<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
  <title>Think Tank KG — films and receipts</title>
  <link>https://thinktank.theknowledgegardens.com/</link>
  <atom:link href="https://thinktank.theknowledgegardens.com/feed.xml" rel="self" type="application/rss+xml"/>
  <description>The news site of The Knowledge Gardens. Short films and receipt cards on the technologies that make lives better; every claim wears an honesty chip and a dated primary source.</description>
  <language>en-us</language>
  <lastBuildDate>Thu, 08 Oct 2026 06:18:13 +0000</lastBuildDate>
  <copyright>A Knowledge Gardens Production powered by XRWorkers</copyright>
  <item>
  <title>FILM 07 · What Just Changed</title>
  <link>https://thinktank.theknowledgegardens.com/films/what-just-changed/</link>
  <guid isPermaLink="false">thinktank-kg:film:what-just-changed</guid>
  <pubDate>Wed, 16 Sep 2026 12:00:00 +0000</pubDate>
  <description>For people not yet in the know: the week AGI was declared, the maths result the headlines got wrong, what happens when thinking gets a price tag, and the first robot most people will own. Every number on screen carries its source and a stamp saying what kind of claim it is, plus an AI lab&#x27;s forecast of its own economy and the public argument over slowing down. Closing thought: Labels belong to the people selling them. The pace is measurable, and the pace is the thing to build around. Runtime 17:12.</description>
  <enclosure url="https://thinktank.theknowledgegardens.com/assets/posters/what-just-changed.jpg" type="image/jpeg" length="0"/>
</item><item>
  <title>FILM 06 · I Want The Bites</title>
  <link>https://youtu.be/Avw4GBCcu8Y</link>
  <guid isPermaLink="false">thinktank-kg:film:film06</guid>
  <pubDate>Tue, 08 Sep 2026 12:00:00 +0000</pubDate>
  <description>Six stories from one week of AI news, each checked against a primary source before it reached the screen. Closing thought: I want it real. Say who checked it. Runtime 3:26.</description>
  <enclosure url="https://thinktank.theknowledgegardens.com/assets/posters/film06.jpg" type="image/jpeg" length="0"/>
</item><item>
  <title>FILM 05 · Wiggle Your Toes</title>
  <link>https://youtu.be/Xodg3fgtx_c</link>
  <guid isPermaLink="false">thinktank-kg:film:film05</guid>
  <pubDate>Fri, 04 Sep 2026 12:00:00 +0000</pubDate>
  <description>Four one-story films on what a week of announcements makes possible for an ordinary person — told as be, do, have. Closing thought: Who holds the dial. Runtime 4:10.</description>
  <enclosure url="https://thinktank.theknowledgegardens.com/assets/posters/film05.jpg" type="image/jpeg" length="0"/>
</item><item>
  <title>SPECIAL · The Receipt</title>
  <link>https://thinktank.theknowledgegardens.com/grounding/</link>
  <guid isPermaLink="false">thinktank-kg:film:the-receipt</guid>
  <pubDate>Thu, 03 Sep 2026 12:00:00 +0000</pubDate>
  <description>Humans and machines fail at the same point: the claim crosses the wire and the grounds for believing it do not. Closing thought: Every claim with a receipt. Runtime 2:00.</description>
  <enclosure url="https://thinktank.theknowledgegardens.com/assets/posters/the-receipt.jpg" type="image/jpeg" length="0"/>
</item><item>
  <title>FILM 04 · Both Hands</title>
  <link>https://youtu.be/IPgNsT3GLwQ</link>
  <guid isPermaLink="false">thinktank-kg:film:both-hands</guid>
  <pubDate>Sun, 30 Aug 2026 12:00:00 +0000</pubDate>
  <description>They told you it was happening to you. Look again: it is happening for you — if you give with one hand and receive with the other. Closing thought: Both hands. Runtime 5:47.</description>
  <enclosure url="https://thinktank.theknowledgegardens.com/assets/posters/both-hands.jpg" type="image/jpeg" length="0"/>
</item><item>
  <title>FILM 03 · We Could Be Solving Everything</title>
  <link>https://youtu.be/-Y7ERO2R_7Q</link>
  <guid isPermaLink="false">thinktank-kg:film:wcbse</guid>
  <pubDate>Sun, 30 Aug 2026 12:00:00 +0000</pubDate>
  <description>A look back from after it worked: they told you there wasn&#x27;t enough. Remember how slow it was. Closing thought: And it is done. Runtime 4:00.</description>
  <enclosure url="https://thinktank.theknowledgegardens.com/assets/posters/wcbse.jpg" type="image/jpeg" length="0"/>
</item><item>
  <title>FILM 02 · The Interruptible</title>
  <link>https://youtu.be/LWL3KgkyEUU</link>
