Top Stories
3 stories + 6 briefsA judge denies xAI's motion to dismiss, and the NAACP suit over data-center turbine pollution goes on
TechCrunch reported on October 31 that a U.S. federal judge denied xAI's motion to dismiss a lawsuit by the NAACP. The Southern Environmental Law Center, suing for the NAACP, alleges that the company, now part of SpaceX, installed and ran at least 35 gas turbines and other air-pollution sources at the site powering the Colossus data center outside Memphis, without the required pre-construction and operating air permits, and in violation of the Clean Air Act.
xAI had argued that the turbines were mobile equipment and needed no permit, that the plaintiffs lacked standing, and that the citizen-suit provision is unconstitutional. Judge Debra M. Brown rejected each point. The plaintiffs have standing under the Clean Air Act. Under Mississippi's implementation plan the turbines are not a mobile source. A citizen suit is not an exercise of federal executive power. A similar motion to dismiss, filed earlier this month by the U.S. Department of Justice, was also denied. The NAACP's lead for environmental and climate justice described the ruling as another important win. TechCrunch has asked SpaceX for comment. The report notes that public feeling about data centers has turned sharply against them. New York State recently passed a pause on new data centers, and Amazon and Microsoft have promised to stop using nondisclosure agreements in talks with local governments.
Mathematician Calegari, on Terence Tao's blog: mathematics has changed, and hiring and evaluation should change with it
On November 1, Terence Tao published a guest post by number theorist Frank Calegari on his blog. The Chinese edition gives the title as mathematics having changed, and the field needing to act accordingly. The post URL carries the same sense. Calegari argues that OpenAI's release of more than 700 research papers on October 6 is a watershed in the history of mathematics, and that AI can already make breakthrough contributions in many areas. He also says the papers are badly written, do not follow the field's authorship norms, and that some are wrong. Rather than dwell on what a trillion-dollar company does, he thinks mathematicians should spend their effort where they have control: how they hire, how they publish, and how they train students.
He uses the classification of finite simple groups as an analogy. Once a famous problem is declared solved, the heat goes out of the field, but the problem itself is far from finished. He acknowledges that many people are now using AI to race to be the first to prove something, and that the ability to produce a proof is pulling away from the ability to understand the subject and move it forward. In his view, choosing the problem worth doing may matter more than producing the proof. He offers several concrete suggestions: evaluate every piece of work as if it might have been AI-assisted; hire more through face-to-face conversation, not only from the papers; require journals to adjust, and leave the ones that will not; give graduate training more breadth without dropping technical training; run better seminars; and take more responsibility for results that come from AI. In the Chinese edition's paraphrase, an optimist is told that the game has changed rather than ended.
What's new (Terence Tao's blog) | November 1, 2026
WIRED: more hikers plan with a chatbot, and search-and-rescue teams see a new risk
WIRED on November 1 republished a High Country News report. Three 20-year-old hikers called for help on Mount Shasta in California. They later said that poor preparation and a flawed plan came partly from relying on Google's Gemini. They did not bring enough food, they overestimated their fitness, they did not turn back when they missed the summit in the afternoon, and they reached the top at 7:30 p.m. They had to descend in the dark, got lost the next morning, and one of them injured a knee. A U.S. Forest Service rescuer said the group really had done the homework, and also that they had somewhat switched their judgment off. The report also describes people who camped beside a no-camping sign because an AI said they could, and people carrying an AI-generated map that placed a landmark on the other side of the mountain. In British Columbia, a 16-year-old who planned a trip with Claude was lifted off a ridge by helicopter. He later said he had also used Google Maps and AllTrails.
The superintendent of Grand Teton National Park said that over the past year the share of visitors who said they got trip information from AI rose from zero to 26 percent. It is not yet clear that AI is raising the number of accidents. Wyoming had fewer search-and-rescue calls this year than last. Chris Flathmann, a Clemson professor who studies human-AI collaboration, warns that a chatbot will confidently offer a route that sounds plausible and may come from an outdated post or from something made up, and that once a person hands over judgment it is hard to think critically again. The experts' advice is to check the latest map from the public-land agency, a trail-association site, or a guidebook before going somewhere unfamiliar, and better still to ask a guide or a ranger.
