Been a very immersive few weeks, so not written a newsletter since the last issue. To make up for that, and to keep my markets chat sharp, here’s my take on a tricky dilemma that tech retailers face amid the vertiginous memory boom. And what’s underpinning the strange sell-off in memory manufacturers amid their record earnings. Also, going to try a fun headline-to-headline structure that flows towards an overall argument, rather than different sections for Governance, Business/Markets, as in previous newsletters.

AI’s tightest bottleneck: Memory chips.

https://www.dbresearch.com/PROD/IE-PROD/PDFVIEWER.calias?pdfViewerPdfUrl=PROD0000000000631087&rwnode=REPORT

By now, I’d say it’s settled that the ongoing explosion in memory stocks represents a structural shift in a historically cyclical industry. Historically, memory has been a cyclical commodity. Since memory chips are pin-compatible (you can swap a certain, say, SK Hynix DRAM chip with one made by Samsung), manufacturers working with memory chips can rapidly change suppliers to arbitrage the lag between supply and demand dynamics. So, the memory market would move in cycles where undersupply increased prices which spurred greater manufacturing which then decreased prices.

But as demand for specific high-bandwidth memory (HBM) for AI is spiralling, a manufacturing bottleneck has been reached. Producing one bit of HBM requires approximately three times the silicon wafer capacity compared to more broadly used types of dynamic random-access memory (DRAM), creating a production squeeze that’ll probably not be overcome until after 2027 given the present rate of fabrication facility buildout. The Big Three manufacturers — SK Hynix, Micron, and Samsung — account for 90% of DRAM market share and hold extreme pricing power. Overall DRAM contract prices are up 90-95% quarter-on-quarter by some estimates, and predicted to increase a further 58-63% during Q2 of this year. These manufacturers have shifted to locking in multi-year contracts with buyers, thereby locking in this higher price floor.

This is a windfall for manufacturers, but bad news for electronics retailers who have to fight over the remaining 30% (!) of global memory after data centres consume the rest…

Apple stock plunges after raising prices on MacBooks, iPads amid global memory crisis.

https://uk.finance.yahoo.com/news/apple-stock-plunges-after-raising-prices-on-macbooks-ipads-amid-global-memory-crisis-140406938.html

To preserve its gross profit margins (the percentage of revenue remaining after subtracting the cost of goods sold), Apple has increased the prices of its full rang of Macbooks and iPads by around 20%. This wiped $263bn from its market capitalisation, its second-biggest loss in one day. That a company with the pricing power of Apple would inflict this hike on their customers represents the scale of the memory crunch. As Apple seeks to support Apple Intelligence on its upcoming devices, limiting them to devices with a minimum of 12GB RAM (the iPhone 17 has 8GB of RAM), these memory-driven hikes will continue.

Faced with these prohibitive memory prices, no wonder that…

Apple seeks to buy memory chips from blacklisted Chinese company.

https://archive.ph/gmsp3#selection-1477.0-1477.64

Apple has been lobbying the Department of Commerce for informal approval to buy memory chips from CXMT, a Chinese manufacturer placed on a blacklist for its alleged ties to the People’s Liberation Army and the broader Chinese military-industrial complex. Naturally, China hawks like Secretary of State Marc Rubio direly warned CEO Tim Cook against doing so.

But what much coverage of this — including this FT article — missed is that…

Memory chip vendors debunk rumours of cheap CXMT DDR5 pricing.

https://www.pradeepdive.com/blog/memory-chip-vendors-say-cxmts-ddr5-prices-match-samsung-sk-hynix-and-micron-and-it-has-a-supply-advantage-in-client-markets-as-it-is-not-prioritizing-hbm-hassan-mujtabawccftech/

Despite China’s industrial model of vicious price undercutting, various industry sources have confirmed that CXMT’s memory chips are at cost parity with the Big Three’s. Since CXMT isn’t sacrificing conventional DRAM to manufacturer HBM like its competitors, its abundance of DDR5 — the second-most powerful system memory after HBM in terms of raw bandwidth and speed — gives it a supply advantage. But this hasn’t yet translated into a discount, as CXMT’s less efficient fabrication processes generate less yield (the percentage of usable chips per silicon wafer) and high domestic and international demand give no incentive to lower prices.

If there’s record demand amid critical shortages and no price undercutting from China, you’d expect the Big Three’s share prices to be exploding. Particularly after Micron’s earnings report on Wednesday revealed that its revenue increased by 346% YoY and its gross margins expanded to 85%, far surpassing those of some of the biggest tech companies like Nvidia’s 74%.

But last week saw multiple spasms from memory stocks, with Samsung and SK Hynix falling 12.31% and 12.47% during Tuesday’s session. Even after Micron’s share price catapulted by 15% on Thursday after Wednesday’s earnings report, it slid 4% on Friday.

Commentators were — as ever — quick to wail about the AI Bubble popping amid unsustainable capex and increasing prices of critical inputs like memory. But take a deeper look and it appears that…

South Korea leveraged ETF crisis sparks global chip selloff.

https://uk.finance.yahoo.com/news/south-korea-leveraged-etf-crisis-125014555.html

After South Korea’s Financial Supervisory Service Governor approved single-stock leveraged ETFs tracking Samsung and SK Hynix in late May, their combined assets swelled from $3bn at launch to $9.1bn with 92% of holders being retail investors. Samsung and SK Hynix account for 60% of the KOSPI. Leveraged ETFs act as a multiplier; if you invest in the CSOP Samsung Electronics Daily 2x Leveraged Product, both your daily profits and losses are doubled.

Last week, the Governor warned that leveraged ETFs were overheated and may require intervention. Following these comments, the KOSPI — South Korea’s stock exchange — fell by 10% as investors rushed to take profits. Given their leverage, this sell-off cascaded; as ETFs plunge, fund issuers and liquidity providers must rebalance by selling the underlying physical shares (e.g., Samsung Electronics and SK Hynix). This algorithmic selling drives the index down, triggering further retail margin calls. Goldman Sachs analysts reckon that a 5% swing in the Korean market could force options dealers to buy or sell $4.7bn worth of shares to rebalance their positions — roughly one-eighth of the average daily trading on the KOSPI.

As runaway demand and manufacturing constraints disrupt memory’s usual commodity cycle to structurally embed higher prices, the fundamentals underlying the Big Three are a perfect storm. Sharp pull-backs in their share prices reflect institutional liquidity rebalancing exacerbated by leveraged ETFs and aggressive retail investors, not changes in fundamentals. But, these liquidity dynamics are now strongly competing with fundamentals to determine share price.

Electronics retailers can absorb higher memory prices into their gross margins or pass them onto consumers. Apple, the tech brand with the most pricing power, chose the latter and the market punished it. But while Apple can absorb short-term share price pain easier than almost any other retailer, smaller retailers may be hurt badly. With memory supply forecast to alleviate pricing pressure in 2028, they should weather the margin squeeze to retain customers until then.

Photo taken by me in Mumbai’s business district, Lower Pare.

India is the fastest-growing major economy, with the third-most startups and second-most AI skills penetration worldwide. Yet, unlike other Asian markets like South Korea’s KOSPI which has more than doubled this year alone, Western investors have largely neglected India’s technology sector as having “no straightforward AI play”. The more pressing question, however, is whether its domestic institutional investors have too.