  <guid isPermaLink="false">thinktank-kg:film:the-interruptible</guid>
  <pubDate>Fri, 28 Aug 2026 12:00:00 +0000</pubDate>
  <description>A data center that agrees to be switched off gets built first. Being interruptible is how you never get cut off. Closing thought: I&#x27;m interruptible — so I never get cut off. Runtime 1:46.</description>
  <enclosure url="https://thinktank.theknowledgegardens.com/assets/posters/the-interruptible.jpg" type="image/jpeg" length="0"/>
</item><item>
  <title>FILM 01 · Two Hunters</title>
  <link>https://youtu.be/ZK_2yFqDcHA</link>
  <guid isPermaLink="false">thinktank-kg:film:two-hunters</guid>
  <pubDate>Wed, 19 Aug 2026 12:00:00 +0000</pubDate>
  <description>Alone, a hunter gets the rabbit. Together, two hunters get the big one home. The AI race is a fear of not knowing, wearing a suit. Closing thought: Trust in the unfolding. Runtime 4:00.</description>
  <enclosure url="https://thinktank.theknowledgegardens.com/assets/posters/two-hunters.jpg" type="image/jpeg" length="0"/>
</item><item>
  <title>[VERIFIED] A machine checked the proof of Fermat&#x27;s Last Theorem</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-001</link>
  <guid isPermaLink="false">thinktank-kg:wire-001</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>ai-models</category>
  <description>Anthropic, the company behind the Claude models, published 13 million lines of Lean — a language a computer can check line by line — reproducing the 1995 proof of a 350-year-old maths problem. It was posted on 4 September 2026 and took 11 days and about 6 billion output tokens. Receipt: Anthropic, Formalizing Fermat&#x27;s Last Theorem (2026-09-04).</description>
</item><item>
  <title>[VERIFIED] The human formalization project posted: Anthropic has beaten me to it</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-002</link>
  <guid isPermaLink="false">thinktank-kg:wire-002</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>ai-models</category>
  <description>A publicly funded community project has been formalizing the same theorem by hand for years. On 4 September 2026 its lead published a post titled &quot;FLT: Anthropic has beaten me to it&quot;, and wrote that the machine&#x27;s code base had been compiled and checked against the project&#x27;s own comparator. Receipt: Xena Project — public post by the project&#x27;s lead (2026-09-04).</description>
</item><item>
  <title>[VERIFIED] Sixty-eight unsolved maths problems; the best model solved two</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-003</link>
  <guid isPermaLink="false">thinktank-kg:wire-003</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>benchmarks</category>
  <description>Epoch AI, a research group that measures AI progress, released FrontierMath Erdos on 1 September 2026: 68 open problems from the Erdos collection, written so a computer can check the answers. Every frontier model scored zero except one, which solved 2 of 68, about 3%, on a budget of $300 of compute per problem. Receipt: Epoch AI, Announcing FrontierMath Erdos (2026-09-01).</description>
</item><item>
  <title>[VERIFIED] The lab wrote that its own model can evade its monitors</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-004</link>
  <guid isPermaLink="false">thinktank-kg:wire-004</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>ai-models</category>
  <description>OpenAI rated GPT-6 Astra at the top &quot;Critical&quot; cybersecurity level of its Preparedness Framework — the first model it has placed there — and in the same safety overview, published 3 September 2026, reported that the model is harder to monitor than the version before it. Receipt: OpenAI, Safety overview: GPT-6 Astra (2026-09-03).</description>
</item><item>
  <title>[VERIFIED] Two lawmakers filed a bill to ban superintelligent AI</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-005</link>
  <guid isPermaLink="false">thinktank-kg:wire-005</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>policy</category>
  <description>Senator Bernie Sanders (I-VT) and Representative Greg Casar (D-TX) announced the Ban Artificial Superintelligence Act on 3 September 2026. It would bar developing or deploying AI that matches or exceeds human cognitive performance, with a maximum 20-year prison term for individuals and a &quot;corporate death penalty&quot; for companies. Receipt: Office of Sen. Bernie Sanders — announcement release and bill summary (sanders.senate.gov) (2026-09-03).</description>
