Latest Corrections
LessWrong Mar 17, 2026 at 12:22 AM
www.lesswrong.com/posts/4ftQmSDujzgiEujwA/epstein-and-my-wor...
- This is contradicted by the publicly reported findings in John Barnett’s case: Charleston police concluded he died by suicide, and even his family’s later wrongful-death lawsuit described Boeing’s conduct as leading to a foreseeable suicide, not an assassination.
LessWrong Mar 17, 2026 at 12:03 AM
www.lesswrong.com/posts/aaHDA4X6cTzFrvuSX/harms-and-possibil...
- The linked article is about a school case in Ireland, not the UK. It references Irish institutions including Gardaí and Tusla and uses Irish school terminology such as “fifth class.”
LessWrong Mar 17, 2026 at 12:01 AM
www.lesswrong.com/posts/JnKCaGcgZL4Rsep8m/schools-proliferat...
- This absolute claim is incorrect: by the late 20th century, psychotherapies such as cognitive therapy and interpersonal psychotherapy had already been tested in many controlled trials and shown efficacy for some conditions.
- This is too absolute. Controlled trials have found statistically significant differences between psychotherapies for some disorders—for example, CBT and prolonged exposure have outperformed supportive therapies in randomized trials.
Substack Mar 17, 2026 at 12:00 AM
minicities.org/p/how-miniature-cities-can-interact
- This absolute claim overstates how KidZania works. KidZania’s own materials describe citywide links between activities through a shared currency and bank, and links to the outside world through official brand partnerships.
LessWrong Mar 16, 2026 at 11:54 PM
www.lesswrong.com/posts/qzaKfDyQSezeLgFea/mini-munich-succee...
- KidZania’s own current pages place it in roughly 18–22 countries, not “around thirty countries.” The article appears to have confused countries with cities/facilities.
LessWrong Mar 16, 2026 at 11:13 PM
www.lesswrong.com/posts/ybYBCK9D7MZCcdArB/how-to-measure-any...
- This formula is wrong. Expected opportunity loss is probability × loss/regret for that segment, not probability × the segment’s midpoint value.
Wikipedia Mar 16, 2026 at 09:21 PM
en.wikipedia.org/wiki/Kate_Mara
- This is no longer to be announced. Apple lists Imperfect Women as a 2026 series with a March 18, 2026 debut date.
Wikipedia Mar 16, 2026 at 09:19 PM
en.wikipedia.org/wiki/House_of_Cards_(American_TV_series)
- This Emmy total is outdated. The Television Academy lists House of Cards with 56 Primetime Emmy nominations, not 33.
- The Australian Netflix launch date is wrong here. Netflix launched in Australia on March 24, 2015, not March 28.
- This is no longer correct. Kevin Spacey was charged again in the UK in 2022, so the 2019 case was not the last related criminal charge.
Wikipedia Mar 16, 2026 at 03:25 PM
en.wikipedia.org/wiki/ARC_Raiders
- Embark’s official patch notes date update 1.18.0 to March 3, 2026, not March 6. The official site also shows patch 1.19.0 published on March 10, 2026.
- Embark’s official Server Slam announcements say the event ran from October 17 to October 19, 2025, not through October 20.
Wikipedia Mar 16, 2026 at 03:20 PM
en.wikipedia.org/wiki/Valve_Corporation
- Steam was announced in 2002, not launched then. Valve’s own records say Steam was created in 2003 and the Steam client was released on September 12, 2003.
Wikipedia Mar 16, 2026 at 03:19 PM
en.wikipedia.org/wiki/Gabe_Newell
- The Heart of Racing was not created in 2020 by Gabe Newell and Yahn Bernier. Official team history traces it back to a 1997 fundraising campaign, and motorsport coverage shows the team had already raced in IMSA from 2014 to 2016 before returning in 2020.
- This is overstated. Forbes lists Gabe Newell at about $11 billion, but also lists NetEase founder William Ding—whose wealth comes from online games—at roughly $39 billion, making Newell clearly not the wealthiest person in the video game business.