So when flight cancellations amid the war against Iran let me extend my planned stay in Bangalore, dubbed “India’s Silicon Valley” because it houses half of India’s billion-dollar startups and venture capital investment, I took this as fortuitous. I was lucky enough to receive an audience with Mohandas Pai, the former Chief Financial Officer at Infosys. Now, he’s the Chairman of Aarin Capital and Chairman of the Board at 3one4 Capital, Bangalore-based venture capital funds that specialise in healthcare and life sciences and between them have around $500mn in assets under management.

I meet Mr Pai at the offices of 3one4 Capital just off MG Road, Bangalore’s uptown shopping district. He welcomes me warmly, eyes and teeth twinkling in a wide smile behind his grizzled beard. I dive in, asking him about the central criticism of India’s tech sector. “Western commentators say that India and Bangalore merely provide ‘back-end’ services to foreign multinationals. Is this accurate?”.

His face darkens. “Software development, AI-driven data analytics, end-to-end product delivery — would you call these ‘back end’?”. It’s a narrative that he’s heard countless times but still irks him. As multinational companies seek to integrate their operations for AI-augmented workflows and maintain control over sensitive data, he explains, Bangalore’s global capability centres now perform full-stack services ranging from research and development to project management.

For example, Amazon’s largest and second-largest global headquarters are in Hyderabad and Bangalore. Engineers code, deploy, and perform real-time maintenance of software architectures like Amazon Web Service that are critical to global e-commerce. Global capability centres constitute around 3% of Indian GDP and are predicted to grow at a compound annual growth rate (CAGR) of 8%.

Given its centrality in global digital infrastructure, why is India’s tech sector still portrayed as so underdeveloped? To Mr Pai, Western media can’t accept being overtaken. Far easier to default to a “third-world” framing that validates assumptions of superior Western innovation and denigrates Indian engineers as what he calls “cyber coolies”, referring to the indentured workers of the 19th and 20th centuries. He’s particularly dismissive of my former workplace, the Financial Times, whose India coverage he’s publicly criticised in the past.

Bangalore and other cities may be sophisticated operational headquarters instead of mere “back-offices”, but does this risk funnelling India’s top IT talent into multinationals rather than domestic companies? He instantly rattles off statistics: “6mn IT professionals, 1.5mn new IT and engineering graduates each year, 250,000 registered startups collectively worth $700bn. The value is being created here in India”, he declares.

Mr Pai is right. From healthcare to e-commerce, India continues to digitise at a dizzying pace. Bugworks and Dozee, two health-tech startups in 3one4 Capital’s portfolio, are valued at $110mn and $125mn each and the native AI healthcare market is forecast to grow at 42% CAGR. One of PM Modi’s greatest achievements is extending online payments to every Indian with a mobile telephone. UPI (Unified Payments Interface) now accounts for 50% of global real-time digital payment volume. Enabled by this banking infrastructure, the Economic Times forecasts India’s fintech sector to reach $1tn by 2032. Yet, this record domestic growth is coupled with record declines in investment.

In fiscal year 2025-2026, India saw a record $94.53bn of FDI inflows that was neutralised by $86.88bn of domestic and foreign outflows as investors aggressively took profits amid oppressive capital gains taxes and rotated to higher returns and interest rates in America. “Yes, yes”, Mr Pai says when I pose this to him. “There is less foreign investment. But what is more important is, why are we lacking domestic investment?”. He tells me that between 2014-2024, America and China invested $2.32tn and $835bn into startups and R&D, while India invested just $160bn.

He attributes much of this gap to regulatory barriers that have no equivalent at scale in either country. India’s Security and Exchange Board guidelines prevent Category II AIFs — the funds through which institutional capital typically flows to startups — from accepting pension fund capital. Such restrictions discourage the kind of ten-year fund cycles through which America’s university endowments and China’s state-backed institutional funds have compounded their R&D investment. “Why should other countries invest in us if we won’t invest in ourselves?”, Mr Pai laments.

If the national regulatory environment constrains the flow of capital, state-level policy determines how productively existing capital gets deployed. Given his complaints, Hyderabad makes a compelling example of integrating active support infrastructure with procurement access. T-Hub and WE Hub — the world’s largest innovation campus and India’s first state-led women’s incubator, respectively — operate under the ITE&C Department, which also administers Telangana’s Startup Cell. Together, they grant startups approved by the central government’s Department for Promotion of Industry and Internal Trade direct access to government procurement contracts without prior turnover or experience criteria, a 15% price relaxation on public tenders, and 100% reimbursement on the State Goods and Services Tax for five years. Separately, Telangana’s TS-iPASS system guarantees industrial approvals across 23 government departments within 15 days, with a deemed-approval clause if the deadline is missed.

This coordination has precedents in nationally competitive tech ecosystems. Israel’s Innovation Authority is a statutory national body that combines regulatory guidance, direct non-dilutive grants, and co-investment with private venture capital under a single mandate, letting the world’s fourth-highest valued high-tech ecosystem deploy its privileged access to American R&D capital. Telangana’s ITE&C Department, which integrates access to financial and procurement benefits under one institutional roof, is an emerging state-level equivalent.

Karnataka’s Startup Cell offers analogous facilitation — and a larger seed grant under ELEVATE — but its support functions are dispersed across administrative bodies compared to Telangana’s integrated system. Bangalore retains its advantages in startup funding concentration and R&D infrastructure, but Hyderabad has quietly built tighter institutional coordination.

Underlying these regulatory differences, however, is an acute employment squeeze. “There’s more than enough talent, but not enough employer demand”, Mr Pai says when I ask him about graduate unemployment. Following the surge in hiring amid the post-pandemic AI boom, technology job openings fell to a 28-month low in June and contracted sharpest amid low-level positions, according to data from talent intelligence platform Xpheno. This supply-demand imbalance is partially attributable to declining domestic investment, Mr Pai says, but exacerbated by a deepening skills shortage. As routine work is outsourced or augmented by AI, hirers increasingly prioritise frontier developers who write the advanced algorithms and models underlying scientific research and computational workflows in highly specific contexts. Despite this, India’s exam-oriented education system still produces generalist knowers over specialised doers.

The National Programming Skills Report, commissioned by technology and educational consultants Saville and Holdsworth Limited, found that while 92% of engineering students could correctly identify technical concepts in a multiple choice format, only 36% could even partially write the relevant code and only 14% wrote the correct code. To Mr Pai, “education should reflect society and its demands” and is failing to. He doesn’t just blame the outdated education system for this skills shortage, however. Education is the “joint responsibility” of government and business, and he exhorts businesses to actively invest in upskilling new graduates instead of “waiting for government to lead”.

Saanvi, a Telugu girl in her mid-twenties who I met at a gallery and tests semiconductors at Google’s office in the outskirts of Bangalore, is candid about the trade-offs. She’d learned more from two years of training programmes and industry exposure at Google than from her engineering degree, but recognises how niche her occupation is. “If Google downsizes here or the industry finds a way to further automate semiconductor assessment, I’m not sure how easily I can re-train for another specialist position”, she worries. Specialising can boost employability amid this hiring contraction, but rapid technological developments are raising barriers to entry between niches.

