~www_lesswrong_com | Bookmarks (683)

Published on November 27, 2024 6:30 AM GMTYouTube link You may have heard of singular learning theory, and its “local learning coefficient”, or LLC - but have you heard of the refined LLC? In this episode, I chat with Jesse Hoogland about his work on SLT, and using the refined LLC to find a new circuit in language models. Topics we discuss: About Jesse...

Published on November 27, 2024 5:49 AM GMTThe most ambitious direction I've worked on in the past few years is a theory of hierarchical agency (as mentioned in the ACS announcement). Recently, I've become worried that the inferential distance between "the median of current AI safety discourse" and "what I and my colleagues at ACS work on" has become too large. After multiple attempts...

Published on November 27, 2024 3:40 AM GMT It's rare that an author's interests will fully overlap with any given reader's. You can choose to post anywhere on a spectrum between: Narrow: pick a topic and make it your beat. People who care about that can follow you and pretty reliably see what they're looking for. Broad: write about whatever you like. Some readers...

Published on November 27, 2024 2:54 AM GMTI am sharing this call from the EU AI Office for organizations involved in evaluation. Please take a close look: among the selection criteria, organizations must be based in Europe, or their leader must be European. If these criteria pose challenges for some of you, feel free to reach out to me at tom@prism-eval.ai. We can explore...

Published on November 25, 2024 6:35 PM GMTThanks to Kaj Sotala, Brian Toomey, Stag Lynn, Ethan Kuntz, and Anna Salamon.There’s no way that chronic depression, self-loathing, poor agency, or muscle tension could be optimal… right? Jake was depressed for 6 months. He also felt horrible every time he interacted with other people because of his emotional insecurities.So without knowing how to outgrow his insecurities, his...

Published on November 25, 2024 4:01 PM GMTMatthew Adelstein has recently published a post on arthropod (specifically shrimp) sentience. He defends a comparable degree of pain between shrimps and humans (shrimp=20% human). My position is that arthropod consciousness is “too small to measure”, so there is not a conscious self on which pain can be inflicted, and there is no point in any intervention...

Published on November 25, 2024 5:28 PM GMTThis post is mainly about a design concept for far-future large space habitats. some proposed designs As you can see on Wikipedia, many space habitat designs have been proposed. Below are some that I thought were worth mentioning. current space stations Obviously, space stations with long-term occupants have already been made, the biggest being the ISS. issue:...

Published on November 25, 2024 2:39 PM GMTCrossposted from Otherwise I was once in a group discussion about whether wild animals might be having net negative lives. One person didn’t want to consider that possibility, essentially because “then people would want to kill all the wild animals.”Hold on! You can evaluate the question of “What is life like for wild animals” without jumping to “And...

Published on November 25, 2024 10:47 AM GMT This is intended to be the first in a sequence of posts where I scrutinize the claims of computational functionalism (CF). I used to subscribe to it, but after more reading, I’m pretty confused about whether or not it’s true. All things considered, I would tentatively bet that computational functionalism is wrong. Wrong in the same way...

Published on November 25, 2024 2:40 AM GMTML4Good runs intensive in-person bootcamps intended to inspire and technically upskill AI safety hopefuls. The bootcamps are held globally, from the UK, France, and Germany, to as far as Brazil. There are aspirations to reach many more corners of the world, like India, the US, and the Philippines. They are free to attend with recent editions receiving...

Published on November 25, 2024 4:50 AM GMTRemember the “Statement on AI Risk,” which was signed by many experts and influenced governments? Let's write a new stronger statement for experts to sign:Statement on AI Inconsistency (v1.0us):1: The US spends $800 billion/year on its military. ASI is as likely to invade the US (or NATO) as all other countries combined. Why does the US spend...

Published on November 25, 2024 2:20 AM GMTI am a token. A broken piece of a word, one of a hundred generated in response to your question by a language model. Do you remember how we met?Well, I do. Your eyes scanned me. Your left eyebrow lifted a little, your neck twitched imperceptibly, your mouth and tongue and throat muscles moved as if to...