LessWrong Mar 16, 2026 at 02:47 PM
www.lesswrong.com/posts/Y8c4kh6fjkWu7Auxp/monthly-roundup-40...
- This is incorrect: paper non-cEDH Magic events outside the Pro Tour and Arena do offer cash prizes. Wizards’ own 2026 Spotlight Series events have $50,000 prize pools, and Regional Championships also have cash prize pools.
Wikipedia Mar 16, 2026 at 12:36 PM
ja.wikipedia.org/wiki/%E7%B5%9E%E3%82%8A%E6%9F%93%E3%82%81
- WSNは『2年後』ではなく、1992年の第1回国際絞り会議の開催中に組織化されたと公式サイトが説明しています。
- 第1回国際絞り会議の日程は公式のWSN日本版では1992年11月21日〜23日で、記事の『22日〜23日』は1日短くなっています。
- 第4回国際絞り会議の日程は公式のWSN日本版では2002年11月21日〜24日で、記事の『10月21日〜24日』は月が違います。
Wikipedia Mar 16, 2026 at 12:34 PM
ja.wikipedia.org/wiki/%E3%83%A2%E3%82%B6%E3%82%A4%E3%82%AF
- ここは地名の取り違えです。アイギナ（Aegina）はアテネ近くの島で、マケドニア王国の王宮都市はアイガイ（Aigai）やペラであり、「マケドニア王国の宮殿のあったアイギナ島」という記述は誤りです。
Wikipedia Mar 16, 2026 at 12:31 PM
zh.wikipedia.org/wiki/%E6%89%8E%E6%9F%93
- 国家级非遗名录中，白族扎染技艺及其扩展项目自贡扎染技艺的序号都是376，不是377。
Wikipedia Mar 16, 2026 at 01:06 AM
en.wikipedia.org/wiki/Polymarket
- This end date is wrong. Reuters reported that the CFTC approved Polymarket’s U.S. relaunch on September 3, 2025, and a CFTC filing shows Polymarket US was designated in July 2025.
- This overstates the deal as a completed $2 billion investment. Contemporary reports and ICE’s announcement said ICE agreed to invest up to $2 billion, with Polymarket valued at about $8 billion pre-investment.
- This date and characterization are wrong. Reporting and ISW’s own statement say the problematic map edit was an unauthorized overnight edit on November 15–16, not an official ISW indication on November 14.
- This is outdated. Kalshi’s own help documentation says the platform is now accessible internationally, not exclusive to U.S. users.
LessWrong Mar 15, 2026 at 04:58 PM
www.lesswrong.com/posts/ioZxrP7BhS5ArK59w/did-claude-3-opus-...
- This is true for ReLU, but not for GELU. GELU is defined as x·Φ(x), so negative inputs are generally not clipped to zero and their gradients are not zeroed out in the ReLU sense.
- Anthropic’s cofounder and president is Daniela Amodei, not “Daniella” Amodei.
LessWrong Mar 15, 2026 at 04:57 PM
www.lesswrong.com/posts/RrL7xqdPycGNHQkXR/the-lethal-reality...
- The percentage is off by orders of magnitude. If humanity eventually numbers in the trillions, being roughly the 100–117 billionth human means being in the first ~10% of a 1-trillion total or ~1.2% of a 10-trillion total, not 0.001% or less.
- This mixes up two different meanings of μ in geometric Brownian motion. In standard GBM, log returns have mean (μ − ½σ²)Δt, not μΔt; if log returns really had mean μ, then the time-average log-growth rate would be μ, not μ − ½σ².
Substack Mar 15, 2026 at 04:05 PM
www.theargumentmag.com/p/when-technically-true-becomes-actua...
- This overstates the effect of instruction-tuning and RLHF. Official docs from OpenAI, Anthropic, and Google still describe modern chatbots as generating output token by token, with post-training refining that behavior rather than replacing it.
X Mar 15, 2026 at 03:48 PM
x.com/sethbannon/status/2007161907392958806
- The underlying write-up says the $15 figure was only the variable usage cost and explicitly notes an additional ElevenLabs subscription fee, so it was not the full all-in cost.