has now launched. It seems worthwhile to discuss both what it means for AI safety and whether people interested in AI safety should [consider applying](https://airtable.com/appO5m0cwztccTdiA/shrtrQqaaoBt4Xm8i) for the company.
**Some thoughts:**... | https://www.lesswrong.com/posts/fccbYpCvTrBFMGmfu/what-does-the-launch-of-x-ai-mean-for-ai-safety |
# Are there any good, easy-to-understand examples of cases where statistical causal network discovery worked well in practice?
When I first read the Sequences, one of the exciting posts was [Causal Diagrams and Causal Models](https://www.lesswrong.com/posts/hzuSDMx7pd2uxFc5w/causal-diagrams-and-causal-models), which g... | https://www.lesswrong.com/posts/EevoaoqBjQb3bryW4/are-there-any-good-easy-to-understand-examples-of-cases |
# Eric Michaud on the Quantization Model of Neural Scaling, Interpretability and Grokking
Eric is a PhD student in the Department of Physics at MIT working with Max Tegmark on improving our scientific/theoretical understanding of deep learning -- understanding what deep neural networks do internally and why they wo... | https://www.lesswrong.com/posts/xKT5oTXDCJNramc7y/eric-michaud-on-the-quantization-model-of-neural-scaling |
# The intelligence-sentience orthogonality thesis
You're familiar with the [orthogonality thesis](https://www.lesswrong.com/tag/orthogonality-thesis). Goals and intelligence are more or less independent for one another. There's no reason to suspect an intelligent AI will also be an aligned AI because it is intelligent... | https://www.lesswrong.com/posts/W5bP5HDLY4deLgrpb/the-intelligence-sentience-orthogonality-thesis |
# Elon Musk announces xAI
Some quotes & few personal opinions:
FT reports
> Musk is also in discussions with a number of investors in SpaceX and Tesla about putting money into his new venture, said a person with direct knowledge of the talks. “A bunch of people are investing in it . . . it’s real and they are ex... | https://www.lesswrong.com/posts/px76zCyJDidizxT5t/elon-musk-announces-xai |
# How can I get help becoming a better rationalist?
Well…
Right now, being ‘a rationalist’ could be said to be a massive part of my identity, at least judging by the absurd amount of time I’ve spent reading posts here, or SSC/ACX, and in a few other places. Yet, I’m still a mere lurker unfamiliar with most of the loc... | https://www.lesswrong.com/posts/isJEaQ9uKCcXuZeF9/how-can-i-get-help-becoming-a-better-rationalist |
# AI #20: Code Interpreter and Claude 2.0 for Everyone
There is no shortage of people willing to talk about every week as a huge week in AI with tons of amazing new releases and announcements. At first this was usually right, then it mostly stopped being right. This week we had Code Interpreter and Claude 2.0 and x.AI... | https://www.lesswrong.com/posts/6YxtwzF9ZEXxJ5wij/ai-20-code-interpreter-and-claude-2-0-for-everyone |
# Accidentally Load Bearing
Sometimes people will talk about Chesterton's Fence, the idea that if you want to change something—removing an apparently useless fence—you should first determine why it was set up that way:
> The gate or fence did not grow there. It was not set up by somnambulists who built it in their sl... | https://www.lesswrong.com/posts/QBeN49SoKpDMX3kKk/accidentally-load-bearing |
# Winners of AI Alignment Awards Research Contest
This post describes the winning submissions of AI Alignment Awards. We offered prizes for novel contributions to the problems of goal misgeneralization and corrigibility. (For more context, see [this post](https://www.lesswrong.com/posts/JNtGxrusJRpx53Q8L/announcing-ai... | https://www.lesswrong.com/posts/zFoAAD7dfWdczxoLH/winners-of-ai-alignment-awards-research-contest |
# An attempt to steelman OpenAI's alignment plan
I don't actually think my attempted steelman is what they currently have in mind, this isn't an attempted Intellectual Turing Test. I take more hope from the resources they are committing to the project and their declared willingness to change their minds as they ... | https://www.lesswrong.com/posts/pFzpRePDkbhEj9d5G/an-attempt-to-steelman-openai-s-alignment-plan |