TechCrunch on November 1, citing Axios, says OpenAI is preparing a public-relations push after Tuesday's U.S. midterm elections. Chief global affairs officer Chris Lehane said the midterms will be a preview of 2028, and he expects AI to be a central issue in the presidential election. The ideas he mentioned include setting a speed limit for AI, making people feel that AI can help daily life, pushing the United States into global standard-setting, and renaming data centers "AI compute campuses". Lehane also said the argument around data centers is healthy, and that communities should hold companies responsible.
Source: TechCrunch · November 1, 2026 · The TechCrunch article; the Axios original was not read
This follows the November 1 brief. Qbitai reports that Anthropic technical staff have confirmed two sentences added to the Opus 5.5 system prompt. The sense of them is: original games, games of a given genre, and games made from a user's own design and assets are allowed; copying a specific game that already exists is not, even under another name or under a claim of originality. Qbitai's view is that the rule binds Claude itself and is not hard to route around. Developers have also complained, in Claude Code's GitHub issues, that original games of their own were misread as copies. This follows Qbitai's account. The Anthropic staff post was not read.
Source: Qbitai · November 1, 2026 · The Qbitai article; the Anthropic staff post and the system prompt were not read
TechCrunch on October 31 reports Steven Levy's WIRED interview with Apple chief executive John Ternus. Apple has often been described as behind on AI over the past few years, and a stronger Siri arrived only this summer. Ternus, who formally became chief executive in early September, said everyone has a view, and that he likes where Apple is. Apple will absolutely use AI in new products, he said, and it does not want to ship AI for its own sake. If the company gets it right, a user should not have to think about whether the thing underneath is AI or ordinary software. Asked whether agents will make apps unnecessary, he said that might make life very easy, and that making a product is a deeply human and personal thing. The reported remarks follow the Chinese edition's paraphrase of the interview.
Source: TechCrunch · October 31, 2026 · Relevant passages of the TechCrunch article and the WIRED interview; the WIRED piece is dated October 30
Intelligent Emergence, part of 36Kr, reported on November 1 that Tencent, at its Global Digital Ecosystem Summit on October 29, released Buddy One, a cross-app workplace system. It connects Tencent Docs, Tencent Meeting, ima, and Tencent's cloud storage, and it moves account, AI credits, and some service benefits across them. WorkBuddy will also open its architecture so companies can connect agents they build themselves, third-party agents, and third-party communication systems. Asked about the business, Dowson Tong, chief executive of Tencent's Cloud and Smart Industries group, gave no specific number. He said commercialization is there to support the product, not to hand a team a figure and push it to hit that figure. Partners who deploy on site for enterprise customers need a business model too.
36Kr · Intelligent Emergence | November 1, 2026
Source: 36Kr · Intelligent Emergence · November 1, 2026 · Relevant passages of the 36Kr article
TechCrunch on November 1 published an excerpt of a discussion from the Equity podcast. The prompt was Dazzle AI chief executive Marissa Mayer using an AI tool to finish a school assignment to describe her daughter. Before that, OpenAI chief executive Sam Altman had suggested parents make a personalized podcast for a child and said he could not imagine parenting without ChatGPT. Mark Zuckerberg has also said it will become more acceptable to meet social and emotional needs with AI. The hosts argue that executives have a strong financial reason to persuade people to use AI in every part of life, and that the picture they draw is not attractive and can look dystopian. Reporter Amanda Silberling said it is both a marketing problem and a product problem, and that the industry is pushing a future in which ordinary human contact keeps getting optimized away. This is a commentary.
Source: TechCrunch · November 1, 2026 · The TechCrunch excerpt of the podcast; a commentary, not a news report
Industry Insights
Conversation SpotlightJustin Solomon is an associate professor of electrical engineering and computer science at MIT. He studies geometric data processing. This conversation is from Bloomberg's business podcast Odd Lots. The Concourse record dates the episode to October 9, 2026. What follows is a paraphrase of caption excerpts the Concourse page arranged from automatic subtitles, not a verbatim English quotation.
