Shinome Brief · 2026-11-05

AI & Tech

6 stories

The lead

Kimi K3.1 Pro ships open weights, and open-weight labs now carry 75 percent of OpenRouter traffic, with DeepSeek alone at 55 percent

Moonshot reports 94.2 on GPQA-Diamond and 75.5 on DeepSWE at maximum reasoning effort; a LocalLLaMA thread already argues the scores are tuned to the benchmarks. The traffic split is the harder number: on the main paid router, most tokens now run on models anyone can download.

Moonshot published the Kimi K3.1 Pro weights: a 2.8 trillion parameter mixture-of-experts model with 104 billion active, a 1 million token context and a custom licence. On 4 November, DeepSeek took 54.9 percent of 25.9 trillion OpenRouter tokens, open-weight labs 75.1 percent, OpenAI 10.4 and Anthropic 8.9.

Hugging Face / dirac.run

CONVOIn an interview on AI cost, say it plainly: on OpenRouter three tokens in four now go to open-weight models, so the price of a capable model is heading to the cost of compute.

1
Anthropic releases FlashProve, an open-source prototype that proves which model produced each answer
A provider commits to its weights once, then attaches a zero-knowledge proof to each response. On Llama 4 Maverick a single H100 proves about 110 tokens a second, at 70 to 95 times the GPU time of normal serving, 19 MB per 1,024-token request and a 1,024-token cap.
Anthropic Research
JOBUse it in a compliance or KYC interview as the example of audit moving into AI: a customer could one day check that a vendor really ran the model it billed for.
2
Uber says third-party AI agents booked four rides and about 500 Eats orders in the third quarter
Uber executives gave the figures on the earnings call to answer the fear that AI agents would cut it out of the customer relationship.
The Verge
CONVOWhen someone says agents will replace apps next year, answer with Uber's number: four rides in a quarter.
3
Voters back data centre limits in the midterms, and Big Tech's 600 million dollar campaign spend falls flat
The FT reports that anger at AI and Silicon Valley fed Democratic support despite about 600 million dollars of tech campaign money; The Verge counts winning candidates and ballot measures that favour moratoriums or new hurdles for data centre building.
FT / The Verge
CONVOPair it with this morning's Asian chip slide in a recruiter call: the AI build-out now carries political risk, and markets priced it in a day.
4
Ricursive's Ripple-65B Preview scores 80.5 on TerminalBench 2.1, ahead of open models eight to fifteen times its size
The 65 billion parameter dense model edges Reflection Beam 501B (80.1) and Meta Muse Spark Open 816B (76.4); the page does not state a licence.
Ricursive Intelligence
CONVOKeep for interviews on AI efficiency: a 65B model matching 500B+ ones is the clearest current case that size is no longer the moat.
5
ChatGPT now makes animated GIFs from a prompt or a photo, with the new Images 2.6 model
Users can set animation speed, edit single frames and share the result; before this, ChatGPT could not make a GIF in one step.
The Verge
SHINOMETest it on one Shinome exercise cue this week: a looping GIF of a hinge or squat brace for the client PWA, instead of filming a video.

Finance & Markets

7 stories

The lead

Federal Council lifts the LPP minimum interest rate from 1.25 to 2 percent, above the LPP commission's recommendation

The minimum rate is what pension funds must credit on mandatory savings, so a jump of 75 basis points lands on every fund's liabilities at once. The government went further than its own expert commission, which is the part senior people will argue about.

Le Temps calls it the largest rise in the history of the second pillar. Unions welcome it; employer organisations are angry and warn about the financial balance of pension funds.

Le Temps

JOBHave one line ready for any Swiss finance interview this month: the 2 percent floor pushes pension funds toward riskier assets or lower conversion rates, and you can say which side you would take.