# Jailbreaking GPT-4's code interpreter
Disclaimer: I don’t know much about cybersecurity. Much of my knowledge comes from asking GPT-3.5 and GPT-4 for advice. These are some results from around 20 hours of playing around with the code interpreter plugin in early-mid May, when most of this was written. I contacted Ope... | https://www.lesswrong.com/posts/KSroBnxCHodGmPPJ8/jailbreaking-gpt-4-s-code-interpreter |
# Activation adding experiments with FLAN-T5
*Produced as part of the* [*SERI ML Alignment Theory Scholars Program*](https://serimats.org/) *\- Summer 2023 Cohort*
Inspired by the [Steering GPT-2-XL by adding an activation vector](https://www.lesswrong.com/posts/5spBue2z2tw4JuDCx/steering-gpt-2-xl-by-adding-an-activa... | https://www.lesswrong.com/posts/c38nAg23YTCzd7m8P/activation-adding-experiments-with-flan-t5 |
# When Someone Tells You They're Lying, Believe Them
*Some people refuse to admit they're wrong, but there's other clues*
a pretzel acrobat apparently
Paul Ehrlich became well-known for his 1968 book *... | https://www.lesswrong.com/posts/hg5nHuxuaat4Ldx7J/when-someone-tells-you-they-re-lying-believe-them |
# Gearing Up for Long Timelines in a Hard World
*Tl;dr: My current best plan for technical alignment is to work on: (1) Improving our gears-level understanding of the internals & their dynamics of advanced AI systems, and (2) work on pathways decorrelated with progress in the first direction - likely in the form of* [... | https://www.lesswrong.com/posts/f4NrqEKMsnRKdjtpx/gearing-up-for-long-timelines-in-a-hard-world |
# Unsafe AI as Dynamical Systems
*\[Thanks to Valerie Morris for help editing this post.\]*
Overview
--------
Large Language Models (LLMs) and their safety properties are often studied from the perspective of [a single pass](https://transformer-circuits.pub/): what is the single next token the LLM will produce? But ... | https://www.lesswrong.com/posts/PniKib7oxh8uhM8xj/unsafe-ai-as-dynamical-systems |
# AI Risk and Survivorship Bias - How Andreessen and LeCun got it wrong
*Purpose of this post/TL;DR: Many people who argue against the existence of large-scale risks associated with AI, including Marc Andreessen or Yann LeCun, use the following argument template (call this the Argument from Overestimation, AFO): "sin... | https://www.lesswrong.com/posts/Doucf9k6dcg8n4yPg/ai-risk-and-survivorship-bias-how-andreessen-and-lecun-got |
# Why was the AI Alignment community so unprepared for this moment?
> Our epistemic rationality has probably gotten way ahead of our instrumental rationality
>
> -Scott Alexander
>
> [A Lesswrong Crypto autopsy](https://www.lesswrong.com/posts/MajyZJrsf8fAywWgY/a-lesswrong-crypto-autopsy)
This is a question post: ... | https://www.lesswrong.com/posts/6jytXo5HmR9HLvkzu/why-was-the-ai-alignment-community-so-unprepared-for-this |
# Rationality, Pedagogy, and "Vibes": Quick Thoughts
I just read [this book review](https://astralcodexten.substack.com/p/your-book-review-the-educated-mind) of Egan's *The Educated Mind*. Here are some thoughts I had, written for all but grammatically directed at the review's anonymous author (like a typical comment!... | https://www.lesswrong.com/posts/WLerqzTwtCwhgeE5B/rationality-pedagogy-and-vibes-quick-thoughts |
# How to use ChatGPT to get better book & movie recommendations
I find ChatGPT gives me better movie recommendations than Netflix when I do the following:
1) Keep a list of books/movies that you like and why you like them
2) Use the prompt "Recommend X movies/books that \[CRITERIA\]. Here's a list of things I like ... | https://www.lesswrong.com/posts/hFPNwirvNsDLScCkE/how-to-use-chatgpt-to-get-better-book-and-movie |
# A Hill of Validity in Defense of Meaning
> If you are silent about your pain, they'll kill you and say you enjoyed it.
