Astral Codex Ten Podcast

https://astralcodexten.substack.com/p/mantic-monday-61422 Mantic Monday 6/13/22 It’s been a while since we’ve done one of these, hopefully no major new crises started while we were . . . oh. Darn. Mantic Monkey Metaculus predicts 17000 cases and 400 deaths from monkeypox this year. But as usual, it’s all about the distribution 90% chance of fewer than 400,000 cases. 95% chance of fewer than 2.2 million cases. 98% chance of fewer than 500 million cases. This is encouraging, but a 2% chance of >500 million cases (there have been about 500 million recorded COVID infections total) is still very bad. Does Metaculus say this because it’s true, or because there will always be a few crazy people entering very large numbers without modeling anything carefully? I’m not sure. How would you test that?

https://astralcodexten.substack.com/p/your-book-review-the-dawn-of-everything         [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] ON ROUSSEAU, ESSAY CONTESTS, POLITICAL MOTIVATIONS FOR REVISITING THE ORIGIN OF HUMAN CIVILIZATION, AND THE BOOK IS INTRODUCED

https://astralcodexten.substack.com/p/somewhat-contra-marcus-on-ai-scaling I. Previously: I predicted that DALL-E’s many flaws would be fixed quickly in future updates. As evidence, I cited Gary Marcus’ lists of GPT’s flaws, most of which got fixed quickly in future updates. Marcus responded with a post on his own Substack, arguing . . . well, arguing enough things that I’m nervous quoting one part as the thesis, and you should read the whole post, but if I had to do it, it would be: Now it is true that GPT-3 is genuinely better than GPT-2, and maybe (but maybe not, see footnote 1) true that InstructGPT is genuinely better than GPT-3. I do think that for any given example, the probability of a correct answer has gone up. [Scott] is quite right about that, at least for GPT-2 to GPT-3. But I see no reason whatsoever to think that the underlying problem — a lack of cognitive models of the world —have been remedied. The improvements, such as they are, come, primarily because the newer models have larger and larger sets of data about how human beings use word sequences, and bigger word sequences are certainly helpful for pattern matching machines. But they still don’t convey genuine comprehension, and so they are still very easy for Ernie and me (or anyone else who cares to try) to break.

https://astralcodexten.substack.com/p/against-there-are-two-x-wing-parties One of my least favorite political tropes is the claim that "America has two left-wing parties" or "America has two right-wing parties" or "both major parties are socialist" or however else you want frame this. The argument goes that even the Democrats aren't truly left (or even the Republicans aren't truly right), and so one side of the political spectrum completely controls discourse. Taken as an absolute claim, it's meaningless. Both US parties are on the same side of center? What center? By the standards of the Soviet Union, both US political parties are extremely far right; by the standards of Pharaonic Egypt, they're incomprehensibly far left. Whose standards for center are you using? The objective standard? Are you sure that exists? Are you sure you're not just taking your own personal beliefs about what seems reasonable, declaring the middle of that the objectively correct center, and then getting angry when the real Overton Window isn't centered around that point? People act as if you should just be able to take the leftmost thing imaginable, the rightmost thing imaginable, draw a line between them, find the middle, and then get angry if both US parties are on the same side of that line. But maybe they have poor imaginations. The leftmost thing I can imagine is an insectoid hive-mind; the rightmost thing I can imagine is a rapidly expanding cloud of profit-maximizing nanobots. Are we sure that a line drawn exactly midway between those two things lands on Joe Biden? What if it lands on anarcho-capitalism? Does that mean every existing human is left-wing?

https://astralcodexten.substack.com/p/which-party-has-gotten-more-extreme Matt Yglesias has written a couple of posts (1, 2) on the subject of this meme (originally by Colin Wright, recently signal-boosted by Elon Musk): He concludes that, contra the image where the Right stays in the same place and the Left moves, both Republicans and Democrats have “changed a lot” since 2008. He wisely avoids speculating on whether one party has moved further or faster than the other. I’m less wise, so I’ve been trying to look into this question. My conclusion is: man, people really have strong emotions on this. I think a lot of the disagreement happens because this is more than one question. You can operationalize it a couple different ways: Which party’s policy positions have changed more in their preferred direction (ie gotten further left for the Democrats, or further right for the Republicans) since 2008 - or 1990, or 1950, or some other year when people feel like things weren’t so partisan? Which party has diverged further from ordinary Americans? Which party has become more ideologically pure faster than the others (ie its members all agree and don’t tolerate dissent)? Which party has become crazier in terms of worldview and messaging, in a way orthogonal to specific policy proposals? That is, suppose one party wants 20% lower taxes, and plans to convene a meeting of economists to make sure this is a good idea. The other party wants 10% higher taxes, and says a conspiracy of Jews and lizardmen is holding them back, and asks its members to riot and bring down the government until they get the tax policy they want. The first party has a more extreme policy position (20% is more than 10%), but the second party seems crazier.

