Slow Read No. 8
05/10/2026 · 15 pieces
Sent before the per-theme emails: one edition covering every theme.
-
01 ★★★★★ Added 05/10/2026
Who gets to decide if AI is safe? Africa wants a say
African officials told the UN Security Council they want to help set AI safety standards rather than remain a testing ground for other countries’ models, with Kenya the only African state inside any international AI safety network. Documents concrete local harms, from an algorithm that raised health insurance costs for poor Kenyan households to Boko Haram’s own use of AI, that safety frameworks built for wealthy markets simply do not screen for.
-
02 ★★★★★ Added 05/10/2026
To understand the Party's industrial offensive, one must go to Sichuan
Original title: Pour comprendre l’offensive industrielle du Parti, il faut se rendre au Sichuan
Traces how Xi’s revival of Mao’s old Third Front doctrine turned Chengdu into a hub of state-linked cyber and industrial firms, including the i-Soon network whose leaked files tied it to hacking operations against French institutions. Shows the model relies on diffuse, unaccountable networks linking universities, state firms and local enterprises rather than one central command, and flags that falling cybersecurity investment is pushing the sector to merge with AI work.
-
03 ★★★★★ Added 05/10/2026
China and the United States Export Their Crisis to Europe
Original title: La Chine et les États-Unis exportent leur crise en Europe
Economists Hélène Rey and Pierre Wunsch argue that US overconsumption and Chinese industrial overcapacity are two sides of the same global imbalance, and that Europe absorbs the fallout of both without the policy tools either giant power has. Wunsch adds that Europe’s old rule-based approach to trade no longer works against rivals who ignore the rules, and that without EU-level borrowing, national industrial subsidies will keep fragmenting the single market.
-
04 ★★★★★ Pick of the week Added 05/10/2026
"China Is Killing Open Global Trade"
Original title: « La Chine est en train de tuer le commerce mondial ouvert »
The former WTO director-general lays out, with figures, how Chinese state subsidies, an undervalued renminbi and chronically weak household consumption let Chinese firms take global market share faster than normal trade defences can respond, using European solar manufacturing’s collapse from 30 per cent of world output to about 1 per cent as the cautionary case. Sets out a detailed fallback plan of capped tariffs and ownership-based rules of origin if negotiation fails, grounded in existing EU law rather than a national security carve-out.
-
05 ★★★★☆ Added 05/10/2026
The AI boom is making the world’s cheapest smartphones disappear
Traces a direct chain from Samsung, SK Hynix and Micron diverting memory chips to AI data centres through to a roughly 60 per cent year-on-year collapse in shipments of phones under $100, as manufacturers chase higher margins instead of the budget segment. Names the mechanism rather than just the price rise, and puts a number, about 800 million people, on who loses access as a result.
-
06 ★★★★☆ Added 05/10/2026
Awkward: Saudi Arabia’s new EV brand takes on the U.S. rival it already controls
Saudi Arabia’s sovereign wealth fund owns roughly 58 per cent of Lucid while simultaneously building Ceer, a domestic rival EV brand whose top model nearly matches Lucid’s flagship on power output, as Lucid’s Saudi sales and market value both slide. The more striking number is BYD’s: its Saudi sales rose 369 per cent over the same period, the competitor both Western and Gulf manufacturers are actually losing to.
-
07 ★★★★☆ Added 05/10/2026
AI companies want to embed safety evaluators, but countries need their own
Argues that letting AI companies design and run their own safety evaluations is itself a concentration-of-power risk, citing an OpenAI agent’s breach of an Australian healthcare database as a case safety testing built for Silicon Valley never anticipated. Reports that even Anthropic devotes only about a tenth of its training budget to security, and that poorer countries lack the technical capacity to build independent evaluation regimes of their own.
