Astral Codex Ten Podcast

I was going back over yesterday's post, and something sounded familiar about this paragraph: A very careless plagiarist takes someone else's work and copies it verbatim: "The mitochondria is the powerhouse of the cell". A more careful plagiarist takes the work and changes a few words around: "The mitochondria is the energy dynamo of the cell". A plagiarist who is more careful still changes the entire sentence structure: "In cells, mitochondria are the energy dynamos". The most careful plagiarists change everything except the underlying concept, which they grasp at so deep a level that they can put it in whatever words they want – at which point it is no longer called plagiarism. After rereading it a few times, it hit me. A few days ago, I'd come across this quote from Miss Manners: There are three possible parts to a date, of which at least two must be offered: entertainment, food, and affection. It is customary to begin a series of dates with a great deal of entertainment, a moderate amount of food, and the merest suggestion of affection. As the amount of affection increases, the entertainment can be reduced proportionately. When the affection IS the entertainment, we no longer call it dating. I laughed at it, I thought it was great, and I stored it in my head as the sort of thing I should quote at some point in order to sound witty. And although I wasn't consciously thinking about it at the time, I'm sure the last sentence of my paragraph comes from the last sentence of Miss Manners'. It would be easy to dismiss it as a coincidence, it probably seems like a coincidence to you, I can't explain how I know that the one comes from the other, but when I replay in my mind the process that made me write that, it's obvious that it did.

A machine learning researcher writes me in response to yesterday's post, saying: I still think GPT-2 is a brute-force statistical pattern matcher which blends up the internet and gives you back a slightly unappetizing slurry of it when asked. I resisted the urge to answer "Yeah, well, your mom is a brute-force statistical pattern matcher which blends up the internet and gives you back a slightly unappetizing slurry of it when asked." But I think it would have been true. A very careless plagiarist takes someone else's work and copies it verbatim: "The mitochondria is the powerhouse of the cell". A more careful plagiarist takes the work and changes a few words around: "The mitochondria is the energy dynamo of the cell". A plagiarist who is more careful still changes the entire sentence structure: "In cells, mitochondria are the energy dynamos". The most careful plagiarists change everything except the underlying concept, which they grasp at so deep a level that they can put it in whatever words they want – at which point it is no longer called plagiarism. GPT-2 writes fantasy battle scenes by reading a million human-written fantasy battle scenes, distilling them down to the concept of a fantasy battle scene, and then building it back up from there. I think this is how your mom (and everyone else) does it too. GPT-2 is worse at this, because it's not as powerful as your mom's brain. But I don't think it's doing a different thing. We're all blending experience into a slurry; the difference is how finely we blend it.

Last week OpenAI announced its latest breakthrough. GPT-2 is a language model that can write essays to a prompt, answer questions, and summarize longer works. For example (bold prompt is human-written, all other text is the machine-generated response): Prompt: For today's homework assignment, please describe the reasons for the US Civil War. It is easy to identify why the Civil War happened, because so many people and so many books and so much television and films tell us that it was the cause, that it has something to do with race or economics or religion. There is much agreement that it was essentially a war of slavery on behalf of capitalism, about a century of slavery. But that's not what most people think of when they think of the Civil War. Many people think of the war as about states rights. You're not wrong about that, but it's kind of misleading to say that the Civil War was a conflict between states' rights and federalism. So let's try again. What's the idea that has inspired so many people to go onto the battlefields of Gettysburg and Antietam and Gettysberg and Petersburg and Fredericksburg? The American idea of the republic–a notion of limited government–is a great part of the history. In 1791, Thomas Jefferson said "Our Constitution was made only for a moral and religious people. It is wholly inadequate to the government of any other." The "founding fathers" were basically religious men who were not very concerned with states' rights. The Civil War occurred in reaction to the idea that the federal government was taking far too much power away from the states, that they weren't sovereign. Jefferson himself, as you well know, is an author of the Declaration of Independence who had some really crazy ideas about, well, you know, secession for a variety of reasons. The War Between the States (later the Great Rebellion) was a radical re-imagining of the American idea and not surprisingly, Jefferson's grandson, who took after both his father and his grandfather and wrote about it, was named Thomas, after the president. "The federal and state governments are both contrived for the benefit of the rich and large property," This view of the Founding Fathers has been promoted by some of today's most prominent politicians, including some who hold prominent national offices and have a direct stake in a Republican Party that has been hijacked by the "federalist" wing of the party. So why is it that so many people have an easy-to-spot way of understanding the Civil War that has everything to do with the South and nothing to do with the South? The Confederate flag has been a symbol of racism for a long time, but when the Civil War started and the Civil War ended, many people suddenly began to see it differently. There was a change of heart, which I think is what led people to take the question of why the Civil War happened quite seriously… Although OpenAI calls this a "language model", modeling language necessarily involves modeling the world. Even if the AI was only supposed to learn things like "texts that talk about the Civil War use the word 'Confederate' a lot", that has flowered into a rudimentary understanding of how the Civil War worked. Its training corpus (8 million web pages) was large enough that in the course of learning language it learned the specific idiom and structure of all sorts of different genres and subtopics. For example:

I. Chefs. Hundreds of them. Tall chefs, short chefs, black chefs, white chefs. I pushed forward through them, like an explorer hacking away at undergrowth. They muttered curses at me, but I was stronger than they were. I came to a door. I opened it. Sweet empty space. I shut the door behind me, sat down in the chair. "Hello," I said. "Detective Paul Eastman, pleased to make your acquaintance." "Doctor Zachary LaShay," said the man behind the desk. His little remaining hair was greying; his eyes showed hints of the intellect that had been buried beneath the dullness of an administrative career. "I hope you didn't have any trouble getting here. Did my secretary warn you about the chefs?" "She did not," I said. "Well, forewarned is forearmed," he answered, inanely and incongruously. "But I trust you got my message about the federal investigators?" "Once a federal investigation has started, we'll retreat and let them take over. But two women died here. We can't just not investigate because you tell us you're trying to get the Feds involved." "Yes, ah, of course. It's just that we're a sort of, ah, defense contractor. None of our projects are officially classified, yet, but we were hoping to get someone with a security clearance, in case this touched on sensitive areas." "I won't pry further than I have to, but until someone from the government says something official, this is a matter for city police. Maybe you could start by telling me more about exactly what you do here." "We're the United States' only proverb laboratory. Our mission is to stress-test the nation's proverbs. To provide rigorous backing for the good ones, and weed out the bad ones." "I'd never even heard of your organization before today, I have to admit. And now that I'm here…it's huge! Who pays for all of this?" "Everybody who uses proverbs," said the Doctor, "which is to say, everybody. Consider: he who hesitates is lost. But also: look before you leap. Suppose you're a business executive who spots a time-limited opportunity. What do you do? Hesitate? Or leap without looking? Eggheads devise all sorts of fancy rules about timing the market and relying on studies, but when push comes to shove most people are going to rely on the simple sayings they learned as a child. If you can keep your stock of proverbs more up-to-date than your competitor's, that gives you a big business advantage." A smartly-dressed woman came in, handed Dr. LaShay a cup of boiling liquid. He put it to his lips, then spat. "This is terrible!" he said. "Try it!" I had been expecting it to be tea, but it wasn't. I didn't know what it was. But it was terrible. Somehow too plain, too salty, and too bitter all at once. I gagged. "That settles it!" said the Doctor. "Too many cooks really do spoil the broth. Tricia, tell the chefs they can all go home now." "So that's what you were doing!" I said.

SSRIs are the most widely used class of psychiatric medications, helpful for depression, anxiety, OCD, panic, PTSD, anger, and certain personality disorders (Why should the same drug treat all these things? Great question!) They've been pretty thoroughly studied, but there's still a lot we don't understand about them. The SSC Survey is less rigorous than most existing studies, but its many questions and very high sample size provide a different tool to investigate some of these issues. I asked fifteen questions about SSRIs on the most recent survey and received answers from 2,090 people who had been on SSRIs. The sample included people on all six major SSRIs, but there were too few people on fluvoxamine (15) to have reliable results, so it was not included in most comparisons. Here's what we found: 1. Do SSRIs work? People seem to think so: Made me feel much worse: 6% Made me feel slightly worse: 7.4% No net change in how I felt: 23.7% Made me feel slightly better: 41.4% Made me feel much better: 21.4% Of course, these statistics include the placebo effect and so cannot be taken entirely at face value. 2. Do some SSRIs work better than others? I asked people to rate their experience with the medication, on a scale from 1 to 10. Here were the results: Lexapro (356): 5.7 Zoloft (470): 5.6 Prozac (339): 5.5 Celexa (233): 5.4 Paxil (126): 4.6 Paxil differed significantly from the others; the others did not differ significantly among themselves. In a second question where participants were just asked to rate their SSRIs from -2 ("made me feel much worse") to +2 ("made me feel much better"), the ranking was preserved, and Lexapro also separated from Celexa. This ranking correlates at r = 0.98 (!?!) with my previous study of this taken from drugs.com ratings. I don't generally hear that Paxil is less effective than other SSRIs, but I have heard that it causes worse side effects. The survey question (probably wrongly) encouraged people to rate side effects as "negative efficacy". My guess is that the difference here is mostly driven by side effects.

The podcast discusses the evolution of societal acceptance of gay individuals, from respectable advocates to openly identifying non-respectable individuals. It also explores the impact of controversial figures embracing causes like endocrine disruption. The chapters delve into the influence of respectability in advocacy and the dynamics of respectability within society, including handling disrespectful voices.

