TL;DR
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
Astral Codex Ten published its October 2026 links roundup, covering topics including Jeremy Bentham’s cookbook, AI-assisted translation and art, marriage statistics, and AI forecasting. The post’s author says the links have not all been independently verified and that readers often identify errors in these roundups.
Astral Codex Ten has published its October 2026 links roundup, a collection of commentary on subjects ranging from historical recipes to AI translation and forecasting. The author cautions that the links have not all been independently verified, and says readers typically identify errors or misleading claims in a few items in each roundup.
The roundup presents a set of short items rather than a single investigation. It points readers to recipes from a cookbook associated with Jeremy Bentham, described in the post as a utilitarian cookbook intended for use at his Panopticon prison. The author says its recipes use inexpensive ingredients and include meat, despite Bentham’s association with animal rights. The post also links to a riddle about a chemistry paper containing the phrase “1 ml of the mass killing of an ethnic group,” but the supplied text does not explain the answer.
Several entries focus on AI and creative work. The author praises an AI-assisted image inspired by Hieronymus Bosch while questioning whether it qualifies as AI art because it draws on a human artist’s work. Another item describes an attempt to render the opening of the Chinese novel Romance of the Three Kingdoms in rhyming pentameter. The author says the result looked impressive but notes that they do not speak Chinese, limiting their ability to evaluate the translation.
The roundup also discusses an AI bot’s performance in the Metaculus Cup forecasting competition. Citing a September result and coverage by Metaculus and The Economist, it says a bot named Laertes placed first. The post attributes its development to Jeffrey Liang and reports his estimate that he spent fewer than 150 hours and a few thousand dollars on computing and data. It also notes that bots from Mantic, FutureSearch and Preseen performed well, and links to a report that Mantic raised $25 million in seed funding.
AI Moves Into Forecasting
The reported competition result offers a concrete example of AI systems taking part in a task usually associated with human judgment: predicting future events. The roundup’s author connects the result to an earlier forecast of AI-human forecasting parity by the end of 2026. One competition win is not enough to establish parity across forecasting tasks, but it gives readers a specific result to assess against that prediction.
The account also draws attention to the resources reportedly used to build Laertes. If Liang’s estimate is accurate, the system was developed with less than 150 hours of work and a few thousand dollars in computing and data costs. The post does not provide enough information to compare that figure with other systems or independently assess the bot’s performance, but it raises practical questions about how accessible competitive forecasting tools may become.
The roundup’s other AI examples point to broader questions about evaluation. A fluent poetic translation can appear successful to a reader who cannot check the source language, while an image inspired by a known painter raises questions about authorship and influence. These examples show why impressive output is not, on its own, proof of accuracy or originality.
Popular right now
De'Longhi 15-Bar Pump Espresso Machine with Milk Frother & Temperature Control
KitchenAid Artisan 5-Qt Tilt-Head Stand Mixer – Dried Rose (KSM150PS)
Sony BRAVIA 5 85-Inch Mini LED 4K HDR Smart TV (K-85XR50)
As an affiliate, we earn on qualifying purchases.
A Roundup With Caveats
Astral Codex Ten’s links posts gather outside writing, research and online discussion, with brief commentary from the author. The October edition includes both links and the author’s own assessments; those assessments should not be confused with independent verification of every linked item.
The post is unusually explicit about that limitation. Its author says commenters typically find evidence that around two or three links in each roundup are wrong or misleading, and that corrections are made when found. That is the author’s account of the review process, not a verified error rate for this specific edition. The supplied material does not identify which October links, if any, have already been corrected.
Other entries cover claims about marriage satisfaction, the history of modern polyamory, and an older Claude model completing text prompted to resemble an internal company email. On the marriage item, the author disputes how a linked writer interpreted survey categories and says their reading of the General Social Survey indicates 61% described their marriage as “very happy”. The post says the author checked that particular figure but had not confirmed other alleged errors.
“I haven’t independently verified each link.”
— Astral Codex Ten author, on the roundup’s reliability
Claims Still Need Checking
The roundup does not establish that every linked report is accurate, and the author expressly says individual links were not independently verified. The source material also does not include the answer to the chemistry-paper riddle, the full underlying competition results, or independent documentation of Laertes’s development costs and methods.
Several judgments are presented as the author’s own view. These include the assessment of the AI-generated art, the claim that AI may be especially strong at poetic translation, and the interpretation of the marriage survey. The author says they checked the “not too happy” issue but had not confirmed other errors they believed Claude identified. The translated verse’s quality cannot be assessed from the author’s account alone because they do not read Chinese.
The source ends partway through an item beginning “Fee fi fo fum,” so its claim and any supporting evidence are unavailable in the supplied material. It is not possible to report what that item says or whether the roundup received later corrections.
Corrections and Results to Watch
The author says they correct links as readers flag problems and plan to highlight important corrections later. Readers should check the original linked material and any subsequent notes before treating a disputed claim as settled. The source supplied here does not identify a schedule for those corrections.
For the forecasting item, later results from Metaculus and other forecasting competitions would help show whether Laertes’s first-place finish reflects sustained performance or a result specific to one contest. The roundup also points readers toward further coverage of the other high-performing bots and Mantic’s reported funding, but it does not provide follow-up dates or additional performance data.
Key Questions
What is the October 2026 links post?
It is an Astral Codex Ten roundup of links and short commentary on topics including AI, history, relationships and public debate. It is not a single reported investigation.
Did an AI win a forecasting competition?
The roundup reports that Laertes placed first in the Metaculus Cup in September 2026, citing Metaculus and The Economist. The supplied text does not include the full results or independent performance analysis.
How reliable are the claims in the roundup?
The author says they have not independently verified every link and that readers often identify errors or misleading material. Treat the roundup as a guide to sources and commentary, not confirmation of every linked claim.
What does the post say about marriage survey figures?
The author disputes a linked writer’s interpretation of marriage-satisfaction categories and says their reading of the General Social Survey found that 61% of respondents described their marriage as “very happy.” The post says this specific point was checked, while other alleged mistakes remained unconfirmed.
Source: astralcodexten.com