The oversupply of tech graduates is perhaps a microcosm for the broader labour-surplus that keeps real wages stagnating behind inflation and leaves tens of millions of Indian youth forced to find or supplement their income through the booming gig economy. Shouting over the cacophony of Mumbai traffic, Aarush tells me that he had to quit his consulting job to support his family after his father died to diabetes complications. An engineering graduate only in his late-twenties, he intended to quit only briefly and then re-enter the sector. But he finds himself outcompeted by the glut of younger graduates who are able to accept lower salaries than him. He currently works as a motorbike taxi and delivery driver to make ends meet.

Mayank Khemka, senior managing partner at ASK Asset and Wealth Management which manages portfolios for high-net-worth family funds, is keenly aware of these challenges. After sharing a presentation about India’s tech sector and growth forecasts with me at his Lower Pare office, he switches off his computer. “All these headline numbers are impressive, but the demographic dividend is not enough”, he says referring to India’s population windfall. “This wealth must reach the youth, the biker who delivers your meal. They are who really drive India’s growth”. I think of Aarush, long vanished into the motorway outside.

And what of Bangalore’s own prospects? Its office rental market is seeing record supply and demand, accounting for 61% of all new office constructions in India this year while having India’s lowest level of vacancies at 7.8%. But surging energy costs and deepening infrastructural bottlenecks constrain its growth. March’s tariff revisions for commercial and industrial electricity users constitute an additional 9% electricity tax atop the commercial bill. 50% of water is sourced from borewells that according to the Bangalore Water Supply and Sewerage Board are seeing a 20-25% spike in demand amid failing monsoons and filling-in of lakes for construction. It’s the second-most congested city worldwide after Mexico City. Arranging to meet Saanvi for a drink after work was hopeless given the time it’d take her to reach the city centre and then back into the outskirts. Competitors like Hyderabad don’t face these acute structural constraints.

Still, Bangalore’s mature tech ecosystem and steady growth continues to attract young entrepreneurs. Shreyas, a young man from a Konkani community in Mangalore whom I shared my AirBnB with, graduated from engineering college in Hyderabad. Rather than stay in the city’s thriving tech sector, he’s come to Bangalore as a frontier developer for Superlend, a decentralised finance startup that directs liquidity across blockchain markets. When I ask why he and Superlend chose the saturated Bangalore rather than the rising Hyderabad, he laughs and says, “Yarr, Bangalore is too big for sure. But now it’s too big to stop growing”. Mr Pai smiles magnanimously; “Yes, Hyderabad has done very well. But this is still Bangalore”.

Bangalore’s lead looks secure, but not unconditional. The next generations of young IT professionals will inherit a rapidly specialising job market whose demands their education system and employers aren’t yet training them for. If Bangalore is indeed too big to stop growing, it will need to be on the shoulders of these rising entrepreneurs and India’s domestic capital.

Governance

Kenya seeks $20.8m for AI-powered social media monitoring system

https://www.mexc.com/en-GB/news/1111153

President William Ruto’s government is seeking $20.8mn (KES 2.7bn) for an artificial intelligence system that will monitor online discourse and to establish a National Communications Centre for government to distribute information. Ruto’s United Democratic Alliance party argues that the system and Centre will combat misinformation and improve civic outreach. A cynical interpretation is that he aims to quash digital dissent before it can manifest into in-the-streets protests — as happened during the youth-led protests from June-July 2025.

Then, Kenyan youth mobilised in retaliation to tax increases and the murder of schoolteacher Albert Omondi Ojwang in policy custody after being arrested on charges of “false publication” after sharing a Twitter post that criticised the Deputy Inspector General of Police. The protests were organised almost entirely over social media, marking the first time such demonstrations have been organised outside of political or tribal structures. During the protests, state-sponsored individual trolls and coordinated groups launched online smear and intimidation campaigns against organisers, often deploying Islamophobic and sexist language to harness resentment towards these groups.

Kenya’s digital repression is significant for its coordination with the private sector and Chinese hardware. The Safe City Project, initiated by government in 2014 to combat crime and terrorism in Nairobi and Mombasa, is a joint project between Safaricom and Huawei. Safaricom is Kenya’s largest mobile-service provider and the most profitable company in east and central Africa. The Project uses Huawei CCTV cameras and facial recognition algorithms and is financed by soft loans from Chinese banks. Its systems were deployed to track and target dissidents in the 2025 protests. Safaricom itself has provided police with the locations of digital dissidents without any court order or disclosure of this information sharing.

Clearly, this requested funding follows a troubling precedent of increasing digital repression. It still awaits final review and approval by Parliament's Budget and Appropriations Committee. If granted, it’ll mark an unprecedented leap in the Kenyan government’s powers to stifle online dissent and target its sources.

Business and markets

DeepSeek V4 Pro tops global bang-for-buck ranking after 75% price cut

https://www.scmp.com/tech/tech-trends/article/3354668/deepseek-v4-pro-tops-global-bang-buck-ranking-after-75-price-cut

Since its release in January 2025, Chinese AI model DeepSeek’s unique selling point has been its blend of cheapness and performance that wiped one trillion dollars in market value from US tech stocks in a single day. Now, its cheapness has been consolidated by a permanent 75% price cut that makes it ~95% cheaper than the models of rival companies like OpenAI and Anthropic.

But it isn’t clear whether these price cuts will stimulate demand for its services. DeepSeek already has 89% domestic market share and its international expansion faces severe headwinds — not least by being partially or fully banned by Asian countries including India and South Korea over security concerns, obstructing it from the world’s fastest-growing chatbot market in the Asia-Pacific. Contrastingly, ChatGPT and Claude are booming in the APAC, with India and Japan as Claude’s leading headcounts.

OpenAI and Anthropic have sufficient capital to heavily subsidise users’ token consumption — particularly as they approach world-beating IPO’s and court significant foreign investment — and are moving to per-token pricing structures (as I wrote last week…). With such subsidies and localised pricing plans for APAC nations, their relative expensiveness over DeepSeek is unlikely to diminish demand for them. And, they face far fewer regulatory barriers.

Last year, I made a similar argument about Chinese electric vehicle manufacturer BYD’s aggressive price cuts. BYD had also saturated domestic demand and faced slews of regulatory probes into its attempts to expand into Europe which still continue. However, domestic sales accounted for 80% of BYD’s revenue whereas at most 44% of DeepSeek activity is from China. DeepSeek has a better international footing, but it’s far from certain that these 75% price cuts will overcome the significant obstacles to its overseas adoption.

Regulation

Palantir hits back at Sadiq Khan after £50m contract with Met police blocked

https://www.theguardian.com/technology/2026/may/22/palantir-hits-back-sadiq-khan-contract-met-police-blocked

After London Mayor Sadiq Khan blocked a £50mn contract between Palantir and the Metropolitan Police, advocates of the American defence-tech firm argue that Khan is “putting politics above public safety”. They highlight Khan’s previous comment that Londoners want public money paid to companies that “share the values of our city”.

In fact, Khan’s rejection of the contract that would’ve automated some aspects of police investigations is about procurement, not politics. The Met had breached procurement rules by failing to present its procurement strategy or detailed cost-benefit analysis to the Mayor’s Office for Policing and Crime (MOPAC) before seeking to award the contract. Moreover, the Met didn’t engage with any other potential suppliers. This willingness to throw £50mn at Palantir without a formal and diversified tender process is sufficient grounds to reject the contract.

