~www_lesswrong_com | Bookmarks (664)

Published on August 19, 2024 4:27 PM GMTThere are a lot of points in AI control and scheming that are kind-of-obvious to those who have worked in the field. A lot of those points are unwritten, however; hence this post. Many of the ideas here are from Buck Shlegeris and Ryan Greenblatt, thanks to them and Cody Rushing for feedback. Suppose an AI company has...

Published on August 19, 2024 3:30 PM GMT I maintain a Boston apartment price map, scraping Padmapper's listings to make a heatmap. Each month I run a script to update it, but last month it failed due to an API change. It had been impressively stable: I'd been running the same code unmodified for three years, with only minor modifications since a big rewrite...

Published on August 19, 2024 3:04 PM GMTA lot of the debate on long-term, structural technological unemployment can be summarized in three short statements:Concerns about the speed or scope of labor substitution have often been premature or exaggerated in the past.Labor substitution has been very positive for humanity so far. As many old tasks have been automated, human labor has moved into many new,...

Published on August 19, 2024 2:05 PM GMTOne big hole in my set of frames and models of the world is that I don't understand how government works. I don't understand how the government of my native country (France) works, I don't understand how the government of the country I live in (UK) works, I don't understand how the governments of countries that matter...

Published on August 19, 2024 1:57 PM GMT[Cross-posted from my substack, davekasten.substack.com.  {I never said that I was creative at naming things}]Hi,So here are some things I’ve been thinking about while traveling on policy consulting business over the past few months.Bureaucracy, by James Q. WilsonAs we’ve discussed previously, a decent description of the goal of bureaucracy is that it’s the organizational form that society...

Published on August 19, 2024 10:56 AM GMT"The AI bubble is reaching a tipping point", says Sequoia Capital.AI companies paid billions of dollars for top engineers, data centers, etc. Meanwhile, companies are running out of 'free' data to scrape online and facing lawsuits for the data they did scrape. Finally, the novelty of chatbots and image generators is wearing off for users, and fierce...

Published on August 19, 2024 9:23 AM GMTWarning: This post was written at the start of 2024 as part of the AISC project "Evaluating Alignment Evaluations". We are not especially satisfied with the quality reached, but since we are not planning to work on it anymore, we are releasing it as a Work-In-Progress document.Introduction In this post, we cover:What are propensity evaluations?How do they differ...

Published on August 18, 2024 4:05 PM GMTMeta: I recently entered the dating market, so naturally I have lots of random thoughts on the subject which you all get to suffer through for a while. Your usual diet of dry math and agency theory will resume shortly.Obviously the phrase “true love” has been so thoroughly overdone in so much fiction as to lose all...

Published on August 18, 2024 3:00 PM GMTIntroductionRecently we (Elizabeth Van Nostrand and Alex Altair) started a project investigating chaos theory as an example of field formation.[1] The number one question you get when you tell people you are studying the history of chaos theory is “does that matter in any way?”.[2] Books and articles will list applications, but the same few seem to come up...

Published on August 18, 2024 2:10 PM GMT One of the things I like about playing for contra dances is that you have a lot of freedom about what to play. As long as you meet the minimum requirements for danceable music (108-122bpm, contra phrasing) you can do almost anything. And then if you're a bottom left corner band you might want to play...

Published on August 18, 2024 7:02 AM GMT"Since you are reluctant to provide us with the location of the Rebel base," said Grand Moff Tarkin, "I have chosen to test this station's destructive power on your home planet of Alderaan." "No. Alderaan is peaceful. We have no weapons there. It is a loyal planet under Imperial control. Striking Alderaan would destroy your own resources...

Published on August 18, 2024 5:35 AM GMT"My father didn't fight in the wars. He was a navigator on a spice freighter," said Luke. "That's what your uncle told you. Basic security protocol. You don't tell a young child sensitive information about your participation in an ongoing civil war. Which reminds me. I have here something for you. Your father wanted you to have...

Published on August 18, 2024 3:19 AM GMTIntroduction from BenZhukeepa is a LessWronger who I respect and whose views I'm interested in. In 2018 he wrote the first broadly successful explication of Paul Christiano's research ideas for AI alignment, has spent a lot of time interviewing people in AI about their perspectives, and written some more about neuroscience and agent foundations research. He came...

Published on August 18, 2024 1:36 AM GMTMy heart leaps up when I beholdA rainbow in the sky:So was it when my life began;So is it now I am a man;So be it when I shall grow old,Or let me die!The Child is father of the Man;And I could wish my days to beBound each to each by natural piety.-William WordsworthI am exquisitely attuned...

Published on August 17, 2024 11:42 PM GMTSummary: The collection of a massive first-person video dataset for use in large model pretraining is technologically & economically feasible with modest resources. Such a dataset might enable high-fidelity imitation of human behavior, with positive implications for the capabilities and safety of trained models. Such models may also enable a type of “digital immortality” and provide a...

Published on August 17, 2024 10:47 PM GMTTLDR: Through concrete scenic descriptions illustrate how I expect naive goal specifications to fail, for getting diamonds in Minecraft. Not much beyond the concrete examples is original. Also checkout the excellent Specification Gaming video by Rational Animations. The Setup I am playing Minecraft. I'd like to have an AI companion that can perform all sorts of tasks,...

Published on August 17, 2024 4:41 PM GMTI don't usually post stories on LessWrong so I'm curious to see how this is received.The first time we spoke, you asked me some questions that felt really invasive. I didn’t want that to happen again, so I avoided you the entire following year.So when you said “Hi” at a party and suggested catching up, I hesitated....

Published on August 17, 2024 3:30 PM GMTShort version It has always felt like a gift that people read things I write, and I enjoyed reciprocating that by sharing my writing for free.  Until recently I had enough income to pull that off and still cover material needs. That is no longer true, and while it is not an urgent problem, I would like...

Published on August 17, 2024 2:08 PM GMTI recently argued for Cohabitive Games, games that are designed for practicing negotiation, or for developing intuitions for applied cooperative bargaining, which is one of the names of preference aggregation.I offered the assets for P1, my prototype, to whoever was interested. About 20 people asked for them. I told most of them that I wanted to polish...

Published on August 17, 2024 11:07 AM GMTI have started a new personal blog!I intend to use it to write about more esoteric/lower confidence/epistemologically-sticky topics than elsewhere.You can already read the first post: "Mysticism 101, or: In defence of Natural Language DSLs"Discuss

Published on August 17, 2024 9:57 AM GMTThe agent-like structure problem is a question about how agents in the world are structured. I think rationalists generally have an intuition that the answer looks something like the following:We assume the world follows some evolution law, e.g. maybe deterministically like xn+1=f(xn).mjx-chtml {display: inline-block; line-height: 0; text-indent: 0; text-align: left; text-transform: none; font-style: normal; font-weight: normal; font-size: 100%;...

Published on August 17, 2024 2:36 AM GMTCan SRAM be stacked like NAND flash? background types of memory SRAM has, for each bit, multiple transistors that connect power supplies to the gates of each other. Typically 6 transistors are used, but many variations have been proposed. DRAM uses transistors to connect capacitors to an input/output line when they're written or read. The capacitors lose...

Published on August 17, 2024 1:16 AM GMTTL;DR: We demonstrate that the decoder directions of GPT-2 SAEs are highly structured by finding a historical date direction onto which projecting non-date related features lets us read off their historical time period by comparison to year features.Calendar years are linear: there are as many years between 2000 and 2024, as there are between 1800 and 1824....