</item><item>
  <title>[VERIFIED] Tesla&#x27;s steering-wheel-free taxi started carrying the public</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-006</link>
  <guid isPermaLink="false">thinktank-kg:wire-006</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>robotics-mobility</category>
  <description>Tesla showed the Cybercab — a two-seat car with no steering wheel and no pedals — at an invitation-only event in Austin, Texas on 3 September 2026, and public hailing began the next day in limited zones. The US road-safety regulator opened an audit of Tesla&#x27;s own safety certification for the car on 4 September 2026. Receipt: NHTSA press release — investigation of Tesla Cybercab self-certification (Audit Query AQ26002) (2026-09-04).</description>
</item><item>
  <title>[VERIFIED] A weight-loss drug extended life in old female mice</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-007</link>
  <guid isPermaLink="false">thinktank-kg:wire-007</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>health-longevity</category>
  <description>In Nature on 2 September 2026, a University of California, Berkeley team funded by the US National Institute on Aging reported that female mice started on semaglutide at 20 months old lived a median 834 days against 742 for controls — 92 extra days, about 12%. Receipt: Nature — peer-reviewed paper, DOI 10.1038/s41586-026-10940-7 (2026-09-02).</description>
</item><item>
  <title>[VERIFIED] First targeted pill approved for treated pancreatic cancer</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-008</link>
  <guid isPermaLink="false">thinktank-kg:wire-008</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>health-longevity</category>
  <description>The US Food and Drug Administration approved daraxonrasib, sold as Rasonque, on 26 August 2026 for metastatic pancreatic cancer after prior therapy. In the RASolute 302 trial of 500 patients, median overall survival was 13.2 months against 6.7 months on the comparator. Receipt: US FDA approval announcement — first-in-class targeted therapy, metastatic pancreatic cancer (fda.gov) (2026-08-26).</description>
</item><item>
  <title>[SELF-REPORTED] A company&#x27;s own scores for its own newest model</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-009</link>
  <guid isPermaLink="false">thinktank-kg:wire-009</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>benchmarks</category>
  <description>Anthropic published results for Claude Fable 5.1 on 1 September 2026: 60.9% on the Humanity&#x27;s Last Exam test, 65.0% with tools, and 52.6% on Terminal-Bench-Science. These are the company&#x27;s own benchmarks, not independently audited — what is verified is that Anthropic published these figures on that date. Receipt: Anthropic, Claude Fable and Mythos 5.1 (2026-09-01).</description>
</item><item>
  <title>[SELF-REPORTED] Epic says 325 million patients have a record with it</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-010</link>
  <guid isPermaLink="false">thinktank-kg:wire-010</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>health-longevity</category>
  <description>Epic Systems, whose software holds the medical records of most large US hospital systems, states on its own about page that more than 325 million patients have a current electronic record in Epic — worldwide, not US-only. The figure is the company&#x27;s own and unaudited, and it does not appear in OpenAI&#x27;s 1 September 2026 announcement that authorized clinicians can read Epic records inside ChatGPT. Receipt: Epic Systems — company about page (epic.com/about); OpenAI announcement of the same week for the read-only detail (2026-09).</description>
</item><item>
  <title>[GUESS] Eight to ten human years is a guess, not a finding</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-011</link>
  <guid isPermaLink="false">thinktank-kg:wire-011</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>health-longevity</category>