It isn’t the first time Palantir’s benefitted from clandestine contracts. Under MOPAC’s Scheme of Delegation and Consent, spending over £500k requires formal approval by the Deputy Mayor for Policing and Crime. In February, the Met awarded Palantir an almost £10k contract for a pilot project that was then extended with another contract worth almost £490,000 — coming just under the £500k threshold for formal approval.

Scotland Yard warns that Khan’s decision will force it to cut officer numbers. But the Met’s £125mn funding shortfall is a consequence of political decisions, beginning with then-Home Secretary Theresa May cutting policing budgets by 18% in 2010, not technological inefficiency. Without even providing an argument for how outsourcing policing casework to AI systems will result in significantly improved margins and officer deployment, the Met’s behaviour is both unprofessional and ill-informed.

Culture

How Saudi Arabia is using AI to transform the Hajj experience

https://english.alarabiya.net/News/saudi-arabia/2026/05/23/how-saudi-arabia-is-using-ai-to-transform-the-hajj-experience

You may have heard Pope Leo XIV’s dire warnings about AI’s economic and political disruption. Interestingly, the previous Pope Leo, Leo XIII, warned about the exploitation of workers in the Industrial Revolution in his 1891 encyclical Rerum Novarum (“Of New Things”). Only 135 years separates both Popes’ similar warnings in unimaginably different contexts. But here’s another example of AI meeting religion.

As an estimated 1.5mn Muslims make the Hajj pilgrimage to Mecca to commemorate this year’s Eid al-Adha, Saudi Arabia is again deploying sophisticated AI techniques to coordinate their movements. Led by the Saudi Data and Artificial Intelligence Authority (SDAIA) and the Ministry of Interior, the kingdom is using computer vision, thermal imaging, and AI-driven analytics from live video feeds to predict and influence crowd density and movement patterns in real time. Crowd control has become a central concern after a tunnel stampede killed 1,462 pilgrims in 1990.

In this recent Hajj, however, the leading cause of deaths has become extreme temperatures. The 2024 Hajj saw over 1,300 heat-related deaths. The following year saw a 90% reduction in such deaths as authorities took measures like updating fitness requirements and reducing wait times for prayers. This year, the General Authority of Civil Aviation has given first-of-its-kind permission to Terra Drone Arabia to use its drones to deliver medicine and water among temperatures up to 50 degrees celsius. 80% of heat-related deaths were unrelated pilgrims, and authorities have already used their camera technology to arrest multiple pilgrims entering through unauthorised routes.

Given Hajj’s significance amid the five pillars of Islam and the world’s religious festivals, the adoption of AI to meet today’s climate-related exigencies is a striking example of old meets new.

Governance

GPs and hospitals in England to be required to share data to create single patient records

https://www.theguardian.com/society/2026/may/10/gps-and-hospitals-in-england-to-be-required-to-share-data-to-create-single-patient-records?utm_campaign=The%20Week%20in%20Data%20TWID&utm_medium=email&utm_content=418930885&utm_source=hs_email

Overlooked amid Starmer’s death throes and the local elections, last week’s Kings Speech announced the NHS Modernisation Bill. One of its central proposals is a legal requirement for all NHS service providers, including GPs and hospitals, to share patient data in a Single Patient Record. Currently, some emergency information like ongoing medication and known allergies is shared. But GPs often have to wait for emails from hospitals describing their patients’ last visits, which can take weeks to arrive. Proponents of the Bill argue that this centralisation will streamline healthcare provision by eliminating these delays.

Clearly, different service providers’ patient data must be accurate before centralisation into a SPR. That’s not the case. Research commissioned by Healthwatch, an independent body that monitors NHS services’ effectiveness, found that 23% of respondents had noticed errors or inconsistencies in their medical records. 16% of these errors were about their previous medication history. A separate audit of over 30,000 surgical patient records found coding errors in 51% (!?) of admissions.

Given the fragmentation of healthcare services and their varying legal distances from the Department of Health and Social Care which will be the legal data controller and custodian under the Bill, establishing accountability for incorrect data in SPR’s will be extremely difficult. Finger-pointing and buck-passing will abound. On these grounds, the British Medical Association has argued that “while we welcome future discussions on the SPR… GPs themselves must remain the data controller of the GP record”.

The SPR proposals come at an inauspicious time for patient data security.
In February, the government granted researchers from UK Biobank, a database for the long-term study of 500,000 citizens, access to coded GP patient data. Just two months later, this data was found for sale on three separate listings on Chinese website Alibaba. Despite assurances that it was anonymised, researchers were able to identify a patient from just their date of birth and details of a major surgery.

Speaking of accountability, Former Health Secretary Wes Streeting resigned on Thursday — the day after the Bill was announced. I can’t help but notice the timing…

Business and markets

Samsung granted court injunction against imminent strike action in last-minute reprieve — talks resume as unions barred from occupying or locking facilities, obstructing workers

https://www.tomshardware.com/tech-industry/samsung-granted-court-injunction-against-imminent-strike-action-in-last-minute-reprieve-talks-resume-as-unions-barred-from-occupying-or-locking-facilities-obstructing-workers

As disputes between Samsung and union representatives of a threatened 45,000-person strike at its memory chip plants reached breaking point, Suwon District Court on Monday partially accepted Samsung’s injunction request that would limit two unions’ planned 18-day walkouts from Thursday. The injunction bars strikers from from obstructing or locking facilities and mandates sufficient personnel to ensure normal safety operations. Strikers will be fined $74,000 per day if this is breached.

One way to interpret this injunction is as the triumph of capital over labour.
It’s significant that the strikes are at memory chip plants. Memory stocks are rallying during a critical memory hardware shortage amid exploding demand. AI stocks have so far driven 80% of S&P 500 gains, and 4/5 of the best-performing US stocks specialise in memory and storage chip production. This memory over-performance is starkest in South Korea, where memory manufacturers like SK Hynix and Samsung have accounted for 70% of its KOSPI index’s 45% year-on-year growth — the fastest globally this year. Samsung’s own share price has more than doubled this year.
No wonder Suwon’s District Court stepped in.

The dispute is far from resolved. Unions have rejected Samsung’s one-time offer of a one-time bonus payout of $340,000 per employee, somewhat overshadowed by SK Hynix’s bonuses amounting of $447,000 per employee this year and $900,000 next year. But the court’s injunction against obstructing facilities and risking manufacturing equipment clearly delineate the boundaries within which strikes are permitted.

Regulation

Anthropic to share Mythos cyber flaw findings with global finance watchdog

https://www.theguardian.com/technology/2026/may/18/anthropic-ai-claude-mythos-cyber-financial-stability-board-fsb

Anthropic is to hold a closed-doors meeting with the Financial Stability Board to discuss the implications of its new model, Mythos, that it claims can unprecedentedly detect and exploit cybersecurity vulnerabilities. The meeting will convene senior economists and central bank officials from across the world, and be chaired by Bank of England Governor Andrew Bailey.

Since Anthropic previewed Mythos on April 7th and warned of its ability to find “ground-zero” vulnerabilities in banking systems and other critical infrastructure, there’s been intense speculation about its capabilities and the risk it poses to worldwide digital security. In the two following weeks, cybersecurity stocks lost billions in share prices as markets feared that the offensive capabilities of programmes like Mythos made them redundant.

