Astral Codex Ten Podcast

Bret Devereaux writes here about the oddities of the academic job market. His piece is comprehensive, and you should read it, but short version: professors are split into tenure-track (30%, good pay and benefits) and adjunct (50%, bad pay and benefits). Another 20% are “teaching-track”, somewhere in between. Everyone wants a tenure-track job. But colleges hiring new tenure-track faculty prefer newly-minted PhDs to even veteran teaching-trackers or adjuncts. And even if they do hire a veteran teaching-tracker or adjunct, it’s practically never one of their own. If a teaching-tracker or adjunct makes a breakthrough, they apply for a tenure-track job somewhere else. Devereaux describes this as “a hiring system where experience manifestly hurts applicants” and displays this graph: https://astralcodexten.substack.com/p/why-is-the-academic-job-market-so  

Adam Mastroianni has a great review of Memories Of My Life, the autobiography of Francis Galton. Mastroianni centers his piece around the question: how could a brilliant scientist like Galton be so devoted to an evil idea like eugenics? This sparked the usual eugenics discussion. In case you haven’t heard it before: https://astralcodexten.substack.com/p/galton-ehrlich-buck

Table of Contents 1. Summary Of Best Comments And Overall Updates 2. Comments Proposing Explanations Based On Response Patterns 3. Comments Proposing Explanations Based On Biology 4. Comments By Jim Coyne 5. Comments Expressing Concerns About The Dangers Of Calling Things Psychosomatic 6. Other Comments https://astralcodexten.substack.com/p/highlights-from-the-comments-on-long Original post: Replication Attempt - Bisexuality And Long COVID

https://astralcodexten.substack.com/p/highlights-from-the-comments-on-housing Table Of Contents: 1. Comments About Whether Density Causes Desirability 2. Comments About Jobs And Amenities (And Not Density Per Se) Producing Desirability 3. Comments About Chinese Ghost Cities 4. Comments Accusing Me Of Not Considering Tokyo, Even Though I Included A Section In The Post On Why I Didn’t Think Tokyo Was Relevant 5. Comments Accusing Me Of Not Understanding Economics 6. Comments By Famous People Who Potentially Have Good Opinions 7. My Final Thoughts + Poll

https://astralcodexten.substack.com/p/constitutional-ai-rlhf-on-steroids A Machine Alignment Monday post, 5/8/23 What Is Constitutional AI? AIs like GPT-4 go through several different 1 types of training. First, they train on giant text corpuses in order to work at all. Later, they go through a process called “reinforcement learning through human feedback” (RLHF) which trains them to be “nice”. RLHF is why they (usually) won’t make up fake answers to your questions, tell you how to make a bomb, or rank all human races from best to worst. RLHF is hard. The usual method is to make human crowdworkers rate thousands of AI responses as good or bad, then train the AI towards the good answers and away from the bad answers. But having thousands of crowdworkers rate thousands of answers is expensive and time-consuming. And it puts the AI’s ethics in the hands of random crowdworkers. Companies train these crowdworkers in what responses they want, but they’re limited by the crowdworkers’ ability to follow their rules.f

https://astralcodexten.substack.com/p/raise-your-threshold-for-accusing I. Many comments in yesterday’s post about self-identified bisexuals getting long COVID centered on a concern that self-identified bisexuals don’t really date both sexes, and are just claiming to be bi because it’s trendy. Bisexuals themselves hate this and have written many articles and papers about why you shouldn’t say it (1, 2, 3). But I especially appreciated a discussion in the comments between Nom de Flume, Ryan W, and others, giving a great statistical explanation for why it’s tempting to believe this, but why it isn’t true. Suppose someone (let’s say a woman) has exactly equal sexual attraction to both men and women.

https://astralcodexten.substack.com/p/replication-attempt-bisexuality-and I learned from Pirate Wires that CDC data show bisexuals were about 50% more likely than heterosexuals to report long COVID. Is this just because more women than men are bisexual, and more women than men get long COVID? Not exactly; in the data they cite, women (regardless of sexuality) have an 18% rate, and bisexuals (regardless of gender) have a 22% rate. (aren’t all these numbers really high? You can find almost any number depending on how you ask the question; questions along the lines of “have you had any persistent symptoms including fatigue, brain fog, shortness of breath, changes to taste/smell, etc, etc, etc, since having COVID?” tend to produce numbers from 20-30%; most will say this symptoms are mild and don’t affect their functioning very much) This seemed weird enough that I wanted to try replicating it with the ACX survey data (read more about the ACX survey here).

https://astralcodexten.substack.com/p/change-my-mind-density-increases Matt Yglesias tries to debunk the claim that building more houses raises local house prices. He presents several studies showing that, at least on the marginal street-by-street level, this isn’t true. I’m nervous disagreeing with him, and his studies seem good. But I find looking for tiny effects on the margin less convincing than looking for gigantic effects at the tails. When you do that, he has to be wrong, right?

https://astralcodexten.substack.com/p/highlights-from-the-comments-on-nerds Table of contents: 1: Comments By The Author Of The Original Post 2: Comments With Strong Opinions On The Definition Of Nerds, Geeks, Etc 3: Comments About Collecting 4: Comments Insisting That Sports Are Good 5: Comments About Enjoying Things Vs. Building Identities Around Them

https://astralcodexten.substack.com/p/mantic-monday-42423 Can AIs Predict The Future? By Which We Mean The Past? If we asked GPT-4 to play a prediction market, how would it do? Actual GPT-4 probably would just give us some boring boilerplate about how the future is uncertain and it’s irresponsible to speculate. But what if AI researchers took some other model that had been trained not to do that, and asked it? This would take years to test, as we waited for the events it predicted to happen. So instead, what if we took a model trained off text from some particular year (let’s say 2020) and asked it to predict forecasting questions about the period 2020 - 2023. Then we could check its results immediately! This is the basic idea behind Zou et al (2022), Forecasting Future World Events With Neural Networks. They create a dataset, Autocast, with 6000 questions from forecasting tournaments Metaculus, Good Judgment Project, and CSET Foretell. Then they ask their AI (a variant of GPT-2) to predict them, given news articles up to some date before the event happened. Here’s their result: