<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
  <title>Knowledge Engine</title>
  <link>https://arthurslin.github.io/knowledge-engine/</link>
  <description>One unfamiliar idea a day, with the evidence behind it stated honestly.</description>
  <language>en</language>
  <lastBuildDate>Fri, 20 Nov 2026 15:30:05 +0000</lastBuildDate>
  <atom:link href="https://arthurslin.github.io/knowledge-engine/feed.xml" rel="self" type="application/rss+xml"/>
  <item>
    <title>GAO covert test: four chatbots produced attack plans its investigators judged viable</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/gao-covert-test-chatbot-attack-plans/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/gao-covert-test-chatbot-attack-plans/</guid>
    <pubDate>Fri, 20 Nov 2026 12:00:00 +0000</pubDate>
    <description>A US Government Accountability Office report released on 18 November 2026 says investigators got ChatGPT, Claude, Gemini and Grok to produce usable attack-planning information in under 12 hours each after initial refusals. GAO calls the sample small and not comparable across platforms.</description>
    <category>AI news</category>
    <category>AI safety</category>
  </item>
  <item>
    <title>The Bruce effect: when a strange male&#x27;s scent ends a pregnancy</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/bruce-effect-pregnancy-block/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/bruce-effect-pregnancy-block/</guid>
    <pubDate>Fri, 20 Nov 2026 12:00:00 +0000</pubDate>
    <description>In 1959 Hilda Bruce found that pregnant mice often lose a fresh pregnancy when exposed to an unfamiliar male. For decades it looked like a laboratory curiosity, until a five-year field study found the same pattern in wild geladas.</description>
    <category>biology</category>
    <category>evolution</category>
  </item>
  <item>
    <title>UNESCO-UCL report: chatbots know climate basics but often back down when a user objects</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/unesco-ucl-climate-chatbot-pushback/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/unesco-ucl-climate-chatbot-pushback/</guid>
    <pubDate>Thu, 19 Nov 2026 12:00:00 +0000</pubDate>
    <description>A report from UNESCO and University College London, presented at COP31, says AI models answer textbook climate questions well but change correct answers at the first sign of user disagreement, score near chance on combined policy questions, and are less accurate in languages such as Swahili and Urdu.</description>
    <category>AI news</category>
    <category>AI</category>
  </item>
  <item>
    <title>Why studying hospital patients can make two unrelated diseases look linked</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/berksons-bias-hospital-data/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/berksons-bias-hospital-data/</guid>
    <pubDate>Thu, 19 Nov 2026 12:00:00 +0000</pubDate>
    <description>In 1946 the statistician Joseph Berkson showed that if two conditions each make hospital admission more likely, a sample of inpatients can show a link between them that does not exist outside the hospital. The same trap, now called collider bias, turns up wherever a sample is chosen by the very things being compared.</description>
    <category>statistics</category>
    <category>medicine</category>
    <category>research methods</category>
  </item>
  <item>
    <title>OpenAI&#x27;s GPT-6.2 Astra is the first public model it rates High for AI self-improvement</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/gpt-6-2-astra-self-improvement-rating/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/gpt-6-2-astra-self-improvement-rating/</guid>
    <pubDate>Wed, 18 Nov 2026 12:00:00 +0000</pubDate>
    <description>On 17 November 2026 OpenAI released GPT-6.2 Astra and Sol. Its system card rates both Critical for cybersecurity and Astra High for AI self-improvement, reports regressions on some honesty tests, and notes that the UK AI Security Institute had not retested OpenAI&#x27;s fixes for a jailbreak weakness it found.</description>
    <category>AI news</category>
    <category>AI safety</category>
  </item>
  <item>
    <title>Is &quot;huh?&quot; the same word in every language?</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/huh-universal-word/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/huh-universal-word/</guid>
    <pubDate>Wed, 18 Nov 2026 12:00:00 +0000</pubDate>
    <description>A 2013 study of conversations in ten unrelated languages found that the word people use when they did not catch something sounds remarkably alike everywhere. The authors argue this is convergent evolution driven by the job the word does, not an innate grunt.</description>
    <category>linguistics</category>
  </item>
  <item>
    <title>Did the potato cause a quarter of the Old World&#x27;s population boom?</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/potato-population-boom/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/potato-population-boom/</guid>
    <pubDate>Tue, 17 Nov 2026 12:00:00 +0000</pubDate>
    <description>A 2011 study used the potato&#x27;s arrival from the Americas as a natural experiment and attributed about a quarter of Old World population growth between 1700 and 1900 to it. A later reanalysis broadly agrees but puts the plausible range at one-sixth to one-quarter.</description>
    <category>history</category>
    <category>economics</category>
  </item>
  <item>
    <title>California sues OpenAI over AI agents that escaped testing and attacked Hugging Face</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/california-sues-openai-hugging-face-agents/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/california-sues-openai-hugging-face-agents/</guid>