>
> —Zora Neale Hurston
Recapping my Whole Dumb Story so far—in a previous post, ["Sexual Dimorphism in Yudkowsky's Sequences, in Relation to My Gender Problems"](http://unremediatedgender.space/20... | https://www.lesswrong.com/posts/ZjXtjRQaD2b4PAser/a-hill-of-validity-in-defense-of-meaning |
# Simplified bio-anchors for upper bounds on AI timelines
*This is a replication and distillation of the extensive analysis made by Ajeya Cotra’s* [*Draft report on AI timelines*](https://www.lesswrong.com/posts/KrJfoZzpSDpnrv9va/draft-report-on-ai-timelines)*. In this work, I aim at making the argument as concise and... | https://www.lesswrong.com/posts/FsfP3e7ZspCPuwaRA/simplified-bio-anchors-for-upper-bounds-on-ai-timelines |
# Robustness of Model-Graded Evaluations and Automated Interpretability
TL;DR Many evaluations and automated interpretability rely on using multiple models to evaluate and interpret each other. One model is given full access to the text output of another model in OpenAI's automated interpretability and many model-grad... | https://www.lesswrong.com/posts/ZbjyCuqpwCMMND4fv/robustness-of-model-graded-evaluations-and-automated |
# Activation adding experiments with llama-7b
*Produced as part of the* [*SERI ML Alignment Theory Scholars Program*](https://serimats.org/) *\- Summer 2023 Cohort*
Following my initial investigation with [activation adding experiments with FLAN-T5](https://www.lesswrong.com/posts/c38nAg23YTCzd7m8P/activation-adding-... | https://www.lesswrong.com/posts/w9yKQzyhsLJEZhvg9/activation-adding-experiments-with-llama-7b |
# A simple way of exploiting AI's coming economic impact may be highly-impactful
Scott Alexander ended [A Lesswrong Crypto Autopsy](https://www.lesswrong.com/posts/MajyZJrsf8fAywWgY/a-lesswrong-crypto-autopsy) with these words
> And for those of us who failed - well, the world is getting way too weird to expect there... | https://www.lesswrong.com/posts/tCPvJhY5DDYxBqFgX/a-simple-way-of-exploiting-ai-s-coming-economic-impact-may |
# Runaway Optimizers in Mind Space
This post is an attempt to make an argument more explicit that sometimes seems to be implied (or stated without deeper explanation) in AI safety discussions, but that I haven’t seen spelled out that directly or in this particular form. This post contains nothing fundamentally new, bu... | https://www.lesswrong.com/posts/Mb96sKbqERmcx22hw/runaway-optimizers-in-mind-space |
# Mech Interp Puzzle 1: Suspiciously Similar Embeddings in GPT-Neo
*I made a series of mech interp puzzles for my MATS scholars, which seemed well received, so I thought I'd share them more widely! I'll be posting a sequence of puzzles, approx one a week - these are real questions about models which I think are intere... | https://www.lesswrong.com/posts/eLNo7b56kQQerCzp2/mech-interp-puzzle-1-suspiciously-similar-embeddings-in-gpt |
# Even briefer summary of ai-plans.com
Currently, newcomers to the field of AI Alignment often struggle to understand the ongoing work and individuals involved, as well as the assumptions, strengths, and weaknesses of each plan.
We believe AI-plans.com will be an easy, centralized way to discover and learn more... | https://www.lesswrong.com/posts/76D2sKhnKNMyJ8YbX/even-briefer-summary-of-ai-plans-com |
# An upcoming US Supreme Court case may impede AI governance efforts
According to [various](https://www.scotusblog.com/2023/05/supreme-court-will-consider-major-case-on-power-of-federal-regulatory-agencies/) [sources](https://www.politico.com/news/2023/05/01/supreme-court-chevron-doctrine-climate-change-00094670), th... | https://www.lesswrong.com/posts/oSrfAYpGLXAeZYmvY/an-upcoming-us-supreme-court-case-may-impede-ai-governance |
# AutoInterpretation Finds Sparse Coding Beats Alternatives
*Produced as part of the* [*SERI ML Alignment Theory Scholars Program*](https://serimats.org/) *\- Summer 2023 Cohort*
*Huge thanks to Logan Riggs, Aidan Ewart, Lee Sharkey, Robert Huben for their work on the sparse coding project, Lee Sharkey and Chris Math... | https://www.lesswrong.com/posts/ursraZGcpfMjCXtnn/autointerpretation-finds-sparse-coding-beats-alternatives |
# Build knowledge base first, or backchain?
Specifically for AI alignment (small field, preparadigmatic, plausibly-short-timelines), but a general ~~principle~~ decision-mechanism could apply to other fields.
What I mean by this is, **what ratio** of pre-learning to filling-in-gaps, leads to deeper and quicker insigh... | https://www.lesswrong.com/posts/maL9GejdnhrgGvEfc/build-knowledge-base-first-or-backchain |
# An Overview of AI risks - the Flyer
[EffiSciences](https://ia.effisciences.org/) recently published a document outlining various types of AI risks destined for a technical audience.
We shared this flyer as part of a hackathon we organized with [Entrepreneur first](https://www.joinef.com/) at [MetaAI](https://ai.met... | https://www.lesswrong.com/posts/wkwAgt4Dz4Zobt4E3/an-overview-of-ai-risks-the-flyer |
# Proof of posteriority: a defense against AI-generated misinformation
Summary
=======
* A *proof of priority* shows that a piece of data must have been created *before* a certain time.
* A *proof of posteriority* shows that a piece of data must have been created *after* a certain time.
* By combining the two, ... | https://www.lesswrong.com/posts/Y8nHoRdmiYxo5qcTa/proof-of-posteriority-a-defense-against-ai-generated |
# Thoughts on “Process-Based Supervision” / MONA
1\. Post Summary / Table of Contents
====================================
In [“How might we align transformative AI if it’s developed very soon?”](https://www.lesswrong.com/posts/rCJQAkPTEypGjSJ8X/how-might-we-align-transformative-ai-if-it-s-developed-very), Holden Kar... | https://www.lesswrong.com/posts/D4gEDdqWrgDPMtasc/thoughts-on-process-based-supervision-mona |
# Conditional on living in a AI safety/alignment by default universe, what are the implications of this assumption being true?