But third-party analysis of the methodology behind Anthropic’s assessment of Mythos supports a far more nuanced picture than headlines suggest. A centrepiece of Anthropic’s demonstration was Mythos achieving a 72.4% code execution rate against the open-source web browser Firefox 147. Yet, Mythos didn’t actually attack Firefox 147 but rather a JavaScript shell of it which used a testing configuration stripped of important defence mitigations. The 72.4% success rate relied upon two specific and highly-exploitable bugs without which Mythos’s success rate plummeted to 4.4% which is the same as earlier models. Finally, Anthropic’s human contractors only manually reviewed 198 vulnerability reports generated by the model and then extrapolated this to the “thousands of zero-days” claim.

Ultimately, Mythos must be contextualised. It was announced on the same day that Bloomberg first reported Anthropic’s plan to list on public markets. Securing partnerships with names like Microsoft and JPMorgan Chase make it appear a critical part of global cybersecurity infrastructure. But much of the scare Mythos created rests upon extrapolated success rates, unrealistic and weakened test environments, and pre-patched bugs. Until its successful in realistic testing environments and at realistic computing costs (given the rapidly increasing token costs my last newsletter discussed), any concrete decisions made by the FSB will be unfounded.

Culture

Chelsea flower show garden designers clash over use of AI

https://www.theguardian.com/lifeandstyle/2026/may/13/chelsea-flower-show-garden-designers-clash-over-ai?utm_campaign=The%20Week%20in%20Data%20TWID&utm_medium=email&utm_content=418930885&utm_source=hs_email

Big Tech meets small town. Alarmed parakeets scatter from Chelsea’s private gardens as uproar resounds over award-winning garden designer Matt Keightly using AI to design his garden at the Chelsea Flower Show. His soon-to-launch new app, Spacelift, can reportedly create garden designs from scratch, alarming horticulturists who fear competition from such software. Other horticulturists call Chelsea’s permitting of the AI a “betrayal”.

Sure, there’s an aesthetic discussion to be had about whether using AI replaces or complements the designer’s own creativity, etc. etc. But amid all the more serious discussion of AI, I just find this a refreshingly petty squabble that reads like satire. Perhaps the next season of Bake Off will be marred by AI-enhanced recipes or touched-up pictures of cakes. Amid all the doomsday wails and ecological and societal fuckery wrought by AI, there’s some perverse comfort to be had in how people still stubbornly argue about the small stuff.

Governance


UK immigration officer among two men guilty of working for Chinese intelligence

https://www.bbc.co.uk/news/articles/c0m2wjlkzplo

In the first prosecution of Chinese spies in British history, two men of dual British and Hong Kong nationality were found guilty on Thursday under the National Security Act of assisting a foreign intelligence service. One of them, Chi Leung "Peter" Wai, was working for the UK Border Force and used his access to government immigration databases to track and even attempt to arrest Hong Kong dissidents, feeding this information to Hong Kong authorities.

Wai’s misuse of such databases since 2013 represents not only gross negligence by the Home Office and Metropolitan Police, but the culmination of the gradual erosion of restrictions on the government’s handling of sensitive data. In its pursuit of economic growth and digital IDs, the government has introduced successive legislation — most notably the recent Data Protection (Use and Access) Act 2025 — that relaxes critical compliance mandates and introduced new “legitimate interests” under which sensitive data can be accessed while circumventing the checks-and-balances mandated by the GDPR.

This weakening exposes citizens’ and political refugees’ data not only to misuse by the domestic government, but also opens the door wide to the transnational repression exemplified in this story. Securing this data while enabling necessary access to it is possible, but only with sufficient and lasting political will.

Business and markets

Perspective: AI demand is inflated, and only Anthropic is being realistic

https://www.cnbc.com/2026/04/17/ai-tokens-anthropic-openai-nvidia.html

Tokens, the words and characters that constitute user inputs and model outputs, are the basic unit of AI usage and are used to model demand. Chatting with a LLM consumes a few hundred tokens per paragraph. But the increasingly popular agentic AIs that write code, trawl the web, and execute multi-step workflows, consume 50-100x more tokens. Given this disparity, extracting the actual demand underlying record levels of token consumption is critical for AI companies to judge infrastructural build-out.

Anthropic’s latest pricing model directly responds to this increased token consumption. While it hasn’t changed the price it charges per token, the latest Opus 4.7 model increases the amount of tokens used for processing input categories like code and structured datasets. It also increases the amount used for generating outputs, and output tokens cost 5x more than input! Staff have stated that the new model can use up to 35% more tokens than before. Anthropic has also cut off third-party agentic tools that were large token consumers, alongside tools like OpenClaw through which users would route their commands to bypass their plans’ limits.

By increasing token count and removing these third-party tools, Anthropic can simultaneously increase its token revenue and better estimate demand. As OpenAI pursues its aggressive scaling strategy at a cost of $1.4tn, Anthropic has in contrast invested just $50bn in infrastructure. But it’s projected to break even by 2028, while OpenAI projects losses into the 2030’s. Correctly estimating and monetising token consumption as a proxy of demand is critical for profitability. As Anthropic CEO Dario Amodei said in a recent podcast, “If you’re off by a couple years, that can be ruinous”.

Regulation

What the EU AI Omnibus Deal Changes for the AI Act and What Lies Ahead

https://www.techpolicy.press/what-the-eu-ai-omnibus-deal-changes-for-the-ai-act-and-what-lies-ahead/

One of the most contentious and far-reaching components of the Omnibus is Article 4(a). It will operate as lex specialis alongside existing GDPR to allow the use of sensitive data for bias detection and correction in narrowly defined cases for high-risk AI systems. These systems can be private or public sector and include those use for biometrics and identification, employment and worker management, and migration and border control. Article 4(a)’s application is contingent upon strict assessments of necessity and proportionality, security measures, access controls, and deletion requirements, and doesn’t mandate such bias detection and correction.

I welcome this change. From facial recognition CCTV systems to welfare assessments, there’s extensive evidence that algorithms do bias against ethnicity, health data, and religious beliefs. Processing such sensitive data is necessary to assess for bias, but is prohibited under Article 9 of GDPR. Article 4(a) finally creates a legal route for governments and researchers to do so.

Opponents to Article 4(a) warn of privacy breaches, mission creep, and data leaks, but sensitive data can be anonymised and processed by governmental and intergovernmental bodies like the EU AI Office in narrowly defined contexts and with tightly regulated access/retention durations. I think the biggest risk is data leaks, as swathes of recent examples like the Biobank leaks attest. But there are still various technical methods of storing and processing this data in maximally secure digital environments.

Algorithmic bias is rife and algorithmic governance is necessary. Only governmental bodies can possess the legal mandates and accountability to do it.

Research

LLMorphism: When humans come to see themselves as language models

https://arxiv.org/abs/2605.05419

Dr. Capraro presents a lucid exploration into how the characteristics of LLMs and the vocabulary used to describe them may be projected onto humans. I think some of the most compelling cases are in psychiatry.

LLMs are already being rapidly incorporated into psychiatry, from extracting underlying themes from unstructured data like assessment transcripts to monitoring patients’ progress through analysing their interactions with chatbots. All instances of such use involve making people’s self-descriptions machine-readable by distilling them into statistical distributions of classifiable variables etc. etc. etc. It might seem that the question is how to ensure that psychiatrists still see their patients as more than such data while working with it.

But it isn’t the individual psychiatrist’s perspective that should be considered. Once you place LLM use in psychiatry within an administrative context and recognise the accompanying pressures towards cost-efficiency, scaling, and interoperability, it seems unavoidable that patients must be represented in increasingly abstract and formalised variables to enable them.

