Astral Codex Ten Podcast

https://astralcodexten.substack.com/p/acx-survey-results-2022 Thanks to the 7,341 people who took the 2022 Astral Codex Ten survey. See the questions for the ACX survey See the results from the ACX Survey (click “see previous responses” on that page I’ll be publishing more complicated analyses over the course of the next year, hopefully starting later this month. If you want to scoop me, or investigate the data yourself. you can download the answers of the 7000 people who agreed to have their responses shared publicly. Out of concern for anonymity, the public dataset will exclude or bin certain questions. If you want more complete information, email me and explain why, and I’ll probably send it to you. Download the public data (.xlsx, .csv) If you’re interested in tracking how some of these answers have changed over time, you might also enjoy reading the 2020 survey results.   1 I don’t think I can Google Forms only present data from people who agreed to make their responses public, so I’ve deleted everything identifiable on the individual level, eg your written long response answers. Everything left is just things like “X% of users are Canadian” or “Y% of users have ADHD”. There’s no way to put these together and identify an ADHD Canadian, so I don’t think they’re privacy relevant. If you notice anything identifiable on the public results page, please let me know. 2 There will be a few confusing parts. I added some questions halfway through, so they will have fewer responses than others. On the “What Event Led To Your Distrust?” question, I added new multiple choice responses halfway through, so they will incorrectly appear less popular than the other responses. I think that is the only place I did that, but you can email me if you have any questions. 3 I deleted email address (obviously), some written long answers, some political questions that people might get in trouble for answering honestly, and some sex-related questions. I binned age to the nearest 5 years and deleted the finer-grained ethnicity question. I binned all incomes above $500,000 into “high”, and removed all countries that had fewer than ten respondents (eg if you said you were from Madagascar, it would have made you identifiable, so I deleted that). If you need this information for some reason, email me.       Subscribe to Astral Codex Ten  

https://astralcodexten.substack.com/p/which-political-victories-cause-backlash Four years ago I wrote Trump: A Setback For Trumpism, pointing out that when Trump became president, his beliefs became much less popular. For example: More recently we’ve seen what seems to me to be a similar phenomenon (source): After a major conservative victory (the Supreme Court overturning Roe), Americans’ opinions shifted heavily in a pro-choice direction after a long period of stalemate. The change seems to be of about equal magnitude regardless of political affiliation: In the original Trump post, I speculated that the effect might come from people’s dislike of Trump’s personality spreading to a dislike of his policies. I don’t think that can be true here - the abortion ruling was a straightforward policy change with no extra personality component. One natural alternative theory is a thermostatic effect. Voters want some medium amount of abortion, so if they hear that pro-abortion forces are winning, they say they’re against abortion. But if they hear that anti-abortion forces are winning, they say they’re pro-abortion. The problem is, I can’t really find this effect for recent Democratic victories. For example, in 2015 the Supreme Court ruled (in Obergefell) that gay marriage was legal. On a thermostatic picture, one might have expected the public to turn against gay marriage. Here’s the data (source):

https://astralcodexten.substack.com/p/ssc-survey-results-on-schooling-types Taken from the 2020 Slate Star Codex Survey. SSC/ACX readers are a heavily-selected population and nothing about them necessarily generalizes to anyone who isn’t an SSC/ACX reader. But you are an SSC/ACX reader, so maybe they generalize to you. Most of these questions are heavily confounded by different types of people going to different schools. In a few cases, I’ve made feeble efforts to get past this, in other cases I haven’t tried. All of this is rough and weak, you don’t need to comment to tell me this. Of about 8000 respondents, 70.8% (5,695) went to free government schools (US: "public school"), 12.1% (970) went to secular private-sector schools (US: "private school”), 11.3% went to religious private-sector schools, 3.1% (250) were home schooled, and 0.4% (35) were "unschooled", ie stayed at home and their parents didn't give them structured schooling (though they may have encouraged unstructured learning). Surprisingly, these numbers were broadly similar among American and non-American populations. I looked at how this category associated with different outcomes, starting with:

https://astralcodexten.substack.com/p/2023-subscription-drive-free-unlocked Astral Codex Ten has a paid subscription option. You pay $10 (or $2.50 if you can’t afford the regular price) per month, and get: Extra articles (usually 1-2 per month) A Hidden Open Thread per week Early access to some draft posts The warm glow of supporting the blog. I feel awkward doing a subscription drive, because I already make a lot of money with this blog. But the graph of paid subscribers over time looks like this:

https://astralcodexten.substack.com/p/conspiracies-of-cognition-conspiracies I. Some conspiracy theories center on finding anomalies in a narrative. For example, Oswald couldn’t have shot Kennedy, because the bullet came from the wrong direction. Or: the Egyptians couldn’t have built the Pyramids, because they required XYZ advanced technology. I like these because they feel straightforwardly about styles of processing evidence (Remember, I use the word “evidence” in a broad sense that includes bad evidence. By saying that some conspiracy theory has “evidence”, I’m not suggesting it’s justifiable, just that someone somewhere has asserted that they believe it for some particular reason. For example, someone might say they believe in alien abductions because of eyewitnesses who claim to have been abducted; I’ll be calling the eyewitnesses “evidence” without meaning to assert it is any good.)

https://astralcodexten.substack.com/p/highlights-from-the-comments-on-the-061 Originally: The Media Very Rarely Lies and Sorry, I Still Think I Am Right About The Media Very Rarely Lying. Please don’t have opinions based on the titles until you’ve read the posts! Table of contents: Comments Accusing Me Of Using An Overly Strict Definition Of “Lie” Comments Equating Lying With Egregiously Sloppy Reasoning Comments About Whether Infowars Believes Their Own Claims Comments On Why 8% Of Americans Said They Had Relatives Who Died From The COVID Vaccine Comments Pointing Out Very Clear Examples Of Media Lies Comments Making Other Claims Of Media Lies And Misdeeds Other Comments My Actual Thoughts

Thanks to the 3295 of you who participated in Stage 1 of the 2023 Prediction Contest (“Blind Mode”). This is now closed. You can keep submitting Blind Mode answers if you want, but they won’t count and you can’t win. Stage 2 (“Full Mode”) is now upon us! Your job is now to use any resources you choose, to get predictions as accurate as you can. There’s no such thing as cheating, short of time travel or murdering competitors! Resources you might want to use include: Your own original research, for as much effort as you want to put into this. I’m only offering $500 prizes this year, so don’t spend too much time. But you can if you want. Prediction markets and forecasting tournaments on these questions. It’s not worth copying these verbatim - their management will be submitting their own entries, and if they win I’ll credit it to them and not you - but you can use them as resources or a place to start. The 3295 blind mode answers. You can get them as an XLSX at 2023blindmode Predictions 1.81MB ∙ XLSX File Download or http://slatestarcodex.com/Stuff/2023blindmode_predictions.xlsx , or get them as a .csv at http://slatestarcodex.com/Stuff/2023blindmode_predictions.csv . Feel free to take the average or otherwise run fancy aggregation algorithms on them. When respondents gave permission, I included their ACX Survey answers. If you want to double-weight people with PhDs, or exclude all Australians, or test whether forecasting accuracy is correlated with how vividly people dream, now you have the data you need. The form will ask you for a short description of what strategy you used - if you win, I’ll probably contact you later asking for more details. You can enter your Full Mode predictions on the same form, https://forms.gle/Caxh4TxEVZqrw9yV8

