Shinome Brief

Monday 2 November 2026

Private edition · 9 min read

Web search returned nothing past early October again, so the brief is built from the feeds and primary pages (CNBC, Tao's blog, nolanlawson.com, PubMed). Le Temps blocked the Leclanche interview, so that item uses the feed text. Fitness runs short on purpose: the week's other papers were in mice, rats or cell culture, and the two Athletech business pieces worth running are paywalled.

AI & Tech

5 stories

The lead

Guest post on Terence Tao's blog: after OpenAI released over 700 research papers on 6 October, choosing problems may matter more than proving them

The post sits on one of the most-read maths blogs and drew 286 comments on Hacker News in a day. It is the clearest statement yet from inside a top field that the job moves from answering questions to choosing them.

Mathematician Frank Calegari argues that OpenAI's release of over 700 research papers on 6 October changed the field, and that the scarce skill is now picking worthwhile problems, not producing proofs. He proposes treating all work as possibly AI-assisted, interviewing students and postdocs in person, and pushing journals to change their practices.

Terence Tao's blog (guest post by Frank Calegari)

CONVOInterview line for any finance role: when models and memos get cheap to produce, the value moves to picking the question the CFO needs answered, and he can say that is how he worked on the INJAZ sell-side process.

  1. 1

    JPMorgan and Amazon block personal AI agents such as Meta Muse; an audit of 51 US retailers finds 37 block them only by accident

    Jamie Dimon says JPMorgan blocks outside AI agents acting for retail clients on security grounds, and Amazon asked Meta to remove it from Muse. Of 51 large retailers audited, 11 took a deliberate position and 37 block agents through old scraper rules; Walmart opens its site to agents, and Accenture finds 82 percent of bank customers want to approve every AI action.

    CNBC

    JOBFor KYC and compliance interviews in Geneva and Vaud: who authorises an agent to act for a client is the next onboarding question, and JPMorgan blocking while Walmart opens up gives him both sides in one sentence.

  2. 2

    Same app built by Claude Opus 5.5 on 12 web frameworks: median cost ran from 2.67 dollars on Hono plus htmx to 4.97 on Angular

    Lawson had the model build one app 5 times per framework and tested each against a conformance suite and Lighthouse. Token use ran from 3.85 million on Nuxt to 9.76 million on Angular, and 2 of 5 Angular runs produced no working app.

    Nolan Lawson

    SHINOMEBefore the next big agent build of the Shinome client PWA, check its stack against this table: framework choice alone nearly doubled the cost per build in his test.

  3. 3

    Kyle Kingsbury: LLMs make the signals of mature software and accurate documentation easier to counterfeit

    The Jepsen author, quoted by Simon Willison, says LLM use erodes trust in documentation, software and the organisations behind them, and that engineers may start treating any new project with suspicion.

    Simon Willison

    JOBWhen a recruiter asks how he uses AI in finance work, answer that AI makes polished output cheap, so checking the numbers against the source becomes the part he is paid for.

  4. 4

    South Korea's 2027 budget puts 21.3 trillion won into chips, physical AI and data centres, about double 2026, and 4.7 trillion into a sovereign model

    Total spending rises 12.8 percent to 820.9 trillion won, and the plan aims for a sovereign model of the 'world's highest performance' plus 2,000 chip design specialists. Sam Altman made an unannounced visit to Samsung and reportedly SK Group over the weekend.

    CNBC

    CONVOA sharp number for networking calls: Seoul doubles state AI spending while Leclanche in Yverdon says Swiss industry gets no state support at all.

Finance & Markets

5 stories

The lead

AI sell-off eases after the Nasdaq's worst month in over a year, ahead of Apple's results today and the US midterms on Tuesday

The week stacks three tests in two days: whether the AI trade stabilises, how Apple handles memory costs, and whether affordability swings the midterms. Bond funds paused positioning before the vote, as Sunday's brief noted.

US tech futures point higher after October's rout. Apple reports on a rare Monday, its first quarter under new CEO John Ternus, with memory prices inflated, and US petrol prices are set for a record high under Trump on the eve of Tuesday's vote.

Financial Times

CONVOOpening line for any call this week: October was the Nasdaq's worst month in over a year, and Apple tonight plus the midterms tomorrow decide whether that was a pause or a turn.

  1. 1

    Meta's ad revenue grew 26 percent to 63.1 billion dollars while Google Search grew 12 percent, the widest gap since 2021

    Google Search reached 63.4 billion dollars in the third quarter, slowing even as queries hit a record. Meta credits AI ad-targeting tools, while chatbots such as ChatGPT and Claude pull users away from search links.

    CNBC

    CONVOClean example of AI as winner and loser in one industry: same ad market, a 14-point growth gap, ready for any interview question on where AI shows up in the numbers.