He thinks mathematics is, to a large extent, about taste, not only about writing proofs and checking conjectures other people proposed. What is interesting is the modeling: working out which question to ask, and what insight that question can bring. That is what AI tools cannot do yet. He also says AI-generated Lean proofs can run to hundreds of thousands of lines, too long for a person to check line by line, so the check falls to the verifier. Once a proof can be produced with no friction, the detours and insights along the way that might open a new field can get cut out. This is a view from three weeks ago, not a response to Calegari's post. Both of them put the weight on what question gets asked.
Concourse · Odd Lots | October 9, 2026, historical episode
Source: Concourse · Odd Lots · October 9, 2026, historical episode · Editorial guide, timestamped automatic-caption excerpts, and the original-video link on the published Concourse page; the full recording was not reviewed
Research Briefs
1 paperWhy read it today: the xAI turbine lawsuit and OpenAI's wish to rename data centers point at the same question. What ruler should a community use for a data center? This preprint, submitted to arXiv on September 20, 2026, says that power usage effectiveness, PUE, along with water usage effectiveness and carbon usage effectiveness, each divides a facility's resource use by its IT energy use. It is not a new paper from today. Those ratios do not directly show whether local resources are tight, whether infrastructure has capacity, or what the social effects are. Data the authors cite say that in the first quarter of 2026 alone, about 130 billion dollars of U.S. projects were delayed or cancelled because of local opposition. The main worries were water, electricity demand, infrastructure capacity, and transparency, not internal efficiency.
The authors propose a local-impact audit in five groups: efficiency, water management, carbon emissions and renewable energy, compliance, and local disclosure. It is designed to be assessed each quarter and checked independently against public records. They demonstrate it on three facilities in Illinois that are in the middle of local policy disputes, and they discuss the data-access barriers to an independent check. The reading scope is the abstract. The paper has not been peer reviewed. It was selected from the Paper-Radar library.
arXiv | v1 submitted September 20, 2026; related-paper look-back
Open Source Spotlight
1 projectMemdoor is a single-binary command-line agent for programming and automation, under the Apache-2.0 license. Its example workflow team-solve writes a paper's method as eight YAML files, with no code and no model specified: a planner, two idea generators, a solver, and a manager, plus a repair role. Each starts from a fresh context and talks only through files on disk. A script then runs the samples and decides success or failure. On the same model, Qwen3.8-27B, the author took the first 10 problems of the original paper's LiveCodeBench Hard list and ran each five times. The workflow averaged 9.8. The original authors' framework averaged 9.6. In the paper's own data, a single Fable 5 call scored 8.0. Fifty runs cost 35.79 dollars in total, with a median of about 40 minutes each.
The author also states the limits: only 10 problems were tested, not the paper's 100, and 9.8 against 9.6 is inside the error range, so the claim is only that the workflow matches the original framework. The problems are from spring 2025, and the model may have seen them in training. The Fable 5 column comes from the original authors' file and was not rerun in this framework. The example was already public in the repository in late October and drew discussion on Hacker News on October 31. This issue did not run the code.
GitHub · guregodevo/memdoor | example first committed October 25, 2026, tool discovery
Coverage notes · Incomplete · 43/50 checks recorded
Coverage is incomplete. This issue reused the 05:00 upstream receipt and checked the final hour before 06:00. Fifty sources were planned. In the final window, 41 were complete, 9 partial, 0 failed, and 0 still awaiting a check. Core sources that were not complete: xAI and arXiv. The xAI news page only yielded cards from before the window. Browser sources such as DeepSeek and ByteDance Seed could not be rechecked in the final hour after 05:00 and are not written as having no updates. arXiv was only partly covered. This issue did not adopt a new paper from the day. The paper section is a related-paper look-back, and only the abstract was read. The OpenAI Help Center original could not be opened, and the related brief follows a secondhand account.
The source catalog has 22 items, screened on the 36-hour window from 18:00 on October 31, deduplicated against the previous seven public catalogs by normalized link, and merged where the event is the same. Most catalog entries are title-level. They are not a full-text check or an endorsement. Hitting a count does not mean coverage is complete. The paper look-back, the open-source tool, and the conversation keep their original dates and are not counted in the 36-hour catalog.
Coverage gaps reflect incomplete checks; they do not establish that a source had no updates.