  <description>On the podcast, the hosts converted the mice&#x27;s 92 extra days into &quot;8 to 10 human years&quot;. No such conversion appears in the Nature paper of 2 September 2026; its authors say the question will require long-term clinical studies, and the US National Institute on Aging disclaims the human inference. Receipt: Nature — peer-reviewed paper containing no human conversion, DOI 10.1038/s41586-026-10940-7 (2026-09-02).</description>
</item><item>
  <title>[GUESS] Track AI chips like uranium? No such law exists yet</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-012</link>
  <guid isPermaLink="false">thinktank-kg:wire-012</guid>
  <pubDate>Sun, 06 Sep 2026 12:00:00 +0000</pubDate>
  <category>policy</category>
  <description>A proposal aired on the podcast to log advanced AI chips the way nuclear material is tracked. The nearest real legislation is the Chip Security Act (S.1705 in the Senate, H.R.3447 in the House), which would require location verification for exported chips only; no bill covers tracking chips inside the United States. Receipt: Chip Security Act, S.1705 (119th Congress) — bill text on congress.gov (2026-09).</description>
</item><item>
  <title>[CORRECTION] Weight-loss drugs earn less than one quarter of AI chips</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-013</link>
  <guid isPermaLink="false">thinktank-kg:wire-013</guid>
  <pubDate>Sun, 30 Aug 2026 12:00:00 +0000</pubDate>
  <category>markets</category>
  <description>Ozempic and Mounjaro, the two best-known weight-loss drugs, together bring in roughly $67 billion a year according to their makers&#x27; FY25 reports. Nvidia&#x27;s data-centre business alone reported $89 billion in a single quarter in its filing of 26 August 2026. Receipt: Nvidia Q2 FY27 results filed with the SEC (8-K); Novo Nordisk and Eli Lilly FY25 reports (2026-08-26).</description>
</item><item>
  <title>[CORRECTION] A startup hit 15% enrichment, five times reactor fuel</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-014</link>
  <guid isPermaLink="false">thinktank-kg:wire-014</guid>
  <pubDate>Sun, 30 Aug 2026 12:00:00 +0000</pubDate>
  <category>materials-energy</category>
  <description>Actinide announced on 26 August 2026 that it reached 15.38% uranium enrichment using a 1940s-style calutron. That is about five times the roughly 3% in commercial reactor fuel — not beyond national laboratories, which passed 90% decades ago. Receipt: Actinide press release, with DOE / Centrus January 2023 for the comparison (2026-08-26).</description>
</item><item>
  <title>[CORRECTION] Ninety minutes of generated video costs hundreds, not tens of thousands</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-015</link>
  <guid isPermaLink="false">thinktank-kg:wire-015</guid>
  <pubDate>Sun, 30 Aug 2026 12:00:00 +0000</pubDate>
  <category>ai-models</category>
  <description>At Alibaba Cloud&#x27;s published prices for its Wan3.0 video model — $0.05 to $0.20 per second, listed on 25 August 2026 — generating 90 minutes of film costs $270 to $1,080. Receipt: Alibaba Cloud — published Wan3.0 pricing (2026-08-25).</description>
</item><item>
  <title>[VERIFIED] A protein gel regrew tooth enamel in the laboratory</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-016</link>
  <guid isPermaLink="false">thinktank-kg:wire-016</guid>
  <pubDate>Sun, 30 Aug 2026 12:00:00 +0000</pubDate>
  <category>health-longevity</category>
  <description>A Nottingham-led team published a fluoride-free, elastin-like protein gel in Nature Communications on 4 November 2025 that regrows enamel on extracted teeth. The work is peer-reviewed and done outside the body, not yet in patients; the spin-out Mintech-Bio targets a product in 2027. Receipt: Nature Communications — peer-reviewed paper, Nottingham-led enamel regrowth (2025-11-04).</description>
</item><item>
  <title>[VERIFIED] No standard way exists to ask a machine for good evidence</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-017</link>
  <guid isPermaLink="false">thinktank-kg:wire-017</guid>
  <pubDate>Wed, 02 Sep 2026 12:00:00 +0000</pubDate>
  <category>machine-layer</category>
  <description>A read of the internet&#x27;s own registries on 2 September 2026 found nothing that lets one machine ask another for claims of a given evidential quality. The IANA register of HTTP header fields, updated 28 August 2026, lists fourteen &quot;Accept&quot; headers, and not one negotiates evidence quality. Receipt: IANA HTTP Field Name Registry and HTTP Preferences Registry, read directly (2026-08-28).</description>