I. Zero To One might be the first best-selling business book based on a Tumblr. Stanford student Blake Masters took Peter Thiel's class on startups. He posted his notes on Tumblr after each lecture. They became a minor sensation. Thiel asked if he wanted to make them into a book together. He did. The title comes from Thiel's metaphor that ordinary businessmen like restaurant owners take a product "from 1 to n" (shouldn't this be from n to n+1?) – they build more of something that already exists. But the greatest entrepreneurs bring something "from 0 to 1" – they invent something that has never been seen before. The book has various pieces of advice for such entrepreneurs. Three sections especially struck me: on monopolies, on secrets, and on indefinite optimism. II. A short review can't fully do justice to the book's treatment of monopolies. Gwern's look at commoditizing your complement almost does (as do some tweets). But the basic economic argument goes like this: In a normal industry (eg restaurant ownership) competition should drive profit margins close to zero. Want to open an Indian restaurant in Mountain View? There will be another on the same street, and two more just down the way. If you automate every process that can be automated, mercilessly pursue efficiency, and work yourself and your employees to the bone – then you can just barely compete on price. You can earn enough money to live, and to not immediately give up in disgust and go into another line of business (after all, if you didn't earn that much, your competitors would already have given up in disgust and gone into another line of business, and your task would be easier). But the average Indian restaurant is in an economic state of nature, and its life will be nasty, brutish, and short. This was the promise of the classical economists: capitalism will optimize for consumer convenience, while keeping businesses themselves lean and hungry. And it was Marx's warning: businesses will compete so viciously that nobody will get any money, and eventually even the capitalists themselves will long for something better. Neither the promise nor the warning has been borne out: business owners are often comfortable and sometimes rich. Why? Because they've escaped competition and become at least a little monopoly-like. Thiel says this is what entrepreneurs should be aiming for. He hates having to describe how businesses succeed, because he thinks it's too anti-inductive to reduce to a formula: Tolstoy opens Anna Karenina by observing "All happy families are alike; each unhappy family is unhappy in its own way." Business is the opposite. All happy companies are different: each one earns a monopoly by solving a unique problem. All failed companies are the same: they failed to escape competition. But he grudgingly describes four ways that a company can successfully reach monopolyhood:

At the beginning of every year, I make predictions. At the end of every year, I score them. So here are a hundred more for 2019. Rules: all predictions about what will be true on January 1, 2020. Any that involve polling will be settled by the top poll or average of polls on Real Clear Politics on that day. Most predictions about my personal life, or that refer to the personal lives of other people, have been redacted to protect their privacy. I'm using the full 0 – 100 range in making predictions this year, but they'll be flipped and judged as 50 – 100 in the rating stage, just like in previous years. I've tried to avoid doing specific research or looking at prediction markets when I made these, though some of them I already knew what the markets said. Feel free to get in a big fight over whether 50% predictions are meaningful. US 1. Donald Trump remains President: 90% 2. Donald Trump is impeached by the House: 40% 3. Kamala Harris leads the Democratic field: 20% 4. Bernie Sanders leads the Democratic field: 20% 5. Joe Biden leads the Democratic field: 20% 6. Beto O'Rourke leads the Democratic field: 20% 7. Trump is still leading in prediction markets to be Republican nominee: 70% 8. Polls show more people support the leading Democrat than the leading Republican: 80% 9. Trump's approval rating below 50: 90% 10. Trump's approval rating below 40: 50% 11. Current government shutdown ends before Feb 1: 40% 12. Current government shutdown ends before Mar 1: 80% 13. Current government shutdown ends before Apr 1: 95% 14. Trump gets at least half the wall funding he wants from current shutdown: 20% 15. Ginsberg still alive: 50%

Lots of people have asked me to recommend them a psychiatrist or therapist. I've done a terrible job responding: it's a conflict of interest to recommend my own group, and I don't know many people outside of it. So now I've put together a list (by which I mostly mean blatantly copied a similar list made by fellow community member Anisha M) of mental health professionals whom members of the rationalist community have had good experiences with. So far it's short and mostly limited to the Bay Area. You can find it at the "Psychiat-List" button on the top of the blog, or at this link. My hope is to crowd-source additional recommendations to expand the list to more providers and cities. Please let me know, either on this post or on the comments to the list itself, if you have any extra recommendations to add – especially if you're in a city likely to have many other SSC readers. Please also let me know if you've had any positive or negative experiences with people already on the list, so I can change their status accordingly.

At the beginning of every year, I make predictions. At the end of every year, I score them. Here are 2014, 2015, 2016, and 2017. And here are the predictions I made for 2018. Strikethrough'd are false. Intact are true. Italicized are getting thrown out because I can't decide if they're true or not. Please don't complain that 50% predictions don't mean anything; I know this is true but there are some things I'm genuinely 50-50 unsure of. US: 1. Donald Trump remains president at end of year: 95% 2. Democrats take control of the House in midterms: 80% 3. Democrats take control of the Senate in midterms: 50% 4. Mueller's investigation gets cancelled (eg Trump fires him): 50% 5. Mueller does not indict Trump: 70% 6. PredictIt shows Bernie Sanders having highest chance to be Dem nominee at end of year: 60% 7. PredictIt shows Donald Trump having highest chance to be GOP nominee at end of year: 95% 8. [This was missing in original] 9. Some sort of major immigration reform legislation gets passed: 70% 10. No major health-care reform legislation gets passed: 95% 11. No large-scale deportation of Dreamers: 90% 12. US government shuts down again sometime in 2018: 50% 13. Trump's approval rating lower than 50% at end of year: 90% 14. …lower than 40%: 50% 15. GLAAD poll suggesting that LGBQ acceptance is down will mostly not be borne out by further research: 80%