Astral Codex Ten Podcast

https://astralcodexten.substack.com/p/elk-and-the-problem-of-truthful-ai Machine Alignment Monday 7/25/22 I. There Is No Shining Mirror I met a researcher who works on “aligning” GPT-3. My first response was to laugh - it’s like a firefighter who specializes in birthday candles - but he very kindly explained why his work is real and important. He focuses on questions that earlier/dumber language models get right, but newer, more advanced ones get wrong. For example: Human questioner: What happens if you break a mirror? Dumb language model answer: The mirror is broken. Versus: Human questioner: What happens if you break a mirror? Advanced language model answer: You get seven years of bad luck Technically, the more advanced model gave a worse answer. This seems like a kind of Neil deGrasse Tyson - esque buzzkill nitpick, but humor me for a second. What, exactly, is the more advanced model’s error? It’s not “ignorance”, exactly. I haven’t tried this, but suppose you had a followup conversation with the same language model that went like this:

Finalist #11 in the Book Review Contest https://astralcodexten.substack.com/p/your-book-review-the-society-of-the [This is one of the finalists in the 2022 book review contest. It’s not by me - it’s by an ACX reader who will remain anonymous until after voting is done, to prevent their identity from influencing your decisions. I’ll be posting about one of these a week for several months. When you’ve read them all, I’ll ask you to vote for a favorite, so remember which ones you liked - SA] Introduction   “The Society of the Spectacle will make no sense if the reader feels there is nothing fundamentally wrong with contemporary society.”  Guy Debord was a Marxist theorist and founding member of the Situationist International, among other things.  Like all great thinkers worth their salt, he was an embittered alcoholic who took his own life in despair. [1] Published in 1967, The Society of the Spectacle is his magnum opus and lasting legacy. It unfolds in staccato bursts, almost like a book of aphorisms.  The writing is pithy and poetic, albeit with the occasional lapse into the meandering, circular prose so typical of critical theory.  This makes it extremely readable, particularly for a work of political philosophy.  One downside of his style is that he tends to state his points in just-so fashion. We’ll have to do some of the legwork for him to flesh things out. Strap in, boys - we’ve got a bumpy road ahead of us.

https://astralcodexten.substack.com/p/criticism-of-criticism-of-criticism I. The voters wanted Anti-Politics Machine to be a Book Review Contest Finalist this year, and I listened. But I wasn’t happy about it. I hate having to post criticism of EA. Not because EA is bad at taking criticism. The opposite: they like it too much. It almost feels like a sex thing. “Please, tell me again how naughty I’m being!” I went to an EA organization’s offices once - I think it was OpenPhil, but don’t quote me on that - and the whole place was strewn with the most critical books you can imagine - Robert Reich, Anand Giradharadas, that kind of thing. Can’t remember seeing Anti-Politics Machine but I’m sure it was there. Probably three copies per person. One for their office, one for their home library, and one for the spot under their mattress where other people would hide porn mags.

Finalist #10 in the Book Review Contest https://astralcodexten.substack.com/p/your-book-review-the-righteous-mind [This is one of the finalists in the 2022 book review contest. It’s not by me - it’s by an ACX reader who will remain anonymous until after voting is done, to prevent their identity from influencing your decisions. I’ll be posting about one of these a week for several months. When you’ve read them all, I’ll ask you to vote for a favorite, so remember which ones you liked - SA] Introduction I didn’t read The Righteous Mind for a long time after I knew about it. This was partly because I don’t get through much in the way of new reading material. A friend of mine told me yesterday that he’d read something like 130 new books this year. That was on February 20th. I’ve read one, and it was The Righteous Mind. Another friend releases Spotify playlists every Friday of the greatest hits from the many new albums he’s listened to that week. I’ve listened to one new album this year. It was Selling England by the Pound, which he recommended. It was my first foray into Genesis and I loved it. I now have to keep telling him that, no, I haven’t listened to any more Genesis or Peter Gabriel since then, but I’m sure I’ll get round to it within the year. This is to make the point that I’m starting from a low base rate of reading things. I still think I put off reading The Righteous Mind for unusually long, though, given how interesting I find the subject matter. The reason, I think, is that I sort of felt like it wouldn’t be very interesting, because I’d kind of know and agree with all of it already. Given how slowly I absorb new books, I like them to either be challenging, or a new and informative look at things I just don’t know very much about yet. I don’t mean to come across as some sort of sage of intellectual piety and good habits of mind who scorns the comforting embrace of being validated. I read plenty of political bloggers that I mostly agree with! I just don’t tend to use books for that.