Now, the question shifts from what a psychiatrist thinks to how psychiatrists collectively function. In a future digitised healthcare system that so strongly tends towards this interoperability, what room is left for them to even conduct psychiatry in ways whose purposes aren’t to make patients machine-readable? And what of the tragedy of those psychiatrists who do see their patients as more than this, but who can’t act on it?

Amid the cobbled walls and wildflowers in springtime Ghent, another movement bloomed. The conference, titled “Beyond Borders: Youth Resilience Against the CCP,” brought together rising young activists to enhance their skills and cross-community connections in combating the Chinese Communist Party (CCP)’s transnational repression. From Tibetan, Taiwanese, Hong Konger, Uyghur, and Southern Mongolian backgrounds, the activists represented the communities most central in this fight.

Through its wide range of workshops, the conference uniquely blended all aspects of peaceful resistance. In the first session, titled “AI as a Tool,” Athena Tong, a research associate at the China Strategic Risks Institute, described AI’s opportunities and risks for activists alongside examples of its use. This focus on how activists can leverage open-source techniques and their comparative digital freedom was an inspiring and urgent perspective amid usual one-sided focus on the CCP’s deployment of AI.

One particular example that Athena gave was how activists used AI to capitalize upon rare leaks like the GoLaxy papers, which revealed the CCP’s development of AI propaganda personas, to profile the developers behind them. Another is MonlamAI, a large language model and translation tool developed by Tibetan computer scientists in exile to counteract censorship and Sinicization of online Tibetan content. As she remarked, “AI tools run both ways. Activists are increasingly realizing this.” Following HRF’s AI Hack for Freedom in January, which connected human rights defenders with open-source AI developers, the opportunities for activists to create digital tools to power their work with the help of AI have never been greater.

The Digital Protection workshop led by cybersecurity expert Sarah Moulton expanded upon this practical element. Through an interactive simulation where guests acted as members of an NGO facing cyberattacks, she illustrated how software vulnerabilities form a critical bottleneck to operations and can exponentially drain resources if left unchecked. By encouraging guests to reflect upon instances where their own organizations have faced such cyberattacks, Sarah’s workshop valuably grounded theory within specific examples.

The Legal Pathways workshop complemented the previous two by refining participants’ understanding of their legal entitlements to protection from common techniques of transnational repression, such as extradition and harassment by plainclothes Chinese agents. With her extensive expertise in advocating for targeted dissidents as Safeguard Defenders’ China in the World director and external liaison for the Inter-Parliamentary Alliance on China, Laura Harth illuminated the legal mechanisms behind notable case studies.

Among these, analysis of both the CCP’s extradition treaties with other nations and its increasing use of bogus charges of economic crimes — such as in the sentencing of the father of Hong Kong activist Anna Kwok in February — was particularly useful for the exiled Chinese activists in attendance.

Participants also shared their own tips about legal advocacy. Misran Dolan, the president of the Uyghur American Association, noted the benefits of building long-term relationships with congressional staffers. “In most cases, senators will be voted out or representatives will leave Capitol Hill, but staffers will remain and might even open up new points of contact.”

This pragmatism was echoed by Freddy Lim, the lead vocalist of the Taiwanese heavy-metal band CHTHONIC, a two-term member of Taiwan’s legislature, and Taiwan’s current ambassador to Finland. During the Artistic Resistance workshop, his honesty about balancing his political, personal, and artistic lives struck chords touched upon during the workshop’s screening of “State of Statelessness,” an anthology of short films exploring Tibetans’ fight to maintain their identities and family ties across borders and nationalities. I moderated a panel discussion featuring three other Tibetan, Uyghur, and Southern Mongolian panelists who discussed how these themes resonated across their own diasporas.

Sonya Imin, a Uyghur-American filmmaker and academic on the panel, unpicked the thorny questions of complicity and estrangement facing Uyghurs able to return to occupied East Turkestan, alongside their relative privilege over those who cannot. Rei Xia, a Han dissident and Chinese activist who fled to Europe from Shanghai after being imprisoned for her role in the 2022 White Paper protests, expanded on the film’s nostalgia. For her, the forests of northern India evoked longing for her own grandparents’ rural village to which she may never return.

The experiences of Rei and other brave Han dissidents are often obscured by those of ethnic minorities. Yet, it takes perhaps the most bravery to sacrifice one’s own Han privilege to speak out against others’ oppression. Rei’s own political awakening in Shanghai’s clandestine and persecuted queer activism networks was an important reminder that in her own words, “in China, the patriarchy and the government are intertwined into a new form.”

Across the conference’s various academic and practical threads, collective care was proven to be activists’ most effective weapon. This was particularly acute during the Physical Protection workshop led by Pema Doma, former Executive Director of Students For a Free Tibet, who trained participants in collaborative self-defense. While fun and interactive, the session was a sobering reminder of the violence that activists perpetually face from CCP plainclothes agents and counterprotestors. Similarly, the Psychological Resilience workshop led by clinical psychologist Vanina Waizmann taught communal methods to manage the constant risks of burnout.

The Beyond Borders conference achieved its purpose in transcending not just imposed territorial borders, but the intangible borders dividing different peoples fighting the same transnational repression. Through its intimate conversations and expert workshops, guests found new empathy and knowledge in each other. Perhaps best symbolizing this was the Graffiti workshop, which saw different activists from many groups blend their political slogans and imagery into bold new wholes. In this spirit, we realized the promise of united youth resistance to the CCP’s repression, where each of us contributes our unique skills to the struggle we all share.

Photo by Evgeniy Alyoshin on Unsplash

As silver prices steadily rebound after falling by up to 27% on January 30^th, silver bulls are increasingly optimistic about surging industrial demand amid lasting supply constraints. They are correct about constrained supply, but the demand-side narrative requires qualification. Industrial demand is often attributed to AI infrastructure but the solar panel industry is the largest source. Demand from photovoltaic installations has outpaced other sectors, now accounting for 29% of overall silver demand compared to 11% in 2014[1].

However, while the capital expenditure of giants like Nvidia, who consistently reports mid-70% gross profit margins[2], can absorb pricier silver, photovoltaic manufacturers operate at far slimmer margins of around 15%[3]. With silver already constituting 30% of production costs[4], its rally may prove self-limiting as manufacturers seek to engineer their way out of it.

Silver is already the most conductive metal, and optimising its usage is nearing physical limits. The gridlines of silver “fingers” screen-printed on silicon wafers either side of photovoltaic cells are already ~80µm thin. Further thinning them impedes their ability to conduct electrons released by light striking the wafers. Risks of broken fingers or non-uniform heights also significantly increase when screen-printing at such high resolutions[5], while recycling silver also remains highly cost-intensive[6].

Faced with these engineering constraints, manufacturers may have found a copper bullet. Copper currently trades at 0.5% the price of silver by weight[7], and laboratory experiments have found that copper-silver blends can reduce silver paste usage by up to 70% while preserving efficiency[8]. Chinese manufacturers lead this pivot. Giants like Longi Solar[9], China’s second-largest manufacturer by market share, are aggressively building out production lines for copper-based metallisation.