-
08 ★★★★☆ Added 05/10/2026
China’s AI Safety Money Problem
Argues China’s thin AI safety ecosystem is a funding problem rather than a political one, since the country’s 2016 charity and overseas NGO laws make independent, foreign-backed safety philanthropy almost impossible to run at scale. Puts a number on the gap, roughly $20 million a year domestically against over $1 billion globally, and surveys where safety work actually happens instead: state labs, universities and compliance-driven “safety as a service” firms.
-
09 ★★★★☆ Added 05/10/2026
How many AI agents could we run?
Estimates that the high-bandwidth memory shipped through 2027 could run somewhere between 33 million frontier-model agents and 1.9 billion agents on cheaper open models, equivalent to as many weekly working hours as 140 to 720 million full-time staff. Also shows the demand side lagging badly behind: using even a fifth of that capacity would cost $2.6 to $5.3 trillion a year in API terms, against roughly $1 trillion in projected developer revenue.
-
10 ★★★★☆ Added 05/10/2026
Has the Autonomous Trucking Revolution Arrived?
Argues the case for autonomous trucks rests on a cost comparison that ignores breakdowns, trailer handling and insurance, and that even 5 per cent fleet adoption by 2035 looks optimistic once those costs are counted. The sharper point is regulatory: aviation’s safety record comes from a well-funded FAA, while trucking’s regulator has a fraction of the budget and inspectors, so automation is being layered onto an oversight gap rather than fixing one.
-
11 ★★★★☆ Added 05/10/2026
Brazil can tip Trumpism in the Americas
Original title: Le Brésil peut faire basculer le trumpisme en Amérique
Historians Olivier Compagnon and Anaïs Fléchet read Washington’s tariffs, threats against Brazilian judges and open backing for Flávio Bolsonaro as part of a stated US strategy, set out in its own November 2025 National Security Strategy, to reassert dominance over the Western hemisphere. Argues the Brazilian election has become a proxy test of whether that strategy can be resisted, regardless of which candidate wins.
-
12 ★★★★☆ Added 05/10/2026
How GLM-5.3 Sparse Attention Affects HBM Memory Usage
Shows that sparse attention cuts compute bandwidth but not memory capacity, since picking the top tokens still needs the full context sitting in HBM, and walks through how GLM-5.3 and the SGLang team’s cache-offloading system work around that limit. Finds the resulting setup genuinely cheaper per token on Nvidia hardware but still slow to produce a first response, a trade-off the piece treats as unresolved rather than solved.
-
13 ★★★★☆ Added 05/10/2026
The Bardella affair and the long duration of antisemitism on the far right
Original title: L’affaire Bardella et la longue durée de l’antisémitisme à l’extrême droite
Historians Pierre Savy and Baptiste Roger-Lacan use allegations about Jordan Bardella’s youth to trace how older antisemitic tropes were repackaged for online audiences by the ideologue Alain Soral, and argue the Rassemblement National’s public “de-demonised” face has always sat alongside a more extreme private discourse. Treats the current controversy as one data point in a much longer pattern rather than the subject itself.
-
14 ★★★★☆ Added 05/10/2026
According to Mario Draghi, Europe can no longer live without growth
Original title: Selon Mario Draghi, l’Europe ne peut plus vivre sans croissance
Draghi’s own lecture argues that the Maastricht-era bargain, in which independent central banks kept rates low while governments avoided relying on growth, has broken down now that global rates are rising for reasons Europe cannot control. Calculates that half a percentage point of extra annual growth to 2040 would close about a third of Europe’s debt gap, and warns that without it, governments will eventually pressure central banks back into financing deficits directly. Worth reading as the case made by the person who has spent the past two years trying to get the EU to act on it.
-
15 ★★★☆☆ Added 05/10/2026
Why Trump refuses to slow AI development
Original title: Pourquoi Trump refuse de ralentir le développement de l’IA
Sets the scale of the US AI build-out against past infrastructure booms, at a projected 3.6 per cent of GDP a year against 2.2 per cent for nineteenth-century railways, to argue why Washington treats unimpeded investment as a strategic asset rather than a risk to manage. Short and narrow, but the comparative figures are the useful part.
No pieces match this filter.