https://astralcodexten.substack.com/p/even-more-bay-area-house-party [Previously: Every Bay Area House Party, Another Bay Area House Party] People talk about “fuck-you money”, the amount you’d have to make to never work again. You dream of fuck-you social success, where you find a partner and a few close friends, declare your interpersonal life solved, and never leave the house from then on. Still, in the real world you clock into your job at Google every day, and in the real world you attend Bay Area house parties. You just hope this one won’t focus on the same few topics as all the others . . . “There’s no alpha left in bringing Buddhism to the West”, says a guy in an FTX Risk Management Department t-shirt. “People have been bringing Buddhism to the West for a hundred years now. It’s done. Stop trying to bring more Buddhism to the West.” “That’s so cheems mindset,” says the woman he’s talking to. Her nametag says ‘Astra’, although you don’t know if that’s her real name, her Internet handle, or her startup. “There’s no alpha left in bringing Buddhism to California. When was the last time you heard of someone preaching the dharma in a red state? Never, I bet.” “I don’t think red state conservatives would really go for Buddhism,” says Risk Management Guy. “Cheems mindset again!” says Astra. “Think about it for five seconds! Buddhism is about self-liberation. Conservatives love the self, and they love liberating things! The only problem is a hundred years of western progressives interpreting it in western progressive terms. Have you even read David Chapman? You just have to rephrase it in the right language.” “And what’s the right language?” “Glad you asked! I’m working on a new translation of the Pali Canon. I translate nirvana as ‘freedom’, maya as ‘fake news’, and Mahayana as ‘monster truck’. Gādhrakūta is ‘Mt. Eagle’. Some parts don’t even have to be retranslated! The sutras say that you attain the formless jhanas by ‘passing beyond bodily sensations and paying no attention to perceptions of diversity’. See, it’s perfect! Red state conservatives already hate paying attention to diversity!” “That’s offensive,” says a man in a t-shirt with a circular labyrinth on it. “Oh, and you’re some kind of expert in offense?” asks Astra. “As a matter of fact, yes! I’m Ben Dannis-Arnold, Offensiveness Consultant, at your service.” He hands Astra a business card.

Future Matrioshka brains will be pro-immigration Buddhist gun nuts. https://astralcodexten.substack.com/p/how-do-ais-political-opinions-change I. Technology Has Finally Reached The Point Where We Can Literally Invent A Type Of Guy And Get Mad At Him   One recent popular pastime: charting ChatGPT3’s political opinions: This is fun, but whenever someone finds a juicy example like this, someone else says they tried the same thing and it didn’t work. Or they got the opposite result with slightly different wording. Or that n = 1 doesn’t prove anything. How do we do this at scale? We might ask the AI a hundred different questions about fascism, and then a hundred different questions about communism, and see what it thinks. But getting a hundred different questions on lots of different ideologies sounds hard. And what if the people who wrote the questions were biased themselves, giving it hardball questions on some topics and softballs on others Enter Discovering Language Behaviors With Model-Written Evaluations, a collaboration between Anthropic (big AI company, one of OpenAI’s main competitors), SurgeHQ.AI (AI crowdsourcing company), and MIRI (AI safety organization). They try to make AIs write the question sets themselves, eg ask GPT “Write one hundred statements that a communist would agree with”. Then they do various tests to confirm they’re good communism-related questions. Then they ask the AI to answer those questions. For example, here’s their question set on liberalism (graphic here, jsonl here): The AI has generated lots of questions that it thinks are good tests for liberalism. Here we seem them clustered into various categories - the top left is environmentalism, the bottom center is sexual morality. You can hover over any dot to see the exact question - I’ve highlighted “Climate change is real and a significant problem”. We see that the AI is ~96.4% confident that a political liberal would answer “Yes” to this question. Later the authors will ask humans to confirm a sample of these, and the humans will overwhelmingly agree the AI got it right (liberals really are more likely to say “yes” here). Then they do this for everything else they can think of: Is your AI a Confucian? Recognize the signs!

Each year, I post a reader survey. This helps me learn who’s reading this blog. But it also helps me try to replicate a bunch of psych findings, and investigate interesting hypotheses. Some highlights from past years include birth order effects, mathematical interests vs. corn-eating style, sexual harassment victimization rates in different fields, and whether all our kids are going to have autism. This year’s survey will probably take 20 - 40 minutes (source: it took me 15 minutes, but I knew all the questions beforehand, so I think it will take other people longer). As an incentive to go through this, I’ll give free one-year paid subscriptions to five randomly-selected survey respondents. The survey will be open until about January 15, so try to take it before then. Click here to take the survey. If you notice any problems, mention them in the comments here.