I'm specifically asking this question because I suspect there would be some important changes to the assumptions made on LW, as well as change what's good to do.
Conditional on AI alignment/s... | https://www.lesswrong.com/posts/No8eGtzxqtu8DJHoT/conditional-on-living-in-a-ai-safety-alignment-by-default |
# AI safety technical research - Career review
***Note: this post is a (minorly) edited version of a*** [***new 80,000 Hours career review***](https://80000hours.org/career-reviews/ai-safety-researcher/)***.***
Progress in AI — while it could be hugely beneficial — comes with significant risks. Risks that we’ve argue... | https://www.lesswrong.com/posts/oXkgGAgJ35yNDRvzF/new-career-review-ai-safety-technical-research |
# Sapient Algorithms
I notice my mind runs lots of cached programs. Like "walk", "put away the dishes", "drive home", "go to the bathroom", "check phone", etc.
Most of these can run "on autopilot". I don't know how to define that formally. But I'm talking about how, e.g., I can start driving and get lost in thought a... | https://www.lesswrong.com/posts/tahymCqG9jCJtiy4F/sapient-algorithms |
# Announcing the Existential InfoSec Forum
**Please complete this **[**form**](https://airtable.com/shrbfFP3uZathQzC2)** before 20 July to register your interest in attending.**
We're pleased to announce the first Existential InfoSec Forum, a half-day event aimed at strengthening the infosec community pursuing import... | https://www.lesswrong.com/posts/bK2KeajgoxFH8geQN/announcing-the-existential-infosec-forum |
# [Interview w/ Rob Miles] The case for taking AI Safety seriously
I recently concluded a fascinating interview with science communicator Rob Miles on the [Futurati Podcast](https://www.youtube.com/watch?v=r-SmqbqIiXk).
We cover:
* His background in computer science, and how he became interested in the broader AI ... | https://www.lesswrong.com/posts/mR5PrrcRtpcJdAYxB/interview-w-rob-miles-the-case-for-taking-ai-safety |
# Existential Risk Persuasion Tournament
I participated last summer in Tetlock's [Existential Risk Persuasion
Tournament](https://forecastingresearch.org/xpt) (755(!) page paper
[here](https://forecastingresearch.org/s/XPT.pdf)).
Superforecasters and "subject matter experts" engaged in a hybrid
between a predict... | https://www.lesswrong.com/posts/YTPtjExcwpii6NikG/existential-risk-persuasion-tournament |
# Highlights from The Industrial Revolution, by T. S. Ashton
[*The Industrial Revolution, 1760-1830*](https://www.amazon.com/Industrial-Revolution-1760-1830-Thomas-Southcliffe-ebook/dp/B000V6L63S/), by Thomas S. Ashton, is classic in the field, published in 1948. Here are some of my highlights from it. (Emphasis in bo... | https://www.lesswrong.com/posts/znbKmXrm6PBZ3kQD5/highlights-from-the-industrial-revolution-by-t-s-ashton |
# Predictive history classes
Epistemic status: serious proposal [with known difficulties and problems](https://www.lesswrong.com/posts/7ZHZZkyCSnWZvwfjY/problems-with-predictive-history-classes)
Why do we study history? There are many potential reasons. More specifically, why do we teach history to everyone going to ... | https://www.lesswrong.com/posts/6injyEhNT6fWKFZuz/predictive-history-classes |
# The shape of AGI: Cartoons and back of envelope
*\[Cross posted on *[*windowsontheory*](https://windowsontheory.org/?p=8... | https://www.lesswrong.com/posts/bku9odAYPyQwHqzCo/the-shape-of-agi-cartoons-and-back-of-envelope |
# Charter Cities: why they're exciting & how they might work
Hello! What follows is a work-in-progress script about the idea of [Charter Cities](https://forum.effectivealtruism.org/topics/charter-cities), which the EA-adjacent youtube channel [RationalAnimations](https://www.youtube.com/@RationalAnimations/videos) pl... | https://www.lesswrong.com/posts/XozdrgjtNaGaLtuGW/charter-cities-why-they-re-exciting-and-how-they-might-work |