1
Sweden's Ahlsell bids CHF 492 million for Zurich building-tech group Meier Tobler, which will leave the Swiss exchange
The board and the majority shareholder back the public offer, so the delisting looks likely.
Le Temps
JOBAdd it to your Swiss deal list for interviews: a foreign strategic buyer taking a listed Swiss mid-cap private, the same week Lonza bought five US sites from Merck KGaA.
2
Asian chip stocks slide on fears a Democratic Congress will slow the data centre build-out
The slide follows Micron and other US chip names falling on Wednesday as election results hit data centre optimism.
FT
CONVOSay it as one chain in a market chat: midterms, then data centre permits, then chip orders, then Asian suppliers.
3
Oil falls below 100 dollars on post-election peace hopes, as Vance blames the Iran war for the Republican defeat
Brent was at 117 dollars on 1 November; the drop reflects bets that a Democratic Congress pushes Washington toward ending the war.
FT
JOBUse the move from 117 to under 100 in a week as your example of how fast political risk reprices commodities, useful for Geneva trading-house interviews.
4
Federal audit office finds several hundred firms that took covid loans still paid banned dividends in 2025
Firms backed by federal guarantees in 2020 may not pay dividends while the loan runs; the Contrôle fédéral des finances calls the cases marginal overall.
Le Temps
JOBGood for compliance and KYC interviews: a concrete Swiss case of a covenant that firms ignore and an auditor that catches it.
5
Qualcomm turns to data centres as smartphone sales fall about 20 percent from 2025
On its fiscal fourth-quarter call Qualcomm set out data centre plans built on partnerships with Amazon and Meta; MarketWatch reports the stock fell after the results.
The Verge
CONVOPair it with AMD's 64 percent data centre share from yesterday: the chip story is one market growing while the phone market shrinks.
6
NextDC shares jump 43 percent after Goodman and Digital Realty move to buy up to 9.9 percent
The Australian data centre operator rose more than 40 percent on Thursday morning as two industry players sought stakes on the same day US investors sold the sector.
CNBC
CONVOUse it to show the split view on data centres: strategic buyers still pay up while public markets sell.

Fitness & Performance

6 stories

Last 7 daysTraining science this week keeps saying the fine print of a programme is poorly tested: an umbrella review finds almost no solid evidence on how close to failure to train, while the industry news is strength going mainstream as a product category.

This quarterAcross 90 days of briefs, prescription details keep coming back as second order: load range (4 November), protein source (13 October) and meal timing (14 and 19 October) mattered less than deficit, total protein and showing up; the business side moved from Pilates and boutique growth to strength as the headline offer (Apple Gym Strength, VASA, new strength-tech startups).

The lead

Umbrella review of 33 reviews: 30 rated critically low confidence, and no solid answer yet on training to failure or cluster sets

The review searched seven databases up to 17 September 2026 and kept only direct comparisons in healthy adults with at least six weeks of training for long-term outcomes. Its conclusion is that choices on failure, rest within sets and velocity loss remain conditional on the population and the programme studied.

Non-failure sets produced less fatigue after sessions; some hypertrophy estimates favoured failure, but the authors say preserved adaptation without failure was not established and none of 13 acute-plus-long-term combinations supported a joint reading.

Life (MDPI)

supportssimple-systemThe literature cannot separate these set-design options reliably, so adding them to a client plan adds decisions without proven return.

SHINOMEUse it to keep the Shinome programme simple: no client needs a failure-versus-reserve debate when the best evidence on it is rated critically low.

1
Crossover trial in 15 young adults: low-GI breakfasts gave less fullness than high-GI ones, and the low-carb low-GI breakfast left people hungriest
Four breakfasts at 45 or 65 percent of energy from carbohydrate, each low or high GI, were tracked over 240 minutes; ghrelin, GLP-1 and GIP did not differ between them. Exploratory secondary analysis.
Nutrients
sharpensprotein-satietyFullness tracked the GI and carb mix, not the gut hormones measured, in only 15 people; too small to lean on alone.
SHINOMEFile it with the potato-first rule as a small data point: a high-GI carb was not the less filling option here, which is what clients often assume.
2
Meta-analysis in JOSPT: resistance training may cut low back pain and disability at 3 months against usual care, at very low certainty
Pooled effects were -1.33 for pain and -1.75 for disability against no intervention or usual care; against other active or passive treatments the evidence was very uncertain. Only 25 percent of programmes were reported well enough to copy.
Journal of Orthopaedic & Sports Physical Therapy
supportsmuscle-is-capitalMeasured benefit of lifting beyond body composition, with the caveat that certainty is very low.
SHINOMEWhen a prospect says back pain stops them lifting, you can say trials measured less pain with strength work, without promising a result.
3
Newmark: none of the largest US fitness chains it tracks closed sites this year, keeping 8.1 to 11.8 million square feet open
Landlords now court gyms as tenants; health and beauty locations nearly doubled since 2018, growing 8.4 percent a year.
Athletech News
SHINOMEUse the number in the Shinome pitch deck: physical gyms are stable, so online coaching sells as a layer on top of a gym, not a replacement.
4
Strength-tech startups Suji, Kyla and Fort target workout tracking and recovery as strength training grows
Three young companies are building tools to help people optimise strength workouts and recovery, a week after Apple added adaptive strength plans to Fitness+.
Athletech News
SHINOMELook up each of the three apps and note one feature the coach app does not have yet, before the next build session.
5
HOTWORX CEO on reaching 913 locations by refusing complexity: lean labour and one clear member promise
Stephen Smith says repeatable growth came from keeping operations simple rather than adding features.
Club Solutions
SHINOMEWrite the Shinome member promise in one sentence this week and cut any app feature that does not serve it.