- xAI — Partially checked. Marked partial in the collection receipt. Only cards from before the window were read, so this is not treated as no update.
- Qwen — Partially checked. Marked partial in the collection receipt. A partial page check is not treated as no update.
- StepFun — Partially checked. Marked partial in the collection receipt. A partial page check is not treated as no update.
- arXiv cs.AI / cs.LG / cs.CL — Partially checked. Marked partial in the collection receipt. Coverage of the listing was incomplete, so it is not treated as no update.
- OpenReview — Partially checked. Marked partial in the collection receipt. The check was not complete, so it is not treated as no update.
- MiniMax·News & Updates — Partially checked. Marked partial in the collection receipt. The check was not complete, so it is not treated as no update.
- Hacker News — Partially checked. Marked partial in the collection receipt. The front-page sample does not represent the full window.
Reviewing a source does not constitute independent testing. Paper-reading scope is stated with each item; experiments were not reproduced.
Sources & discoveries 22 links
Titles are retained from the Chinese edition; source names are shown in English. This discovery list includes material not read in full.
News · 21 links
- 法官驳回xAI撤诉动议，NAACP就数据中心燃气轮机污染的诉讼继续审理Judge allows NAACP lawsuit over xAI data center pollution to move forwardTechCrunch
- 数学家Calegari在陶哲轩博客撰文：数学已经改变，评价和招聘方式也要跟着改Mathematics has changed. Let’s act like itWhat's new (Terence Tao's blog)
- WIRED：越来越多徒步者靠聊天机器人做规划，搜救人员看到新的风险Chatbots Are Leading Hikers Astray, With Dangerous ConsequencesWIRED
- 据机器之心：ChatGPT与Codex额度将自12月起合并，统一按token计量Tibo 28连更后，OpenAI合并额度池，计费聊天开始了36Kr AI (article-level checks for original and syndicated reporting)
- 据Axios：OpenAI准备在美国中期选举后发起公关攻势，考虑把数据中心改称“AI算力园区”OpenAI might rebrand its data centers as ‘AI compute campuses’TechCrunch
- 续报：Anthropic人员证实在Opus 5.5系统提示中加入“勿复刻现有游戏”的规则Claude紧急叫停！暗改Harness禁止逆向游戏Qbitai
- 苹果CEO特努斯：不想“为了AI而做AI”Apple CEO says the company doesn’t ‘want to ship AI for AI’s sake’TechCrunch
- 汤道生谈WorkBuddy商业化：收入用来支撑产品与伙伴，不给团队压数字汤道生谈WorkBuddy商业化：不靠收入数字催团队36Kr AI (article-level checks for original and syndicated reporting)
- TechCrunch Equity播客：AI高管谈AI育儿、AI约会，为何总显得脱离常人AI can make tech CEOs’ lives easier. What about everyone else?TechCrunch
- 引述 Kyle KingsburyQuoting Kyle KingsburySimon Willison's Weblog
- 三年前布局，今年爆火丨西湖大学成立新中心Qbitai
- 新入口和旧城墙之下，“中国版Muse”有多远？36Kr AI (article-level checks for original and syndicated reporting)
- 2800万用户，养不活一个AI陪伴应用36Kr AI (article-level checks for original and syndicated reporting)
- Agent改写算力账本，CPU不再是AI服务器的配角？36Kr AI (article-level checks for original and syndicated reporting)
- 算力的尽头是电力？从第一性原理看AI基础设施36Kr AI (article-level checks for original and syndicated reporting)
- Meta拟筹近千亿美元，甲骨文报出了“最后一笔”36Kr AI (article-level checks for original and syndicated reporting)
- 清华姚期智团队新作：代数世界模型，让AI像牛顿一样思考Qbitai
- 老牌便利店机器人售货折戟：实验室锤炼上万次，货架前还是摔瓶子Qbitai
- Jev新成果，再次引爆开源Qbitai
- 跟AI谈恋爱，年轻人图什么？36Kr AI (article-level checks for original and syndicated reporting)
- 我们如何接管了 ChatGPT 账户How We Took Over ChatGPT AccountsSimon Willison's Weblog
Open source and releases · 1 links
- 开源项目 datasette-llm 0.1a9datasette-llm 0.1a9Simon Willison's Weblog