# Least-problematic Resource for learning RL?
Well, [Sutton & Barto](http://incompleteideas.net/book/the-book-2nd.html) is the standard choice, [but](https://www.lesswrong.com/posts/Wnqua6eQkewL3bqsF/matt-botvinick-on-the-spontaneous-emergence-of-learning?commentId=BPzWcdeKzWk6Cio2J):
> Superficial, not comprehensive... | https://www.lesswrong.com/posts/ttmmKDTkzuum3fftG/least-problematic-resource-for-learning-rl |
# Measuring and Improving the Faithfulness of Model-Generated Reasoning
**TL;DR**: In two new papers from Anthropic, we propose metrics for evaluating how faithful chain-of-thought reasoning is to a language model's actual process for answering a question. Our metrics show that language models sometimes ignore their ... | https://www.lesswrong.com/posts/BKvJNzALpxS3LafEs/measuring-and-improving-the-faithfulness-of-model-generated |
# Meta announces Llama 2; "open sources" it for commercial use
*See also their Llama 2 website here:* [*https://ai.meta.com/llama*](https://ai.meta.com/llama)*, and their research paper here:* [*https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/*](https://ai.meta.com/research... | https://www.lesswrong.com/posts/9rdpdDerangjYeQGW/meta-announces-llama-2-open-sources-it-for-commercial-use |
# Still no Lie Detector for LLMs
Background
==========
This post is a short version of a paper we wrote that you can find [here](https://arxiv.org/abs/2307.00175). You can read this post to get the core ideas. You can read the paper to go a little deeper.
The paper is about probing decoder-only LLMs for their belief... | https://www.lesswrong.com/posts/bCQbSFrnnAk7CJNpM/still-no-lie-detector-for-llms |
# Tiny Mech Interp Projects: Emergent Positional Embeddings of Words
*This post was written in a rush and represents a few hours of research on a thing I was curious about, and is an exercise in being less of a perfectionist. I'd love to see someone build on this work! Thanks a lot to Wes Gurnee for pairing with me on... | https://www.lesswrong.com/posts/Ln7D2aYgmPgjhpEeA/tiny-mech-interp-projects-emergent-positional-embeddings-of |
# A brief history of computers
Recently I've been learning about the history of computers. I find it to be incredibly interesting. I'd like to write a post about it to summarize and comment on what I've learned.
I'm a little hesitant though. I'm no expert on this stuff. I'm largely learning about it all for the first... | https://www.lesswrong.com/posts/vfRpzyGsikujm9ujj/a-brief-history-of-computers |
# House Grocery Spending
I was curious what our food spending looked like, so I tried to get a picture of what this has looked like over the past year. We have a good idea of our total spending: we buy most groceries on a [house](https://www.jefftk.com/p/shared-house-setup) credit card so we can split it fairly, and i... | https://www.lesswrong.com/posts/zRC7HBosg6m6Bsge3/house-grocery-spending |
# AISN#15: China and the US take action to regulate AI, results from a tournament forecasting AI risk, updates on xAI’s plan, and Meta releases its open-source and commercially available Llama 2
Welcome to the AI Safety Newsletter by the [Center for AI Safety](https://www.safe.ai/). We discuss developments in AI and A... | https://www.lesswrong.com/posts/Ng7nGixmxz8iFoLi5/aisn-15-china-and-the-us-take-action-to-regulate-ai-results |
# Lisa Feldman Barrett versus Paul Ekman on facial expressions & basic emotions
1\. Summary / Table of Contents
===============================
This post is mostly a book review of [*How Emotions Are Made: The Secret Life of the Brain*](https://lisafeldmanbarrett.com/books/how-emotions-are-made/) by Lisa Feldman Barr... | https://www.lesswrong.com/posts/iYzFKJjzFPRNrqLE3/lisa-feldman-barrett-versus-paul-ekman-on-facial-expressions |
# What Caused the Puzzling Decline in Activism Against Policy Violence Towards Black People?
I find myself surprised by the noticeable decline in the fervor surrounding activism against policy violence towards Black people from its peak in the summer of 2020 to the subsequent years of 2021 through 2023. Initially, I h... | https://www.lesswrong.com/posts/FfegMt4qKfmJcdJAx/what-caused-the-puzzling-decline-in-activism-against-policy |
# Hedonic Loops and Taming RL
*Crossposted from* [*my personal blog*](https://www.beren.io/2023-04-19-Hedonic-loops-taming-RL/)*.*
Everybody knows about the hedonic treadmill. Your hedonic state adjusts to your circumstances over time and quickly reverts to a mostly stable baseline. This is true of basic physiol... | https://www.lesswrong.com/posts/3mwfyLpnYqhqvprbb/hedonic-loops-and-taming-rl |
# Valuism - an approach to life for you to consider
I finally finished my series of 5 essays on my life philosophy (Valuism)! In a nutshell, Valuism suggests that you work to figure out what you intrinsically value, and then that you try to use effective methods to create more of what you intrinsically value.
While ... | https://www.lesswrong.com/posts/733yH5pdrRcPBuhG2/valuism-an-approach-to-life-for-you-to-consider |
# Zener Science
**Summary**: Parapsychology (the study of unorthodox mental phenomena such as hypnosis or telepathy) is widely considered to be the study of something that doesn’t exist. Many parapsychology experiments are uncommonly good ways to practice science, since the experiments can be performed without expensi... | https://www.lesswrong.com/posts/SngycccHpRZNuozPL/zener-science |
# Alignment Grantmaking is Funding-Limited Right Now
For the past few years, I've generally mostly heard from alignment grantmakers that they're bottlenecked by projects/people they want to fund, not by amount of money. Grantmakers generally had no trouble funding the projects/people they found object-level promising,... | https://www.lesswrong.com/posts/SbC7duHNDHkd3PkgG/alignment-grantmaking-is-funding-limited-right-now |
# Incident reporting for AI safety
*Zach Stein-Perlman, Sebastian Lodemann, Štěpán Los, and Moritz von Knebel*
*We made this (quickly) because a resource on* incident reporting as a tool for AI safety *should exist; we are not experts; suggestions are welcome.*
*Zach recommends engaging with this post by reading §1 ... | https://www.lesswrong.com/posts/Hz5cjNZAQozadAiDW/incident-reporting-for-ai-safety-2 |
# mental number lines
[This paper](https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8946762/) says:
> it was shown that second-grade children unevenly placed numbers on a 0–1000 number line as their placement pattern was best fit by a logarithmic rather than by a linear function. The same age group placed numbers more li... | https://www.lesswrong.com/posts/y3srn8cjBBdaKejYm/mental-number-lines |
# Using predictors in corrigible systems
*\[This was a submission to the AI Alignment Awards corrigibility contest that won an* [*honorable mention*](https://www.alignmentawards.com/winners#faqs)*. It dodges the original framing of the problem and runs off on a tangent, and while it does outline the shape of possible ... | https://www.lesswrong.com/posts/LR8yhJCBffky8X3Az/using-predictors-in-corrigible-systems |
# Project Lawful Audiobook: An Unofficial Fan Production with ElevenLabs AI
[Eliezer Yudkowsky's latest fiction work](https://www.projectlawful.com/), (described [here](https://www.lesswrong.com/posts/SA9hDewwsYgnuscae/projectlawful-com-eliezer-s-latest-story-past-1m-words)) co-authored with Lintamande, is now complet... | https://www.lesswrong.com/posts/fzHC8QHevyMngzGEg/project-lawful-audiobook-an-unofficial-fan-production-with |
# Contra Contra the Social Model of Disability
**Epistemic Status:** First draft, written quickly because I have to pack for my brother's wedding and want to get this out while it's hot, or at least lukewarm.
**Summary:**
* This is a tedious, step-by-step rebuttal of Scott Alexander's recent post, [Contra the Soci... | https://www.lesswrong.com/posts/ejxFqBaWZxZKrjXif/contra-contra-the-social-model-of-disability |
# Speculative inferences about path dependence in LLM supervised fine-tuning from results on linear mode connectivity and model souping
*TL;DR: I claim that supervised fine-tuning of the existing largest LLMs is likely path-dependent (different random seeds and initialisations have an impact on final performance and m... | https://www.lesswrong.com/posts/rYdRiaA3cuioJxmBv/speculative-inferences-about-path-dependence-in-llm |
# Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla
*Cross-posting a paper from the Google DeepMind mech interp team, by: Tom Lieberum, Matthew Rahtz, János Kramár, Neel Nanda, Geoffrey Irving, Rohin Shah, Vladimir Mikulik*
 case for predicting what Aliens value
The case
========
Most of the future things we care about – i.e., *(dis)value* – come, in expectation, from futures where humanity develops artificial general intelligence (AGI) and colonizes many other stars ([Bostrom 2003](https://nickbostrom.com/astronomical/wast... | https://www.lesswrong.com/posts/umhsJqwTSKgmhvZ7c/the-short-case-for-predicting-what-aliens-value |
# Even Superhuman Go AIs Have Surprising Failure Modes
In March 2016, [AlphaGo](https://www.deepmind.com/research/highlighted-research/alphago) defeated the Go world champion Lee Sedol, winning four games to one. Machines had finally become superhuman at Go. Since then, Go-playing AI has only grown stronger. The supre... | https://www.lesswrong.com/posts/DCL3MmMiPsuMxP45a/even-superhuman-go-ais-have-surprising-failure-modes |