Moreover, the very photovoltaic technologies that utilise copper are jostling with traditional silver-reliant ones. Copper metallisation is easiest to implement in back-contact cells, where all electrical contacts are on the cell’s rear and circumvent the difficulties of extremely fine screen-printing to minimise shade. Longi’s recent field data[10] found that back-contact cells achieved 4.5% cheaper energy costs compared to TOPCon cells that dominate market share and rely on silver contacts on the cell’s front face. While currently more expensive to produce, back-contact cells are forecasted to achieve cost-parity and equal market share with them by 2030[11].

It is not the first time that surging silver prices catalysed thrift. When its price doubled between 2009 and 2011[12], manufacturers reduced consumption by a third the following year[13]. It has decreased by 15% every year since[14]. Some industry experts forecast that if copper metallisation captures 50% of global photovoltaic production by 2030, solar silver consumption could halve[15]. Given its current 29% share of overall silver demand, this would be a very significant demand shock.

Retail investors appear undaunted, pouring almost half a billion dollars into SLV (the biggest silver exchange traded fund) in the week following January’s crash[16]. As they argue, supply is structurally inelastic since roughly 75% of silver is mined as a by-product of base metals. World-leading miners such as Glencore[17] and Fresnillo[18] have recently cut production forecasts by up to 10%, compounded by the long-term trend of declining ore quality across base metals[19].

Yet, these supply constraints are only half the equation. Industrial demand is proving itself highly inventive and elastic to commodity prices. The very speculative inflows driving silver prices may further compress R&D timelines to hasten innovation towards cheaper inputs. This dynamic has been visible in the solar market for over a decade. Even amid tight supply, it is one that could yet temper the metal’s ascent.

[1] https://silverinstitute.org/wp-content/uploads/2025/12/Silver_The-Next-Generation-Metal_DECEMBER-Release.pdf

[2] https://www.macrotrends.net/stocks/charts/NVDA/nvidia/gross-margin

[3] https://docs.nrel.gov/docs/fy25osti/95135.pdf

[4] https://www.reuters.com/sustainability/climate-energy/solar-industry-accelerates-shift-silver-costs-soar-2026-02-19/

[5] https://www.pv-tech.org/wp-content/uploads/legacy-publication-pdfs/4251012258-cell-metallization-by-screen-printing-cost-limits-and-alternatives.pdf

[6] https://www.sciencedirect.com/science/article/abs/pii/S0038092X24007229

[7]

https://www.copperprice.org/

[8] https://www.pv-magazine.com/2025/09/24/copper-metalized-heterojunction-solar-cells-close-to-achieve-same-efficiency-as-silver-based-counterparts/

[9] https://www.pv-magazine.com/2026/01/06/rising-silver-prices-drive-chinas-longi-shift-to-copper-metallized-solar-cells/

[10] https://eu.longi.com/press/longi-releases-global-bc-technology-field-data-performance-and-reliability-exceed-expectations

[11] https://www.pv-magazine.com/2025/12/08/back-contact-will-reach-cost-parity-with-topcon-before-end-of-decade/

[12] https://silverprice.org/silver-price-history.html

[13] https://www.reuters.com/article/markets/us/solar-industrys-silver-usage-drops-as-high-prices-burn-idUSL4N09D2BB/

[14] https://www.gurufocus.com/news/8607549/solar-hits-654-gw-as-silver-surges-154-on-demand-debate

[15] https://viox.com/solar-de-silvering-technology-trends/

[16] https://www.ft.com/content/3183551c-fd27-4469-8e1d-adac4ca7b5ca

[17] https://www.fastmarkets.com/insights/glencore-cuts-2026-copper-growth-forecast/

[18] https://www.wsj.com/business/fresnillo-cuts-silver-gold-production-guidance-after-output-falls-8f3dc3fa

[19] https://www.spglobal.com/market-intelligence/en/news-insights/research/2026/01/mine-cost-outlook-2026-inflation-new-supply-reshape-global-mining-landscape

This adaptation of The Trial tells the story of K., a clerk who is arrested by policemen from an unspecified agency and for an unspecified crime. He’s allowed to continue with his life as usual except from court summons, but his case consumes him. He seeks legal help from a lawyer named Huld, solace from Huld’s nurse, Leni, and advice from a court painter, priest, and another of Huld’s clients.

The Trial is uniquely suited to India. From repealing fines issued by bent traffic coppers, to obtaining a visa (everyday I’m thankful for my overseas citizenship), to buying a house, virtually every interaction with the legal or civic system must be lubricated by a sprawling network of fixers and friends-of-friends. At its best, these bureaucracies are an idiosyncratic nuisance. At its worst, they are weaponised to facilitate corruption and stifle dissent.

So when I settled down in my seat at the Ranga Shankara Theatre (a short walk from my gaff in J.P. Nagar) to read the Director’s synopsis, I was excited by its promise to focus on exactly this contemporary parallel. “When the Right to Information Act [India’s equivalent to the Freedom of Information Act] is diluted repeatedly… to shield governing agencies from public scrutiny, how can one not be reminded of characters in Kafka’s novels?”, writes Vellani.

Sadly, the few attempts at this contemporary grounding were glibly shallow and the overall promise was broken. Perhaps this was a consequence of Vellani’s choice of reinterpreting The Trial as fragmented love stories. K.’s obsession with Leni, in her various incarnations as the court’s cleaner and his lawyer Huld’s nurse, is the play’s central theme. As he grows increasingly bewildered about his case, he oscillates between latching onto her and her purported influence over the court and Huld, and fits of paranoid suspicion about whether her influence is underpinned by sexual relationships.

Leni controlling Huld’s wheelchair.

Vellani stages Leni impeccably. When her and K. are alone, they’re in a perpetual embrace. Otherwise, she is always directly behind the figures she influences and that K. seeks to win over for his case. Despite the dialogue ostensibly concerning his ongoing case, such scenes are physically characterised by K. trying to circle around these figures to access Leni who coyly evades him. Sanjna Banerjee, the actress who plays Leni, effortlessly transitions between a seductive cleaner provoking K. and a dutiful nurse dodging his advances. Here, Vellani’s choice of costume is also effective. Even when they are not interacting, K. and Leni are tied to each other — and to the court — by the reds and yellows they wear that match the court’s insignia.

K. appealing to the audience as he stands before the judge. Observe the parallels between K’s pinstriped trousers and the bars behind the eye in the insignia, and the yellows shared by K., Leni’s tights, and the insignia.

K.’s own moral breakdown is also artfully enacted. He initially pities Huld’s other client, a dejected wretch named Bun whose case has been ongoing for so long that he sleeps in Huld’s back yard. But once K. realises that he’s been sleeping there by the grace of Leni, his compassion sours into jealousy that Leni is helping other people. With this new knowledge, his cruelty towards him is no better than Huld’s mistreatment of Bun. Worse, perhaps K. is beating bun to earn Huld’s favour.

K. initially pitying Bun.

Yet, this isn’t enough to detract from how Vellani underuses those very characters who are part of the legal system and should’ve been central vectors for Vellani to make his political commentary. The policemen explicitly reference the Right to Information Act, but that’s it. In the book, they are whipped for corruption; surely something that Vellani could riff off. Huld and the judge could’ve been pompous, lecherous officials in any country, not just India.

Those politically relevant aspects of the plot are also ignored. K.’s hearings don’t reference specific Indian legal processes. The painter suggests that K. may be accused of sedition. This is relevant to how Indian courts have widely interpreted “disaffection towards the Constitution of India” to justify censorship, but this link was again unsubstantiated. Very strangely, Vellani omits K.’s meeting with a priest who tells him a fable that’s an allegory for his predicament. An Indian reinterpretation of the fable would’ve been the perfect mechanism for political commentary.