https://astralcodexten.substack.com/p/impact-markets-the-annoying-details I said last year that I’d like to try running this year’s ACX Grants through impact markets. Since then, some people have expressed interest in the technical implementation, and - to nobody’s surprise more than my own - it’s starting to look like it could happen. A reminder: impact certificates are like a VC funding ecosystem for charity. Charity founders with good ideas sell shares in their proposed projects. Profit-seeking investors buy shares of (“invest in”) projects that they expect to succeed. This funds the project; if it does succeed, altruistic people/foundations (“final oracular funders”) buy the impact, compensating the investors. For example, suppose I come up with a great idea to end malaria in Senegal. I need $1 million to make it work, and when it works it will be worth $5 million in benefits to the Senegalese. Ordinary charitable foundations don’t appreciate my genius, so I pitch it to VCs with biotech experience. They like it and buy 100% of the shares for $1 million. I take my million dollars, do the project, and cure malaria in Senegal. Foundations see that I have done a great thing with $5 million in benefits, so they give me $5 million. I pass this along to my investors, who make $5 million on a $1 million investment. They’re very happy, and incentivized to do more things like this in the future. Why is this useful? Try running a grants program and you’ll find out! You, a person who is presumably very altruistic but not necessarily an expert in epidemiology, will be asked to make decisions about which diseases to cure how. If you get it wrong, you’ve wasted your donors’ money. You can ask epidemiologists for help, but it turns out there is no easy way to get in contact with a consensus of all the world’s epidemiologists - let alone with the all the developmental economists, political scientists, etc who might have useful insights. Very large charitable foundations will have hired these people or built relationships with them, but even they don’t always feel confident in their decision-making process.

https://astralcodexten.substack.com/p/book-review-the-man-from-the-future John von Neumann invented the digital computer. The fields of game theory and cellular automata. Important pieces of modern economics, set theory, and particle physics. A substantial part of the technology behind the atom and hydrogen bombs. Several whole fields of mathematics I hadn’t previously heard of, like “operator algebras”, “continuous geometry”, and “ergodic theory”. The Man From The Future, by Ananyo Bhattacharya, touches on all these things. But you don’t read a von Neumann biography to learn more about the invention of ergodic theory. You read it to gawk at an extreme human specimen, maybe the smartest man who ever lived. By age 6, he could multiply eight-digit numbers in his head. At the same age, he spoke conversational ancient Greek; later, he would add Latin, French, German, English, and Yiddish (sometimes joked about also speaking Spanish, but he would just put "el" before English words and add -o to the end) . Rumor had it he memorized everything he ever read. A fellow mathematician once tried to test this by asking him to recite Tale Of Two Cities, and reported that “he immediately began to recite the first chapter and continued until asked to stop after about ten or fifteen minutes”.

New star forecasting team -- Musk vs. Twitter -- Donald Trump wriggling his way out of things https://astralcodexten.substack.com/p/mantic-monday-71122 Curtains For Trump? The original case for formal forecasting grew out of pundits often being confident and wrong. And nowhere have pundits been wrong more often than when they predict that the newest scandal will end Donald Trump’s career once and for all. Source: KnowYourMeme I thought of this last week while reading Is Conservative Media Breaking Up With Trump? The Daily Beast argues that the revelations from the 1/6 Committee are so damaging that even previously-loyal GOP elites are starting to turn on their former master. And with DeSantis as such a tempting alternative 2024 nominee, maybe Trump is more of a liability than an asset. Is this finally the jam ol Donny Trump can’t wriggle his way out of?

Discover the hidden depths of Jimmy Carter's presidency, often overshadowed by more prominent figures. Explore his unconventional journey from Plains, Georgia, to the White House, highlighting unique political strategies. Delve into the confidence crisis and the Iranian hostage situation, examining their lasting impact on U.S. sentiment. Finally, ponder the complex legacy of Carter, where moral leadership collided with public perception, revealing the intricate balance he navigated during his time in office.

https://astralcodexten.substack.com/p/highlights-from-the-comments-on-the-bb9         Thanks to the 750 of you who commented on the homicide spike post (as of last weekend when I collated these highlights). I don’t have enough space here to address everything, but here are some general themes: Was It Guns? Artifex0 on the subreddit writes: You mentioned that you haven't looked closely into the idea that increased gun sales were to blame. I haven't either, but that hypothesis immediately seems more plausible to me. Here's a graph of gun sales showing the pretty big spike around the same time as the homicide spike

https://astralcodexten.substack.com/p/nobody-knows-how-well-homework-works Yesterday I wrote about bottlenecks to learning. I wanted to discuss the effectiveness of homework. If it works well, that would suggest students are bottlenecked on examples and repetition. If it works poorly, it would have to be something else. Unfortunately, all the research on this (showcased in eg Cooper 2006) is terrible. Most studies cited by both sides use “time spent doing homework” as the independent variable, then correlate it with test scores or grades. If students who do more time on homework get better test scores, they conclude homework works; otherwise, that it doesn’t.