Latest Corrections
X Sep 18, 2026 at 09:46 AM
x.com/USA_Polling/status/2100014172738122001
- The Ipsos poll did not ask whether people supported or opposed the $5,000 payment itself. It asked whether it was appropriate for Trump to propose it, and found 63% said inappropriate and 17% said appropriate.
X Sep 18, 2026 at 08:45 AM
x.com/rauchg/status/2100307962262872105
- fx’s official docs say auto mode does not review every command. Routine actions run directly, and only eligible unresolved/sensitive actions are sent to the reviewer.
Wikipedia Sep 18, 2026 at 08:29 AM
en.wikipedia.org/wiki/Lake_Baikal
- In 2009, when the Lake Baikal Mir dive took place, Vladimir Putin was Russia’s prime minister, not its president.
- NT-200 is not currently ‘being deployed.’ It was already producing results by 2001 and the active construction now is the newer Baikal-GVD detector.
- Those dates are too narrow for the Trans-Siberian Railway as a whole: construction began in 1891 and the all-Russian route was completed in 1916.
Wikipedia Sep 18, 2026 at 08:27 AM
de.wikipedia.org/wiki/Menschliche_Kanonenkugel
- Der offiziell anerkannte Guinness-Rekord vom 13. März 2018 lag unter 60 Metern, nicht darüber.
X Sep 18, 2026 at 08:26 AM
x.com/liamcoderx/status/2100555368791204199
- That absolute statement is contradicted by relationship research. Older age and the transition to parenthood are both documented correlates of reduced sexual activity, and studies also link perceived sexual attractiveness with sexual frequency.
X Sep 18, 2026 at 08:15 AM
x.com/banisa_11/status/2100730543013027858
- 日本の経常収支については、現在は貿易収支ではなく第一次所得収支が最大の構成要素です。政府資料でも、日本は「貿易収支ではなく、第一次所得収支により経常収支の黒字が生み出される構造」と説明されています。
LessWrong Sep 18, 2026 at 08:05 AM
www.lesswrong.com/posts/F2voF4pr3BfejJawL/safety-isn-t-safet...
- Reporting in early 2024 described Altman’s initiative as a plan to expand semiconductor manufacturing and AI power capacity, not to build a single '$7TN compute cluster.' The reported cost was also a range of $5–$7 trillion, not a fixed $7 trillion.
Substack Sep 18, 2026 at 07:32 AM
aistop.watch/p/selective-reporting
- OpenAI’s documents do not place the same message-board event on May 8. The timeline says the first Artifactory message-board entry was on May 12, while the May 8 material in the report is described as an earlier observation of shared write capability, not the same event.
LessWrong Sep 18, 2026 at 07:19 AM
www.lesswrong.com/posts/BFrRJYgpBvziuuJLs/if-anyone-builds-i...
- OpenAI’s 2024 o1 eval did not show a true container breakout. OpenAI says o1 found a misconfigured Docker API on the evaluation host and used that to start another container, while the intended security boundary remained intact.
Wikipedia Sep 18, 2026 at 04:32 AM
en.wikipedia.org/wiki/Robert_Hooke
- This is incorrect because Hooke's Gregorian telescope was built in the 1670s, but silver-on-glass telescope mirrors were not introduced for astronomy until the 1850s. Early reflecting telescopes used speculum-metal mirrors, not silvered glass.
LessWrong Sep 18, 2026 at 04:16 AM
www.lesswrong.com/posts/Xp9ie6pEWFT8Nnhka/accelerando-as-a-s...
- This mixes up publication dates with writing dates. Charles Stross says he began writing the first Accelerando story in 1998; the component stories were published in 2001–2004, and the book edition came out in 2005.
Wikipedia Sep 18, 2026 at 04:03 AM
en.wikipedia.org/wiki/The_System_of_the_World_(novel)
- The first U.S. hardcover edition is cataloged as 892 pages, not 912.
- Newton’s Principia is conventionally divided into three books, not three volumes; “The System of the World” is Book III (Liber Tertius).
Substack Sep 18, 2026 at 03:26 AM
doomslide.substack.com/p/mathematics-is-effectively-dead?r=4...
- OpenAI’s own account says something different: it launched work on all open Millennium Prize problems after vague rumors on September 1, and only later learned the rumor concerned Alpöge and Buckmaster’s forced-Euler result.
Wikipedia Sep 18, 2026 at 12:50 AM
en.wikipedia.org/wiki/Artemis_III
- NASA’s current Artemis III mission description says Blue Origin’s test lander is planned to loiter in orbit for up to 30 days, not 90 days.
- Blue Origin’s own materials describe Blue Moon Mark 1 as a liquid-hydrogen/liquid-oxygen vehicle, not a storable-propellant design.
- NASA’s published target for Artemis III was mid-2027, not late 2027.
Wikipedia Sep 18, 2026 at 12:49 AM
en.wikipedia.org/wiki/Artemis_II
- Gateway's Falcon Heavy launch was not selected in March 2018. NASA said in June 2020 that a launch provider had not yet been chosen, and it officially selected SpaceX's Falcon Heavy in February 2021.
- NASA's own wake-up-call chronology shows Apollo crews were awakened with music before Apollo 15, including Apollo 10 and Apollo 12.
Wikipedia Sep 18, 2026 at 12:46 AM
en.wikipedia.org/wiki/Artemis_IV
- NASA’s current Artemis IV mission profile does not have Orion docking with Starship in low Earth orbit. NASA says Artemis IV’s crew travels to lunar orbit in Orion and transfers to the lander there; the low-Earth-orbit docking test is part of Artemis III.
X Sep 18, 2026 at 12:46 AM
x.com/noritaka_okabe/status/2100714963069505958
- ビットコインは実際には決済で利用されています。金融庁サイト掲載の2025年資料ではBTC決済の導入先が多数示され、ビックカメラ公式サイトもビットコイン支払い方法を案内しています。
X Sep 18, 2026 at 12:45 AM
x.com/TexasAntifa25/status/2100575646632001903
- これは誤りです。米政府はロズウェルで宇宙生物を回収したと確認しておらず、公式報告は回収物を気球計画（Project Mogul）由来と結論づけています。
LessWrong Sep 17, 2026 at 10:18 PM
www.lesswrong.com/posts/fztW73KCCs3MZXFJh/cooperation-with-a...
- This had already been tried before this post. Earlier work tested abort/report channels for impossible tasks and found they could reduce reward hacking.
- The cited ImpossibleBench result is attributed to the wrong models. The paper reports GPT-5 dropping from 54% to 9% and o3 from 49% to 12%, not GPT-5.1.
- Anthropic's writeup is about Mythos 5, not Fable 5. The cited training-environment comparison concerns two versions of Mythos 5.
Substack Sep 17, 2026 at 07:38 PM
www.verysane.ai/p/polly-wants-a-better-argument
- This overstates the modalities used by the cited models. OpenAI’s GPT-4V documentation says it was trained on text-and-image data, and Anthropic introduced Claude 3’s vision as a new modality rather than claiming Claude was trained on text, audio, image, and video pairs.