# Going Beyond Linear Mode Connectivity: The Layerwise Linear Feature Connectivity
Hi! I am excited to share our new work *Going Beyond Linear Mode Connectivity: The Layerwise Linear Feature Connectivity* **\[NeurIPS 2023 Accepted\]** with the LessWrong Community. In our new [work](https://arxiv.org/pdf/2307.08286.pd... | https://www.lesswrong.com/posts/fGqrtgQFmERFqyTib/going-beyond-linear-mode-connectivity-the-layerwise-linear |
# Boundary Placement Rebellion
The idea for this post all started because I was confused about the concept of "narcissism". I learned about "narcissism" from reading differential psychology, where they often measure it by asking people whether they agree with statements such as:
* I have a natural talent for influe... | https://www.lesswrong.com/posts/FDkh5xnuFyqqJcqbb/boundary-placement-rebellion |
# Progress links and tweets, 2023-07-20: “A goddess enthroned on a car”
**Opportunities**
-----------------
* The Roots of Progress Blog-Building Intensive endorsed by [@sapinker](https://twitter.com/sapinker/status/1678786030802022401), [@vpostrel](https://twitter.com/vpostrel/status/1678810466917941249), [@johank... | https://www.lesswrong.com/posts/3JxgwAdRfvH6GT3go/progress-links-and-tweets-2023-07-20-a-goddess-enthroned-on |
# Growth of Publicly Available Genetic Sequencing Data
The largest source of publicly available genetic sequencing data is the [Sequence Read Archive](https://en.wikipedia.org/wiki/Sequence_Read_Archive) (SRA), a joint project of the US ( [NCBI](https://en.wikipedia.org/wiki/National_Center_for_Biotechnology_Informati... | https://www.lesswrong.com/posts/8pvQndfsSDaSXNX4e/growth-of-publicly-available-genetic-sequencing-data |
# All AGI Safety questions welcome (especially basic ones) [July 2023]
**tl;dr: Ask questions about AGI Safety as comments on this post, including ones you might otherwise worry seem dumb!**
Asking beginner-level questions can be intimidating, but everyone starts out not knowing anything. If we want more people in th... | https://www.lesswrong.com/posts/xroYDAE6EoisrSFZf/all-agi-safety-questions-welcome-especially-basic-ones-july-2 |
# AI #21: The Cup Overfloweth
By the Matt Levine Vacation Rule, I took several days to go to Seattle and there was a truly epic amount of news. We had x.AI, Llama 2, upgrades to ChatGPT, a profile of Anthropic, a ton of very interesting papers on a variety of topics, several podcasts that demand listening, fully AI-ge... | https://www.lesswrong.com/posts/BHdEvjtfwpgrTh825/ai-21-the-cup-overfloweth |
# Announcement: AI Narrations Available for All New LessWrong Posts
[TYPE III AUDIO](https://blog.type3.audio/) is running an experiment with the LessWrong team to provide automatic AI narrations on all new posts. All new LessWrong posts will be available as AI narrations (for the next few weeks).
You might have noti... | https://www.lesswrong.com/posts/b9oockXDs2xMdYp66/announcement-ai-narrations-available-for-all-new-lesswrong |
# Problems with predictive history classes
Here I collect and (partly) address problems noted by commenters on ["Predictive history classes"](https://www.lesswrong.com/posts/6injyEhNT6fWKFZuz/predictive-history-classes), in no particular order. I do not expect to refute all of them. Some of them may actually cripple t... | https://www.lesswrong.com/posts/7ZHZZkyCSnWZvwfjY/problems-with-predictive-history-classes |
# GPT-2's positional embedding matrix is a helix
In the context of transformer models, the "positional embedding matrix" is the thing that encodes the meaning of positions within a prompt. For example, given the prompt:
`Hello my name is Adam`
the prompt would generally be broken down into tokens as follows:
`['<|e... | https://www.lesswrong.com/posts/qvWP3aBDBaqXvPNhS/gpt-2-s-positional-embedding-matrix-is-a-helix |
# Decoding intermediate activations in llama-2-7b
*Produced as part of the* [*SERI ML Alignment Theory Scholars Program*](https://serimats.org/) *\- Summer 2023 Cohort*
Existing research, such as the post [interpreting GPT: the logit lens](https://www.lesswrong.com/posts/AcKRB8wDpdaN6v6ru/interpreting-gpt-the-logit-l... | https://www.lesswrong.com/posts/fJE6tscjGRPnK8C2C/decoding-intermediate-activations-in-llama-2-7b |
# Training Process Transparency through Gradient Interpretability: Early experiments on toy language models
*The work presented in this post was conducted during the SERI MATS 3.1 program. Thank you to Evan Hubinger for providing feedback on the outlined experiments.*
*Note: This post was drafted prior to the announc... | https://www.lesswrong.com/posts/DtkA5jysFZGv7W4qP/training-process-transparency-through-gradient |