  2. 2

    Fairfax's 5.5 billion dollar purchase of over 60 percent of India's IDBI Bank meets a strike and political opposition

    Fairfax bid 81 rupees a share against a fair value of 75.99, and the deal would be the largest foreign investment in India's financial sector once approved. The employees' union called a two-day national strike on 1 November, and Rahul Gandhi says the sale breaks a pledge to keep 51 percent state ownership.

    CNBC

    CONVOA live case of what sits between signing and closing: price above fair value, union strike, political pledge. He knows from INJAZ that a found buyer is not a closed deal.

  3. 3

    Leclanche's management says competitors get massive state support and Swiss battery makers get none, after covering two months of late salaries in mid-October

    The Yverdon (VD) battery maker is in its second major cash crisis: management is chasing liquidity while, it says, demand is strong and investors would fund growth. It only just paid two months of overdue wages in mid-October.

    Le Temps

    CONVOVaud talking point for Romandie interviews: a firm with demand and willing investors still runs out of cash, which is the treasury and working-capital case a CFO-mission profile can speak to.

  4. 4

    Rubio says Trump wants to end the Iran war peacefully, as reports say both sides are preparing strikes after the midterms

    Gulf diplomats warn a renewed round would hurt more than earlier ones, the FT reports. It lands with Brent at 117 dollars after a 10 percent weekly rise.

    CNBC

    CONVOKnow the date risk: Wednesday, after the vote, is when the reports place the next escalation, so oil and the franc can move just as the midterm results come in.

Fitness & Performance

3 stories
Last 7 daysThe week's diet evidence kept pointing at the deficit itself: probiotics added to calorie restriction bought a third of a kilo of fat across 16 trials, a 36-study review tied diabetes remission to percent weight lost and not to exercise or fasting style, and a 17-cohort review found eating timing and frequency gave inconsistent cancer links. On the training side, extra dairy protein added nothing to home training (1 November).
This quarterSince July the briefs have built a list of inputs that stop mattering once intake is controlled (keto on 13 August with its 1.49 kg exception, protein source on 1 and 19 August and 13 October, eating windows on 10 and 15 October, 5:2 on 2 and 29 October, the ISSN stand on 14 October), and today's probiotic review joins it. The remission review fits the RESET trial on 31 October, where 54 percent remission came with a large deficit; two-meals stays the least tested pillar in trends.py at 10 tags.

The lead

Meta-analysis of 16 trials in 1,343 adults: probiotics on top of calorie restriction added no weight loss and 0.34 kg of fat loss

Deficits ran from 300 to 1,000 kcal a day over 3 to 20 weeks, so the comparison isolates the supplement against the same restriction. The authors call the fat mass gap small and of uncertain clinical relevance.

Against calorie restriction alone, adding probiotics changed body weight by minus 0.41 kg with a confidence interval crossing zero, and had no effect on BMI, body fat percentage, waist or fat-free mass. Fat mass fell 0.34 kg more, which held after removing high-risk-of-bias trials.

Nutrition Reviews (PubMed)

supportsenergy-balanceAt matched restriction the add-on moved weight by less than half a kilo, within noise.

SHINOMEAdd to the shinomelab.com supplement FAQ: 16 trials, no extra weight loss, a third of a kilo of fat, reported as measured and with no gut bacteria explanation.

  1. 1

    Review of 36 studies: more weight lost tracked higher odds of type 2 diabetes remission; exercise and intermittent fasting did not

    Each percentage point of total weight lost went with higher odds of remission (OR 1.05), while longer disease duration, higher HbA1c and more medications went with lower odds. Physical exercise, intermittent fasting, sex and ethnicity showed no association.

    Frontiers in Endocrinology (PubMed)

    supportsenergy-balanceWeight lost, not exercise or fasting style, was the modifiable factor that tracked remission.

    SHINOMEFile next to the 31 October RESET trial in the energy-balance notes; clients with diabetes stay with their doctor, and this is reading, not a client claim.

  2. 2

    Review of 17 cohorts: eating timing, windows and meal frequency show inconsistent links with cancer, at very low quality

    Results for eating duration and night eating varied across studies, and a 3-study pooled estimate for meal frequency and colorectal cancer was not significant (HR 0.81, 95% CI 0.64 to 1.03). Less frequent breakfast and irregular eating times linked to higher cancer mortality in some cohorts only.

    Journal of Nutrition (PubMed)

    sharpenstwo-mealsNo reliable signal for or against meal frequency on cancer; breakfast skipping links appear in some cohorts and need a caveat.

    SHINOMEWhen a prospect asks whether two meals a day or an eating window is risky long term, the honest answer from 17 cohorts is that the evidence is inconsistent and observational.