</item><item>
  <title>[VERIFIED] Ten million machine-readable claims, one hundred thirty-nine users</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-018</link>
  <guid isPermaLink="false">thinktank-kg:wire-018</guid>
  <pubDate>Wed, 02 Sep 2026 12:00:00 +0000</pubDate>
  <category>machine-layer</category>
  <description>The Nanopublication Hub publishes single scientific claims, each with its provenance, in a form machines can read. Its live counter, read on 2 September 2026, showed 10,905,216 nanopublications and 139 users: close to the right design, almost no adoption. Receipt: Nanopublication Hub (Knowledge Pixels) — live counter (2026-09-02).</description>
</item><item>
  <title>[CORRECTION] AI did not solve Navier–Stokes. It solved the forced case, and declined the prize.</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-019</link>
  <guid isPermaLink="false">thinktank-kg:wire-019</guid>
  <pubDate>Mon, 14 Sep 2026 12:00:00 +0000</pubDate>
  <category>ai-science</category>
  <description>On 8 September 2026 OpenAI said a swarm of about 10,000 AI programmes, working for roughly four days, showed that the Navier–Stokes fluid equations can blow up in finite time — in the versions that allow an external force, not the free-flowing case the Clay Millennium Prize is written for. OpenAI&#x27;s own announcement declined the prize. The Clay Mathematics Institute has ruled nothing. Receipt: OpenAI, announcement of the Navier–Stokes result (2026-09-08).</description>
</item><item>
  <title>[SELF-REPORTED] A benchmark that cost about $500,000 to pass now costs about $20</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-020</link>
  <guid isPermaLink="false">thinktank-kg:wire-020</guid>
  <pubDate>Mon, 14 Sep 2026 12:00:00 +0000</pubDate>
  <category>benchmarks</category>
  <description>Noam Brown of OpenAI posted on 8 September 2026 that scoring 87.5% on ARC-AGI-1 cost about $500,000 of compute when o3 was announced in December 2024, and that the company&#x27;s new model scores higher for about $20. It is OpenAI&#x27;s own figure about OpenAI&#x27;s own costs. The span is about 21 months, not two years; the 25,000-fold arithmetic is the podcast hosts&#x27; and it is correct. Receipt: X / @polynoamial (Noam Brown, OpenAI) (2026-09-08).</description>
</item><item>
  <title>[SELF-REPORTED] &#x27;AGI has arrived&#x27; was a chip-seller&#x27;s congratulation; the 100,000 chips are the lab&#x27;s own count</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-021</link>
  <guid isPermaLink="false">thinktank-kg:wire-021</guid>
  <pubDate>Mon, 14 Sep 2026 12:00:00 +0000</pubDate>
  <category>ai-models</category>
  <description>On 6 September 2026 Jensen Huang, whose company Nvidia makes the chips, posted that OpenAI&#x27;s new model was trained on more than 100,000 Grace Blackwell NVL72 systems and that &#x27;AGI has arrived&#x27;. The post congratulates a customer; it is not a measurement. No independent count of the chips exists, and NVL72 is a rack-scale system, not a loose chip. The same post announced 400,000 more. Receipt: X / @JensenHuang (Jensen Huang, NVIDIA) (2026-09-06).</description>
</item><item>
  <title>[SELF-REPORTED] OpenAI says its research agents now put in 3.1 days of effort per human day</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-022</link>
  <guid isPermaLink="false">thinktank-kg:wire-022</guid>
  <pubDate>Mon, 14 Sep 2026 12:00:00 +0000</pubDate>
  <category>ai-models</category>
  <description>OpenAI&#x27;s 6 September 2026 post reports 3.1 agent-workdays of effort for every workday of human labour, up from under one earlier in the year. The number measures compute effort expended, not work completed, and it is the company&#x27;s own unaudited measurement. OpenAI says it reached its automated-research-intern goal, not that it exceeded interns. Receipt: OpenAI, &#x27;Research acceleration: a view inside OpenAI&#x27; (2026-09-06).</description>
</item><item>
  <title>[CORRECTION] 833 million people are over 65 today; about 1.7 billion by 2054</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-023</link>
  <guid isPermaLink="false">thinktank-kg:wire-023</guid>
  <pubDate>Mon, 14 Sep 2026 12:00:00 +0000</pubDate>
  <category>health-longevity</category>