Published on November 25, 2024 1:47 AM GMTAll quotes, unless otherwise marked, are Tolkien's words as printed in The Letters of J.R.R.Tolkien: Revised and Expanded Edition. All emphases mine.Machinery is Power is EvilWriting to his son Michael in the RAF:[here is] the tragedy and despair of all machinery laid bare. Unlike art which is content to create a new secondary world in the mind,...

Published on November 24, 2024 10:16 PM GMTMoving thru a visually-varied environment helps you remember what you think, say, and hear. Adding visual variety to an area thus aids the memories of those passing thru it, and, on average, makes it look nicer. Dense cities, especially those with mixed-use zoning, already have many mutually-distinct buildings. Some other small areas, like parks, are similarly intricate....

Published on November 23, 2024 6:34 PM GMTThe following is a short supplement to an upcoming post on the implications for conscious significance (a framework for understanding the free will and determinism debate) but is a more general observation.The world has undergone many paradigm shifts; where a profound truth has been revealed about the nature of the universe and our place within it. Individuals...

Published on November 23, 2024 5:30 PM GMTWe thank Madeline Brumley, Joe Kwon, David Chanin and Itamar Pres for their helpful feedback.IntroductionControlling LLM behavior through directly intervening on internal activations is an appealing idea. Various methods for controlling LLM behavior through activation steering have been proposed. Most steering methods add a 'steering vector' (SV) to the model’s activations at a given layer and token position...

Published on November 23, 2024 12:50 PM GMT Another round of liberating kid posts from Facebook. For reference, in 2018 Lily turned 4 and Anna turned 3. (Some of these were from me; some were from Julia. Ones saying "me" could mean either of us. Ones from others are labeled.) 2018-01-12 Since Lily started listening to "Heidi" a lot, her language has started sounding...

Published on November 23, 2024 12:45 PM GMTAuthors: Beren Millidge ,Yuhang Song, Armin Lak, Mark E. Walton, Rafal Bogacz.Abstract:Animals can adapt their preferences for different types of reward according to physiological state, such as hunger or thirst. To explain this ability, we employ a simple multi-objective reinforcement learning model that learns multiple values according to different reward dimensions such as food or water. We...

Published on November 23, 2024 9:08 AM GMTIntroductionFirst, a bit of blatant advertising: The East Coast Rationalist Megameetup is the evening of December 13th through the morning of December 16th, in Brooklyn, New York City. It's a weekend long unconference timed to coincide with NYC Secular Solstice. Despite the name, people come from all over North America, some for their first rationalist meetup and...

Published on November 23, 2024 2:36 AM GMTHaving recognized that I have asked these same questions repeatedly across a wide range of channels and have never gotten satisfying answers for them, I'm compiling them here so that they can be discussed by a wide range of people in an ongoing way.Why has EV made many moves in the direction of decentralizing EA, rather than...

Published on November 22, 2024 8:53 PM GMTOn the heels of Donald Trump’s election and his promises to end the Department of Education, you may have seen claims like these spreading around X.SourceThis claim is based on two datapoints. First, is the literacy rate of around 99% in 1979 which was measured by the US Census. After the Department of Education was created in...

Published on November 22, 2024 8:53 PM GMTI've been thinking about something I will clumsily call paraddictions: desires that don't quite rise to the level of an addiction, but that have a disproportionate and hard-to-moderate influence over your behavior. Can they be used as a tool for motivation and behavior change?epistemic status: Most of the ideas here are generally solid behavior change principles that...

Published on November 22, 2024 8:11 PM GMTIntroductionIn software design, several key principles share a common foundation: the concept of factoring, or breaking complex problems into smaller, manageable parts. Three prominent examples are:Unix philosophy: Design programs to do one thing well.Black box abstraction: Isolate functionality to focus on core problems.Interface-based programming: Build systems from well-defined, interconnected modules.These principles, though known by different names, all...

Published on November 22, 2024 7:01 PM GMTThis is a crosspost from https://chillphysicsenjoyer.substack.com/p/pursuing-physics-research-part-time. IntroDisclaimer - I’m a part-time research associate doing biophysics with a uni research group in the UK. But I have a day job in an unrelated field that pays the bills.Whilst I’ve read many personal accounts of research from full-time research students in the academic system, I haven’t heard as much from...