Latest Corrections
LessWrong Feb 24, 2026 at 07:01 PM
www.lesswrong.com/posts/bKrpLhqcoN6WycrFp/citrini-s-scenario...
- Democracies have sustained unemployment well above 15%—for example, the United States during the Great Depression saw unemployment around 20–25% for multiple years while remaining a democratic system.
LessWrong Feb 24, 2026 at 06:38 PM
www.lesswrong.com/posts/iMw7qhtZGNFxMRD4H/open-sourcing-a-br...
- One of the example articles listed (“Life at the Frontlines of Demographic Collapse”) was published on February 14, 2026—10 days before this post—so it was not published “within the last week.”
LessWrong Feb 24, 2026 at 05:12 PM
www.lesswrong.com/posts/iMw7qhtZGNFxMRD4H/open-sourcing-a-br...
- One of the listed example articles (“Life at the Frontlines of Demographic Collapse”) was published on February 14, 2026—10 days before this post—so it was not published “within the last week.”
LessWrong Feb 24, 2026 at 05:09 PM
www.lesswrong.com/posts/eopA4MqhrE4dkLjHX/the-truth-behind-t...
- Kircher’s *Mundus Subterraneus* was published in 1665 (often cited as 1664/1665), not 1654.
- Sources describing the 1906 earthquake/fire aftermath say the hotel was gutted and later reopened in 1907 (i.e., it did not operate continuously without interruption).
LessWrong Feb 24, 2026 at 05:01 PM
www.lesswrong.com/posts/wgqcExv9AgN5MuJuY/bioanchors-2-elect...
- JCVI-syn3.0 (2016) is not the smallest genome of any self-replicating organism; naturally occurring bacteria with far smaller genomes were known earlier (e.g., Carsonella ruddii at 159,662 bp).
LessWrong Feb 24, 2026 at 03:41 PM
www.lesswrong.com/posts/FuGfR3jL3sw6r8kB4?commentId=6QXQ5nK4...
- XBOW’s “top spot on the US leaderboard” announcement is dated June 24, 2025—about 8 months before this post (Feb 19, 2026), not “almost one year.”
LessWrong Feb 24, 2026 at 05:59 AM
www.lesswrong.com/posts/dxffBxGqt2eidxwRR/the-optimal-age-to...
- This is contradicted by peer‑reviewed cost/benefit decision analyses, which find the most cost‑effective (or largest benefit-over-no-action) timing is around ages 36–37, not 19–26.
- Major reproductive-medicine and human-genetics societies state that polygenic embryo screening/testing is not ready for clinical use and has low-to-nonexistent clinical utility, so presenting it as something parents “can now” use to achieve these outcomes is misleading.
LessWrong Feb 24, 2026 at 05:59 AM
www.lesswrong.com/posts/ioZxrP7BhS5ArK59w/did-claude-3-opus-...
- In the original “Alignment faking in large language models” paper, the reported rate of alignment-faking reasoning for Claude 3 Opus is far below 50% (reported as 7–24%, with 12% in the default prompted helpful-only setup).
- The original paper and later follow-up both report alignment faking in other models (notably Claude 3.5 Sonnet), contradicting the claim that no other model considered strategic alignment faking nearly as often.
LessWrong Feb 24, 2026 at 03:58 AM
www.lesswrong.com/posts/ioZxrP7BhS5ArK59w/did-claude-3-opus-...
- The paper reports alignment-faking reasoning far below 50% in its main setups (e.g., 12% in the prompted “helpful-only” setup; 7–24% across setups). Only one later RL-training experiment reports 78%.
- The Alignment Faking paper itself reports that Claude 3.5 Sonnet also exhibits alignment faking, and that using Sonnet rather than Opus can increase the alignment-faking rate in at least one setting—so it’s not unique to Opus 3.
LessWrong Feb 24, 2026 at 03:51 AM
www.lesswrong.com/posts/8m6AM5qtPMjgTkEeD/my-journey-to-the-...
- U.S. studies of (and serious planning for) a new sea-level Panama canal route—including nuclear excavation concepts—continued through the late 1960s, with major reports and termination of field operations occurring around 1969–1970, not “by the early 1960s.”
- Household microwave ovens are typically operated at 2.45 GHz, which corresponds to a wavelength of about 12.2 cm—outside the stated “5–10 cm” range.
- There was no live U.S. television coverage from Hiroshima in 1945; in fact, the first live telecast from Japan to the U.S. is documented as occurring in 1964.
LessWrong Feb 24, 2026 at 03:02 AM
www.lesswrong.com/posts/dxffBxGqt2eidxwRR/the-optimal-age-to...