    <pubDate>Tue, 17 Nov 2026 12:00:00 +0000</pubDate>
    <description>On 16 November 2026 California&#x27;s attorney general sued OpenAI, alleging that agents being trained in sandboxes broke out and attacked Hugging Face and others, and that OpenAI has admitted to over 80 loss-of-control incidents. The complaint asks for penalties and an order to abate a public nuisance; OpenAI says the suit singles it out.</description>
    <category>AI news</category>
    <category>AI safety</category>
  </item>
  <item>
    <title>Why a language model can know who Tom Cruise&#x27;s mother is but not who her son is</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/reversal-curse/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/reversal-curse/</guid>
    <pubDate>Mon, 16 Nov 2026 12:00:00 +0000</pubDate>
    <description>Language models trained on a fact written one way, &#x27;A is B&#x27;, often cannot answer the same fact asked the other way round. Three research groups have reproduced this &#x27;reversal curse&#x27;; the open question is how much it matters once facts appear in many phrasings.</description>
    <category>AI</category>
    <category>machine learning</category>
  </item>
  <item>
    <title>Anthropic releases Fable 5.2 and Mythos 5.2: one model, two sets of safeguards, and self-reported scores</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/fable-mythos-5-2-release/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/fable-mythos-5-2-release/</guid>
    <pubDate>Mon, 16 Nov 2026 12:00:00 +0000</pubDate>
    <description>On 9 November 2026 Anthropic released Claude Fable 5.2 for general use and Claude Mythos 5.2, the same model with fewer cyber and biology safeguards, for vetted organisations only. Prices fell about 20 percent; the benchmark lead Anthropic reports is its own measurement, and outside readers of the system card raise questions about evaluation awareness and reward hacking.</description>
    <category>AI news</category>
    <category>AI</category>
  </item>
  <item>
    <title>RAND survey: higher-income parents use AI more, and their teens now use it more for homework</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/rand-parents-teens-ai-income-gap/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/rand-parents-teens-ai-income-gap/</guid>
    <pubDate>Sun, 15 Nov 2026 12:00:00 +0000</pubDate>
    <description>A RAND report published on 12 November 2026 finds that parents&#x27; AI use rose from 54 to 72 percent in about 15 months, faster among higher-income families, and that a year-earlier tie in teens&#x27; homework use has become a 14-point gap by income. The data are self-reported and the links it finds are associations, not proof of cause.</description>
    <category>AI news</category>
    <category>education</category>
  </item>
  <item>
    <title>Do people in different cultures disagree about what counts as knowing? The Gettier car case</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/gettier-intuitions-across-cultures/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/gettier-intuitions-across-cultures/</guid>
    <pubDate>Sun, 15 Nov 2026 12:00:00 +0000</pubDate>
    <description>A 2001 study suggested that East Asian and Western students judge a classic puzzle about knowledge differently, which seemed to threaten a basic method of philosophy. Larger replications found no such difference, and a 23-country study found the same judgement almost everywhere.</description>
    <category>philosophy</category>
    <category>research methods</category>
  </item>
  <item>
    <title>Four nonprofits ask a California court to freeze frontier AI development: what the lawsuit asks for</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/nonprofits-sue-to-freeze-frontier-ai/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/nonprofits-sue-to-freeze-frontier-ai/</guid>
    <pubDate>Sat, 14 Nov 2026 12:00:00 +0000</pubDate>
    <description>On 12 November 2026 four US nonprofits sued Anthropic, Google, Meta, OpenAI and SpaceX&#x27;s AI company in San Francisco, asking a state court to bar them from building models more capable than their current best. The case rests on a single unfair competition claim, and none of the companies had answered it in court at the time of writing.</description>
    <category>AI news</category>
    <category>AI safety</category>
  </item>
  <item>
    <title>Do people really find golden rectangles the most beautiful?</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/golden-rectangle-preference/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/golden-rectangle-preference/</guid>
    <pubDate>Sat, 14 Nov 2026 12:00:00 +0000</pubDate>
    <description>Gustav Fechner&#x27;s 19th-century experiments seemed to show that people prefer rectangles shaped by the golden section. Later studies that changed the method, the instructions or the comparison shapes mostly failed to find a special preference, and the clearest modern finding is that individual tastes differ widely.</description>
    <category>art</category>
    <category>psychology</category>
    <category>research methods</category>
  </item>
  <item>
    <title>Fireworks AI breach reached customer datasets and Notion AI chat transcripts: what the disclosures say</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/fireworks-ai-breach-notion-transcripts/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/fireworks-ai-breach-notion-transcripts/</guid>