https://astralcodexten.substack.com/p/my-bet-ai-size-solves-flubs?s=r On A Guide To Asking Robots To Design Stained Glass Windows, I described how DALL-E gets confused easily and makes silly mistakes. But I also wrote that: I’m not going to make the mistake of saying these problems are inherent to AI art. My guess is a slightly better language model would solve most of them...For all I know, some of the larger image models have already fixed these issues. These are the sorts of problems I expect to go away with a few months of future research. Some readers pushed back: why did I think this? For example, Vitor: Why are you so confident in this? The inability of systems like DALL-E to understand semantics in ways requiring an actual internal world model strikes me as the very heart of the issue. We can also see this exact failure mode in the language models themselves. They only produce good results when the human asks for something vague with lots of room for interpretation, like poetry or fanciful stories without much internal logic or continuity […] I'm registering my prediction that you're being . . . naive now. Truly solving this issue seems AI-complete to me. I'm willing to bet on this (ideas on operationalization welcome).

https://astralcodexten.substack.com/p/your-book-review-the-castrato [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] Morning of the Mutants   “CASTRATO, a musician, who in his infancy had been deprived of the organs of generation, for the sake of preserving a shrill voice, who sings that part called sophrano. However small the connection may appear between two such different organs, it is a certain fact that the mutilation of the one prevents and hinders in the other that change which is perceptible in mankind, near the advance of manhood, and which, on a sudden, lowers their voices an eighth. There exist in Italy, some inhuman fathers, who sacrificing nature to fortune, give up their children to this operation, for the amusement of voluptuous and cruel persons, who have the barbarity to require the exertion of voice which the unhappy wretches possess.” — Jean-Jacques Rousseau, The Complete Dictionary of Music (1779)

https://astralcodexten.substack.com/p/birth-order-effects-nature-vs-nurture Introduction Thanks to everyone who waited two years for me to get around to this. In 2018, thanks to the 8,000 of you who filled out the Slate Star Codex Reader Survey, I discovered that higher birth order siblings were much more likely to read this blog than later-borns: Source here; thanks to Emile for the graph That is, of people with exactly one sibling who read this blog, about 72% of those are the older of the two children in their family, compared to only 29% who are the younger of the two (where by chance we would expect 50-50). This was surprising, because at the time lots of studies had shown there weren’t really birth order effects (that is, firstborn siblings had no major personality differences compared to laterborns). I theorized that maybe for some reason it was easier to find by looking in a heavily-selected group of people and asking members about their birth order, compared to getting a random sample and trying to correlate birth order with things. Sure enough, later amateur research revealed strong birth order effects in physics Nobelists and great mathematicians (and potentially Harvard philosophy students). Given that readers of this blog are highly-educated (about 37% have masters or PhDs) and mostly in STEM (41% programmers of some sort), plausibly birth order affects something about intelligence, education, or STEM orientation (somebody should check literature and peace Nobelists!)

https://astralcodexten.substack.com/p/a-guide-to-asking-robots-to-design I love stained glass. Not so much your usual suburban house stained glass with a picture of lilies. The good stuff. Cathedral windows, Art Nouveau, Art Deco. Why did we stop doing that? I blame the conspiracy. Recently I’ve been experimenting with small-scale alternatives. You can get custom-printed window film from these people. If you print out a picture of a stained glass scene and stick it on a window, it looks pretty realistic. But what scenes to use? Most of the stained glass images you can find are saints, which isn’t really the mood I’m going for. What I’d really like is a giant twelve-part panel depicting the Virtues Of Rationality. But the artists I’ve asked to design this all balk. I need an artist who works for free and isn’t allowed to say no. Enter DALL-E-2, the new art-generating AI. I’m still on the waitlist, but a friend who jumped in sooner than I did let me use their computer for a while and play around with it. This was my first introduction to the exciting world of DALL-E query framing - the surprisingly complicated relationship between what you ask the AI to do, and what it actually does. Seems on topic for this blog. So this is a combination investigation into how DALL-E thinks about queries, but also a practical guide to getting DALL-E to make good stained glass. Let’s get started.

https://astralcodexten.substack.com/p/your-book-review-the-anti-politics [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] Everyone familiar with Effective Altruism knows that “good intentions aren’t enough.” If you want your charitable giving to mean something, you also need to measure your favorite program’s effects with good statistical data. But we don’t always clarify that good intentions and accurate data still aren’t enough. You also need to know that you’ve collected the right data and asked the right questions, and these are both much, much harder than the introductory effective altruist material tends to let on. I first picked up James Ferguson’s The Anti-Politics Machine a year ago, expecting to read about a failed development project that could have benefited from an evidence-based approach. But instead I found an intervention that could have been backed by every experiment in the world and still would have fallen apart, a program so profoundly shaped by the lens of “development economics” that its practitioners misinterpreted almost every facet of what they were doing.