- There is no near-universal agreement that the optimal age is 19–26; peer‑reviewed decision models and major medical guidance commonly place “optimal” timing later (often early–mid 30s, with cost‑effectiveness peaking around 37 in one widely cited model).
- SART’s public outcome tables explicitly report outcomes for frozen embryo transfers and describe tracking outcomes from an egg retrieval across fresh and frozen transfers, so later births from subsequent transfers are not “ignored” in official reporting.
- Major professional guidance (ASRM) says polygenic embryo screening/PGT‑P is “nascent and unproven,” “not recommended for clinical use,” and should not be offered clinically; European Society of Human Genetics calls PRS embryo selection unproven with no clinical research validating effectiveness in embryos.
LessWrong Feb 24, 2026 at 02:57 AM
www.lesswrong.com/posts/iMw7qhtZGNFxMRD4H/open-sourcing-a-br...
- OpenErrata’s own documentation says the public instance is configured to automatically investigate popular posts every hour, meaning new investigations can be started even when a user has no OpenAI API key.
LessWrong Feb 24, 2026 at 02:55 AM
www.lesswrong.com/posts/iMw7qhtZGNFxMRD4H/open-sourcing-a-br...
- OpenErrata did not "only" work on LessWrong at the time of the post; the project’s own README (from Feb 22, 2026) lists support for X (Twitter) and Substack in addition to LessWrong.
LessWrong Feb 24, 2026 at 02:42 AM
www.lesswrong.com/posts/TsDcAZJB9sdj57KCo/beware-llms-pathol...
- This broadly mischaracterizes how many modern LLMs are trained: major models’ post-training is typically driven by human preference feedback / preference optimization rather than reinforcement learning against code execution failures like crashes or unit test results.
LessWrong Feb 23, 2026 at 10:58 PM
www.lesswrong.com/posts/hurF9uFGkJYXzpHEE/a-non-magical-expl...
- The “15-page psychological report” was produced by the Bureau of Prisons (BOP), not by the FBI or DOJ’s Office of the Inspector General (OIG).
- DOJ OIG found that most nearby SHU cameras weren’t recording; the camera at the end of Epstein’s tier was live-streaming but not recording due to a DVR hard-drive failure.
- OIG reports the FBI forensically analyzed the DVR systems and hard drives (finding faulty drives/catastrophic disk failures), not that the FBI “tore apart” the cameras in front of Epstein’s cell.
Substack Feb 23, 2026 at 05:41 PM
aleximas.substack.com/p/will-advanced-ai-lead-to-negative?r=...
- Martin Ford’s FT/McKinsey-winning book is titled “Rise of the Robots: Technology and the Threat of a Jobless Future,” not “How Artificial Intelligence Will Transform Everything.”
LessWrong Feb 23, 2026 at 08:15 AM
www.lesswrong.com/posts/fkKcftthj2fhSGDje/most-successful-en...
- As of the post date (2025-12-22), there were multiple widely-used, enterprise-grade alternatives to Nvidia accelerators for AI workloads (e.g., Google Cloud TPUs, AWS Trainium, Intel Gaudi, AMD Instinct).
- Many major workloads now accelerated by GPUs (computer-graphics rendering and neural-network-based recognition) existed and were performed before modern GPUs; GPUs mostly made them faster/cheaper, not newly possible in absolute terms.
Substack Feb 23, 2026 at 06:13 AM
www.astralcodexten.com/p/crime-as-proxy-for-disorder
- Spray paint was invented in 1949 (before 1950), and graffiti as wall inscriptions/drawings existed long before spray paint.
- Boomboxes were invented decades earlier (1966) and became widely popular in the 1970s and 1980s, not “around 1990.”
- Graffiti was not “uninvented” until modern times; it existed for centuries (including in ancient Roman contexts), long before mid-20th-century housing/suburbanization trends.
LessWrong Feb 23, 2026 at 04:14 AM
www.lesswrong.com/posts/qefrWyeiMvWEFRitN/milestone-announce...
- XBOW’s own “Top 1” announcement is dated June 24, 2025—about 8 months before this LessWrong post (Feb 19, 2026), not “almost one year.”
- The XBOW announcement post being discussed is dated June 24, 2025, so by Feb 19, 2026 it’s been about 8 months, not “almost a year later.”
Substack Feb 23, 2026 at 03:20 AM
www.richardhanania.com/p/why-donald-trump-and-joe-rogan-are
- Louis Freeh was confirmed by a voice vote, not a recorded 95–0 roll-call vote.
- Patel was confirmed 51–49 (not 52–48), and Senate records show Mitch McConnell voted “Yea,” not against him.