Astral Codex Ten

Thomas Reilly

7 min read

Thomas Reilly

Article

Thomas Reilly is a recurring person in the Astral Codex Ten archive, appearing 4 times across 4 issues between March 16, 2022 and February 27, 2025. The archive places it in contexts such as “Thomas Reilly has a new blog Rational Psychiatry where he’s written up some more info”; “Thomas Reilly (author of Rational Psychiatry ) writes”; “Thomas Reilly says the study was underpowered”. It most often appears alongside Aella, China, Rational Psychiatry.

Metadata

  • Category: People
  • Mention count: 4
  • Issue count: 4
  • First seen: March 16, 2022
  • Last seen: February 27, 2025

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Highlights From The Comments On Zulresso
March 16, 2022 · Original source
— Thomas Reilly has a new blog Rational Psychiatry where he’s written up some more info on premenstrual dysphoria and progesterone. For example:
Inline links: Rational Psychiatry, some more info
Highlights From The Comments On Geography Of Madness
February 27, 2023 · Original source
Thomas Reilly (author of Rational Psychiatry) writes:
Inline links: Rational Psychiatry, writes
Open Thread 303
November 20, 2023 · Original source
2: And some good comments on the ketamine post. Thomas Reilly says the study was underpowered. Awais Aftab compares to a recent very positive trial of ketamine vs. ECT. Eremolalos on a meta-analysis. Nate Praschan argues that anaesthetics might block ketamine directly.
Inline links: says the study was underpowered, a recent very positive trial of ketamine vs. ECT, on a meta-analysis, anaesthetics might block ketamine directly
Links For February 2025
February 27, 2025 · Original source
11: Intrinsic Perspective wants a law saying AI-generated text must be watermarked. I was most interested the article’s claim that there is now “semantic watermarking” - watermarking which operates on the level of ideas, and can’t be defeated by rephrasing an AI-generated text in your own words. I have skimmed the paper explaining this and think I vaguely understand what’s going on, but it still boggles me that this is possible. 12: Aella: How OnlyFans Took Over The World. There have been camgirl sites since forever. How did OnlyFans leap over all of its predecessors and achieve an unprecedented level of success? Aella discusses many factors, but one stands out: traditional camsites advertised the site as a whole, and then once you got to the site you chose which model you wanted to see. OnlyFans encourages models to advertise themselves - often on their own social media accounts, sometimes via scams - which “unlocks human creativity” on the problem of bringing new eyeballs to a porn site. 13: Nate Silver has 113 predictions for Trump’s second term. I’d be interested to see whether making each of these predictions 10% less confident (to account for possible gameboard-overturning AI) ends up beating Nate. 14: Sarah Constantin: What’s Behind The SynBio Bust? Three of the most promising synthetic biology companies - Gingko, Zymergen, and Amyris - all crashed between 2021 and 2023. Why? Producing chemicals in traditional factories is orders of magnitude more efficient than synthesizing them via microbes (except for the sort of large biomolecules that can’t be produced in factories). These companies had brilliant employees and cool tech, but no clear plan to get around this handicap, and used up their runway before they could figure one out. They also focused too hard on designing the microbes, and were too willing to outsource the actual manufacturing to other people without being sufficiently paranoid that those other people were doing quality control. 15: One of the more exciting psychiatric results (which I blogged about a long time ago) was the apparent finding that omega-3 supplementation could prevent high-risk people from having first break schizophrenia. A new RCT says this doesn’t replicate and cites two other recent trials showing it didn’t replicate. There’s also a new meta-analysis which says actually it does replicate, but usually failing a big RCT is a bad sign and I’m pretty skeptical. Thanks to Isaak F for the links. {ETA: Thomas Reilly says: “Although I don't believe omega-3 supplementation has any benefit in psychosis, I also don't think this new trial should shift your opinion much, given the total sample size was n=135 and the total number of transitions to psychosis was n=8.”] 16: Claim that predictions of global warming magnitude are gradually going down thanks to successful pledges/action: Source is CipherNews (h/t Stefan Schubert) apparently citing Climate Action Tracker, but I get the impression that this is just some people eyeballing the size of pledges and not any more sophisticated forecasting. I don’t know how to square this with the claims that such and such a thing (summer temperature, sea ice, etc) is much worse than anyone expected. 17: I don’t know anything about the Lucy Letby case, but all of my smart friends who have been right about this kind of thing before say she’s innocent. 18: A reader asks House of Strauss (edgy sports Substack) whether the vibe shift away from political correctness threatens the edgy Substack business model - as the power of orthodoxy declines, can you still get rich and famous as a brave anti-orthodoxy critic? His answer: nothing that can happen from here is as bad as the Twitter/X link deboost (which made attracting attention harder for everyone). I mostly agree: I think discoverability has suffered, people who are already famous will be able to stay famous without too much extra effort, and everyone else will have to explore new options. 19: Spectator: Could AI Lead To A Revival Of Decorative Beauty? Profiles Not Quite Past, a startup using AI and fancy printing to make customized Delft tiles. It’s a good idea and the tiles are very pretty, but the tiles are sort of a best possible case (a pretty, traditional object that can have a customized 2D image and be mass-printed). I think most forms of lost decorative beauty aren’t bottlenecked by ability to generate 2D images of the type image models are good at, and so will have to wait. 20: Some friends including Kelsey Piper wrote an emergency PEPFAR Report, collecting evidence for why PEPFAR is good/effective/important and deserves to be kept. Some key points: PEPFAR has saved between 7.5 and 30 million lives, at a cost between $1,500 and $10,000 per life saved. The US government is willing to spend at least a thousand times this much to save an American life.
Inline links: wants a law saying AI-generated text must be watermarked, skimmed the paper explaining this, How OnlyFans Took Over The World, 113 predictions for Trump’s second term, What’s Behind The SynBio Bust?, this doesn’t replicate, actually it does replicate, says, https://substackcdn.com/image/fetch/$s_!S0yA!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb75ee3e3-3c0d-48b7-aa98-88f5e33dbf55_748x447.png, CipherNews, Stefan Schubert, Climate Action Tracker, the Lucy Letby case, His answer, Could AI Lead To A Revival Of Decorative Beauty?, Not Quite Past, PEPFAR Report