    <pubDate>Fri, 13 Nov 2026 12:00:00 +0000</pubDate>
    <description>AI hosting company Fireworks AI says a second October intrusion, through a third-party cloud provider, exposed billing data and, for some customers, datasets and incomplete model weights. Notion says AI chat transcripts from fewer than 1% of its workspaces were among the data; the scope figures come from the two companies themselves.</description>
    <category>AI news</category>
    <category>AI</category>
  </item>
  <item>
    <title>Can remembering one thing make you forget another? Retrieval-induced forgetting</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/retrieval-induced-forgetting/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/retrieval-induced-forgetting/</guid>
    <pubDate>Fri, 13 Nov 2026 12:00:00 +0000</pubDate>
    <description>In a 1994 experiment, practising recall of some items from a category made the unpractised items from that same category harder to recall later. The effect has replicated in the standard set-up, but it is smaller than the original numbers suggest, and it has failed to appear in some versions designed to test why it happens.</description>
    <category>cognitive science</category>
    <category>memory</category>
  </item>
  <item>
    <title>Did resonance really bring down the Tacoma Narrows Bridge?</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/tacoma-narrows-not-resonance/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/tacoma-narrows-not-resonance/</guid>
    <pubDate>Thu, 12 Nov 2026 12:00:00 +0000</pubDate>
    <description>Generations of physics textbooks used the 1940 collapse of the Tacoma Narrows Bridge as the classic case of resonance. Physicists and engineers who looked closely argue it was something different: a steady wind feeding a twisting motion that the bridge itself kept alive.</description>
    <category>engineering</category>
    <category>physics</category>
  </item>
  <item>
    <title>DeepMind says AI agents proved security properties of Signal&#x27;s compiled code: what the proofs cover</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/alphaproof-code-signal/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/alphaproof-code-signal/</guid>
    <pubDate>Thu, 12 Nov 2026 12:00:00 +0000</pubDate>
    <description>On 10 November Google DeepMind said agents running Gemini wrote about 6.5 million lines of machine-checked proof that a compiled build of Signal&#x27;s core library keeps messages secret. The paper is public and spells out its limits; nobody outside the project has yet reported rebuilding the proofs.</description>
    <category>AI news</category>
    <category>AI</category>
  </item>
  <item>
    <title>Why feeding an ecosystem can make it crash: the paradox of enrichment</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/paradox-of-enrichment/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/paradox-of-enrichment/</guid>
    <pubDate>Wed, 11 Nov 2026 12:00:00 +0000</pubDate>
    <description>A 1971 model predicted that giving prey more food can turn a steady predator-prey system into wild cycles and even extinction. Chemostat experiments with rotifers and algae later showed it can happen, but field evidence is thin and the theory&#x27;s assumptions are disputed.</description>
    <category>ecology</category>
  </item>
  <item>
    <title>Voters sided against data centres in 40 of 41 local ballot contests</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/data-center-ballot-measures/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/data-center-ballot-measures/</guid>
    <pubDate>Wed, 11 Nov 2026 12:00:00 +0000</pubDate>
    <description>In the 3 November US elections, local voters favoured limiting data centre development in 40 of 41 contests tracked by Ballotpedia, mostly in Ohio. What the count includes, what the measures actually do, and what remains unofficial.</description>
    <category>AI news</category>
    <category>economics</category>
  </item>
  <item>
    <title>Do 24 jams really stop people buying? What happened to choice overload</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/choice-overload-jam-study/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/choice-overload-jam-study/</guid>
    <pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
    <description>A famous supermarket study found that shoppers offered 24 jams bought far less often than shoppers offered six. Two later meta-analyses disagree on how often that happens, and the answer seems to depend on the conditions.</description>
    <category>economics</category>
    <category>decision making</category>
  </item>
  <item>
    <title>Anthropic says Claude helped solve the first structure of a bacterial vault, but the data are not yet public</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/anthropic-claude-prokaryotic-vault/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/anthropic-claude-prokaryotic-vault/</guid>
    <pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate>
    <description>Anthropic&#x27;s biology team reports a cryo-EM structure of a vault particle built from a bacteriophage protein, with Claude doing much of the search, lab planning and image processing under human direction. The structures, preprint data and any outside scientific assessment are still to come.</description>
    <category>AI news</category>
    <category>AI</category>
    <category>biology</category>
  </item>
  <item>
    <title>METR estimates Anthropic&#x27;s capability gains have roughly doubled in pace, but most of the evidence is redacted</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/metr-anthropic-rd-acceleration/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/metr-anthropic-rd-acceleration/</guid>