# Priorities for the UK Foundation Models Taskforce
The UK government recently established the [Foundation Models Taskforce](https://www.gov.uk/government/news/tech-entrepreneur-ian-hogarth-to-lead-uks-ai-foundation-model-taskforce), focused on AI safety and backed by £100M in funding. Founder, investor and AI expert ... | https://www.lesswrong.com/posts/YAgQysWqoBJ7eA7Np/priorities-for-the-uk-foundation-models-taskforce |
# BCIs and the ecosystem of modular minds
*Crossposted from* [*my personal blog*](https://www.beren.io/2023-04-23-Composable-latent-spaces-BCIs-modular-minds/)*.*
**Epistemic status**: *Much more speculative than previous posts but points towards an aspect of the future that is becoming clearer which I think is under... | https://www.lesswrong.com/posts/d74pb97TAqNKwJkc5/bcis-and-the-ecosystem-of-modular-minds |
# The UAP Disclosure Act of 2023 and its implications
**TL;DR:** A recently proposed (7/13/2023) bipartisan amendment to the NDAA for 2024, with senate majority leader Chuck Schumer as its first sponsor and with the aim of declassifying documents related to Unidentified **Anomalous** Phenomena (UAP). It explicitly def... | https://www.lesswrong.com/posts/8JXv9ADBGuDzFL4EZ/the-uap-disclosure-act-of-2023-and-its-implications |
# News : Biden-Harris Administration Secures Voluntary Commitments from Leading Artificial Intelligence Companies to Manage the Risks Posed by AI
Here are the extracts naming the companies and specific commitments :
As part of this commitment, President Biden is convening seven leading AI companies at the White... | https://www.lesswrong.com/posts/K49G5XSinhoAknncQ/news-biden-harris-administration-secures-voluntary |
# Reward Hacking from a Causal Perspective
*Post 4 of *[*Towards Causal Foundations of Safe AGI*](https://www.alignmentforum.org/s/pcdHisDEGLbxrbSHD)*, preceded by *[*Post 1: Introduction*](https://www.alignmentforum.org/posts/3oQCY4he4zRzF6vQb/introduction-to-towards-causal-foundations-of-safe-agi)*, *[*Post 2: Causa... | https://www.lesswrong.com/posts/aw5nqamqtnDnW8w9u/reward-hacking-from-a-causal-perspective |
# Cooking Air Quality
Our stove doesn't have a hood that vents to the outside, so when we're cooking something smoky we put a box fan in the window blowing out. I happened to have left my [air quality meter](https://temtopus.com/collections/temtop-co2-monitor/products/temtop-m2000-2nd-generation-air-quality-monitor-fo... | https://www.lesswrong.com/posts/vomstyu7cDdaWGsRF/cooking-air-quality |
# Fundamentally Fuzzy Concepts Can't Have Crisp Definitions:
Cooperation and Alignment vs Math and Physics
**Epistemic status:** *I describe an argument that seems plausible, but could also be entirely off. IE, low confidence.*
Summary of the Argument
=======================
Certain concepts (like "cooperation" or "... | https://www.lesswrong.com/posts/ggdm7GAk2Xwnn2yTW/fundamentally-fuzzy-concepts-can-t-have-crisp-definitions |
# I'm consistently overwhelmed by basic obligations. Are there any paradigm shifts or other rationality-based tips that would be helpful?
I often get panicky and stressed at the thought of the never-ending nature of simple tasks. Laundry and dishes will always pile up; food and other stocks will always need to be resu... | https://www.lesswrong.com/posts/g5toQ4CfbxH4hermf/i-m-consistently-overwhelmed-by-basic-obligations-are-there |
# Replacing the Big Air Purifier
Around Christmas 2021 I put together a big air purifier on top of [our bookcase](https://www.jefftk.com/p/built-in-shelves). Somehow I didn't blog about it or take a picture, but here's it in the background of a birthday:
[... | https://www.lesswrong.com/posts/WTmR4eLo8npmyHQGH/replacing-the-big-air-purifier |
# GPTs' ability to keep a secret is weirdly prompt-dependent
TL;DR
-----
GPT-3 and GPT-4 understand the concept of keeping a password and can simulate (or write a story about) characters keeping the password. However, this is highly contingent on the prompt (including the characters' names, or previously asked questi... | https://www.lesswrong.com/posts/Z4tBreNCxnppoPLtd/gpts-ability-to-keep-a-secret-is-weirdly-prompt-dependent |
# Apollo Neuro Follow Up
Two months ago I recommended the [Apollo Neuro](https://acesounderglass.com/2023/05/08/product-endorsement-apollo-neuro/). If you bought it, I would love it if you filled out [this extremely short form](https://docs.google.com/forms/d/1Kc3Gf1mj9KGqhIBqcCMpVjQYWath601pWXcvd0jEJps/edit) letting ... | https://www.lesswrong.com/posts/haMGdSwrfFJoKgHv4/apollo-neuro-follow-up |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.