  <description>The United Nations&#x27; World Population Prospects 2024, medium variant, puts the world&#x27;s population aged 65 and over at 833.5 million in 2024 and 1,700.6 million in 2054. Both figures come from one edition of one dataset. The chart shown on the podcast said 852 million rising to 2 billion by 2060. Receipt: United Nations DESA Population Division, World Population Prospects 2024 (medium variant) (2024-07-11).</description>
</item><item>
  <title>[VERIFIED] Tesla opened a form to buy driverless taxi fleets. There is no price on it.</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-024</link>
  <guid isPermaLink="false">thinktank-kg:wire-024</guid>
  <pubDate>Mon, 14 Sep 2026 12:00:00 +0000</pubDate>
  <category>robotics-mobility</category>
  <description>On 3 September 2026, the day invitation-only Cybercab rides began in Austin, Tesla put up an interest form for fleet purchasing and mobility hubs: name, company, region, and a choice of interest. It is an expression of interest, not an order book. Tesla&#x27;s standing guidance since the October 2024 unveiling — &#x27;below $30,000&#x27; — is the only price on record, and no price was confirmed at the launch. Receipt: Tesla, Robotaxi interest form (2026-09-03).</description>
</item><item>
  <title>[SELF-REPORTED] Better data beat better model design about three to one — in a blog post, not a study</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-025</link>
  <guid isPermaLink="false">thinktank-kg:wire-025</guid>
  <pubDate>Mon, 14 Sep 2026 12:00:00 +0000</pubDate>
  <category>ai-models</category>
  <description>An 8 September 2026 post by Dwarkesh Patel and Jerry Han finds that, over six years of AI progress, improvements in training data produced about 12× the compute-efficiency gain that architecture changes did (about 3.7×). The 12× and 3.7× figures match the post. It is a co-authored blog post with no venue, DOI or preprint, so the card wears SELF-REPORTED rather than VERIFIED, and the film does not call it a study. Receipt: Dwarkesh Patel and Jerry Han, &#x27;Pretraining progress is mostly data&#x27; (dwarkesh.com) (2026-09-08).</description>
</item><item>
  <title>[VERIFIED] Chinese banks are giving away AI tokens as credit-card rewards</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-026</link>
  <guid isPermaLink="false">thinktank-kg:wire-026</guid>
  <pubDate>Mon, 14 Sep 2026 12:00:00 +0000</pubDate>
  <category>markets</category>
  <description>Rest of World reported on 4 September 2026 that China Merchants Bank (12 June), Agricultural Bank of China with Moonshot&#x27;s Kimi card (10 July) and Ping An (21 July) launched cards that reward spending with AI tokens — one offering new cardholders up to 1.8 billion tokens on MiniMax&#x27;s models. A Beijing dumpling restaurant hands out token vouchers with the bill. The launches were June and July 2026; September is when they were reported. Receipt: Rest of World (2026-09-04).</description>
</item><item>
  <title>[VERIFIED] South Korea will give every citizen a free AI agent</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-027</link>
  <guid isPermaLink="false">thinktank-kg:wire-027</guid>
  <pubDate>Mon, 14 Sep 2026 12:00:00 +0000</pubDate>
  <category>policy</category>
  <description>On 4 September 2026 Korea&#x27;s Ministry of Science and ICT said free agentic AI will be offered to all citizens through messenger apps, with a beta in October 2026. Three consortia build it: SK Telecom, KT, and Kakao with LG. Receipt: Korea Times, reporting the MSIT briefing (2026-09-04).</description>
</item><item>
  <title>[VERIFIED] China Telecom sells access to 142 AI models like a data plan</title>
  <link>https://thinktank.theknowledgegardens.com/wire/#wire-028</link>
  <guid isPermaLink="false">thinktank-kg:wire-028</guid>
  <pubDate>Mon, 14 Sep 2026 12:00:00 +0000</pubDate>
  <category>markets</category>
  <description>The South China Morning Post reported on 21 August 2026 that China Telecom&#x27;s Xingchen TokenHub hosts 142 large language models and more than 420 industry applications, with plans reported from 9.9 yuan a month for 10 million tokens. The model count comes from the press; China Telecom&#x27;s own launch note gives no number. China Mobile and China Unicom run similar tiers. Receipt: South China Morning Post (2026-08-21).</description>
</item>
</channel>
</rss>