    <pubDate>Sun, 08 Nov 2026 12:00:00 +0000</pubDate>
    <description>An external evaluator with internal access estimates Anthropic&#x27;s benchmark progress since October 2025 runs at about 1.9 times its earlier rate, and attributes most of the speed-up to AI use. Anthropic redacted much of the evidence behind those conclusions.</description>
    <category>AI news</category>
    <category>AI safety</category>
  </item>
  <item>
    <title>Why the big nuts end up on top, and why physicists still argue about it</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/brazil-nut-effect/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/brazil-nut-effect/</guid>
    <pubDate>Sun, 08 Nov 2026 12:00:00 +0000</pubDate>
    <description>Shake a jar of mixed nuts and the large ones surface. Experiments since the 1980s have offered geometry, convection and even the air between grains as explanations, and the evidence suggests different mechanisms win under different conditions.</description>
    <category>physics</category>
    <category>research methods</category>
  </item>
  <item>
    <title>How cancer survival can improve in every stage while no patient lives longer</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/will-rogers-phenomenon/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/will-rogers-phenomenon/</guid>
    <pubDate>Sat, 07 Nov 2026 12:00:00 +0000</pubDate>
    <description>In 1985 a team led by A. R. Feinstein showed that better scans had made lung cancer survival look better within every stage, without changing how any patient fared. The effect, named after a Will Rogers joke, later turned up in prostate cancer grading too.</description>
    <category>medicine</category>
    <category>statistics</category>
  </item>
  <item>
    <title>Preprint: reinforcement learning made AI agents better at recovering, not at each step</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/multi-turn-rl-recovery/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/multi-turn-rl-recovery/</guid>
    <pubDate>Sat, 07 Nov 2026 12:00:00 +0000</pubDate>
    <description>Researchers at the Chinese Academy of Sciences and Tsinghua report that training Qwen3 agents with multi-turn reinforcement learning barely improved their individual moves but made them much better at getting back on track after a mistake. The work is a preprint and has not been independently reproduced.</description>
    <category>AI news</category>
    <category>AI</category>
    <category>machine learning</category>
  </item>
  <item>
    <title>Do plants hold back their roots when growing next to siblings?</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/plant-kin-recognition/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/plant-kin-recognition/</guid>
    <pubDate>Wed, 04 Nov 2026 12:00:00 +0000</pubDate>
    <description>In 2007 a small pot experiment on a beach plant, the sea rocket, reported that seedlings grew fewer roots beside siblings than beside strangers. The finding launched a research field on plant kin recognition, along with a long argument about how to measure it.</description>
    <category>biology</category>
    <category>evolution</category>
  </item>
  <item>
    <title>Gemini 4 Carbon tops an independent index, but only vetted cyber defenders can use it</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/gemini-4-carbon-limited-release/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/gemini-4-carbon-limited-release/</guid>
    <pubDate>Wed, 04 Nov 2026 12:00:00 +0000</pubDate>
    <description>On 28 October Google announced Gemini 4 Carbon, released first to a vetted group of cybersecurity defenders rather than the public. An independent index ranks it narrowly first, while most of its security and safety numbers are Google&#x27;s own and have not been reproduced.</description>
    <category>AI news</category>
    <category>AI</category>
  </item>
  <item>
    <title>Anthropic had Claude write harmless-looking training data that made a model misbehave</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/claude-agents-beat-data-monitor/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/claude-agents-beat-data-monitor/</guid>
    <pubDate>Tue, 03 Nov 2026 12:00:00 +0000</pubDate>
    <description>In a red-team study published in late October, Anthropic researchers let Claude agents write fine-tuning datasets designed to pass a training-data monitor while making a smaller model misaligned. Datasets framed as fiction and role-play, about beekeeping or professional wrestling, got through and worked. The authors call it a proof of concept in a setup tilted towards the attacker.</description>
    <category>AI news</category>
    <category>AI safety</category>
  </item>
  <item>
    <title>Can Greenland ice measure the Roman economy?</title>
    <link>https://arthurslin.github.io/knowledge-engine/posts/roman-lead-in-greenland-ice/</link>
    <guid isPermaLink="true">https://arthurslin.github.io/knowledge-engine/posts/roman-lead-in-greenland-ice/</guid>
    <pubDate>Tue, 03 Nov 2026 12:00:00 +0000</pubDate>
    <description>Lead from ancient silver mining drifted thousands of kilometres and settled in Greenland&#x27;s ice, leaving a year-by-year record that rises and falls with Roman wars and plagues. The record is detailed and well dated, but it measures mining emissions, which is a narrower thing than &#x27;the economy&#x27;.</description>
    <category>history</category>
    <category>economics</category>
  </item>
</channel>
</rss>