Vellani does convey his play’s message that “a person’s struggle to establish his innocence should be…supported by everyone who cherishes the right of individuals…against the arbitrary actions of the state”. But this is a very general sentiment, and not the commentary that was promised. I would’ve minded far less if I found his political commentary to be clumsy. Sadly, I found it all but absent. As a Director, so boldly outlining your play’s purpose in the synopsis sets a very high bar for you to meet. In my opinion, Vellani fell short of it. Without this, his play leaves me entertained but dissatisfied in the important political message it fails to make.

Originally published with Collective Aid, here

Britain’s mooted and underreported plan to deport failed asylum seekers to North Macedonia was incoherent from the outset. Under the plans, Keir Starmer’s government would pay the Balkan nation of 1.8mn inhabitants an undisclosed sum for every person on the move they accepted. People on the move would be encouraged to apply for asylum in North Macedonia to plug labour shortages in critical industries like infrastructure but would not be detained or face restrictions on their movement, making them free to leave the country.

North Macedonia is a key point along the Balkan migration route which is the union of two migration flows: the Eastern Mediterranean route that leads from Türkiye by sea to Greece, and the Western Balkan route. In the latter, people on the move cross through the non-EU states of North Macedonia, Kosovo, Montenegro, Albania, Serbia, and Bosnia and Herzegovina, and then through the EU member states of Romania, Hungary, or Croatia, until they reach western Europe. While traffic along the Western Balkan route has fallen sharply, declining by 46% over the first 10 months of 2025, it remains a critical entry point into Europe for tens of thousands of people.

There are several flaws with choosing North Macedonia specifically as a return hub. Of all the countries along the Balkan migration route, people on the move spend the lowest amount of time in it. A survey by the International Organisation for Migration of 2,680 people on the move transiting through the Balkan route from May 1st to July 31st 2025 found that people on the move on average only spend 4 days in North Macedonia before travelling to Serbia or Kosovo. By comparison, they spend 123 days in Bosnia and Herzegovina.

Commensurately, North Macedonia has one of the lowest rates of asylum applications among Balkan route nations. Of the 4,055 people on the move that registered in its main transit centre in 2024, only 7.5% applied for asylum. That number was even less in 2025.

The 92.5% of people on the move who are intercepted but do not apply for asylum face severe threats from both the North Macedonian authorities and the smuggling gangs they rely on. Border Violence Monitoring Network and Collective Aid have spoken with multiple people on the move who have experienced “pushbacks” to Greece after being beaten by police and having their mobiles, SIM cards, and money confiscated. As defined by the European Centre for Constitutional and Human Rights, pushbacks “entail a variety of state measures aimed at forcing refugees and migrants out of their territory while obstructing access to applicable legal and procedural frameworks”.

Pushbacks occur within North Macedonia’s unique legal limbo. Its asylum applications and transit centres operate under conditions of lex specialis introduced during the 2015. Lex specialis is a legal principle whereby a specific law overrides a more general one if both are applicable. Once a person on the move declares an intention to request asylum, the Law on International and Temporary Protection which governs asylum applications and transit centres takes precedence over the Law on Foreigners.

By not allowing people on the move to declare asylum requests, North Macedonian border forces create a superficial legal justification for pushbacks under the Law on Foreigners which permits the deportation of foreign nationals who illegally enter or stay in the country. Even for the many people on the move who assent to be returned to Greece, they and NGOs working with them say their voluntary return is conducted orally and without any written documentation.

There are widespread reports of people on the move only being allowed to apply for asylum if they testify against their smugglers in illegal and undocumented hearings. Those who do testify face severe risks of retribution, exacerbated by their testimony being outside of any legal framework that could afford them witness protection. Besides this, people on the move face other abuses at the hands of border forces. In testimonies submitted to BVMN, one man recounted that North Macedonian police held a 16-year-old minor on train tracks in front of an approaching train until he confessed that there were two extra people in their group.

Others allege that authorities in return centres pressured them into confessing that they themselves were smugglers. Multiple intercepted people on the move have told Collective Aid that return centres lacked basic hygiene and medical facilities for those who sustained injuries either during their journey or at the hands of border authorities.

There is also a significant risk that Britain would be deporting failed asylum seekers into deeper exploitation by smuggler networks. Since North Macedonia is a transit country and facilitators are paid at each successful border crossing, they exert intense pressure on people on the move to voluntarily return to Greece and try again to enter North Macedonia. Some people on the move even leave the return centres during the night so that authorities do not know which point of the Greek border they will try to re-enter from. If somebody has tried crossing the border multiple times, smugglers also force them to act as a guide for other people on the move.

The Labour government’s plans are unlikely to come to fruition. Speaking at a Brussels conference to counter smuggling in December, North Macedonia’s Prime Minister denied that his government was ever in negotiations with Keir Starmer’s, calling such reports “baseless”.

It is not the first time that Starmer’s overseas deportations ambitions have been shot down. Albania’s Prime Minister denounced similar plans just days after Starmer claimed they were discussed during his visit to the country in May. Montenegro’s Prime Minister has said that it is not part of the Balkan migration route because its railway infrastructure is not developed enough. With unwitting irony, he would only consider its use as a return hub if Britain invested €10bn into its railway infrastructure — over double the cost of Britain’s asylum system in 2024/2025.

Given North Macedonia’s extremely low asylum applications and transition period for people on the move, paying it to receive failed asylum seekers would do little to deter migration flows through it. Since they can retry their journey from North Macedonia, it is further questionable whether deportation there is an effective deterrent for asylum seekers willing to use highly dangerous overland or overseas routes.

Most likely, British taxpayers would pay for North Macedonian authorities to continue conducting cruel and legally unsupported pushbacks that enrich smuggler networks and endanger lives.

Image from Farage’s YouTube upload of his economic speech.

The advantage of being a fringe party hellbent on immigration is that you can afford to gloss over articulating your economic policy. When you’re now the electorate’s favoured opposition and polled to win the next election, this advantage disappears.

Amid Britain’s economic slump, Nigel Farage must now convince voters that he can chart a course out of it. However, under the fiscal limelight on Monday’s speech in the City of London, his latest positions on the most critical economic questions hardly differ from Labour orthodoxy.

Raising the income tax to £20,000? Relegated to an “aspiration”. Scrapping the two-child benefits cap? Only for families of UK nationals where both parents work — this constitutes a very small fraction, since 59% of families affected by the cap have at least one working parent. It’s only the triple lock, Britain’s most explosive political landmine, where he’s wisely maintained strategic ambiguity.

It was easy to tauntingly promise tax cuts before May’s local elections while Labour squirmed under looming tax rises. Likewise, it’s easy to antagonise Starmer by saying you’ll scrap the two-child benefits tax after he suspended several rebel MP’s over it. Yet, these outflanking tactics are no longer suitable. Now Reform has been polled to win the next election for over a year, it’s apparent that delivering an economic strategy palatable to its voters clashes with cold hard economic reality.

For a party whose membership is largely lower and middle-income, Farage has an almost schizophrenic attitude towards growth. He exalts financial services as a silver (sterling?) bullet, advocating in true Thatcherite style for a second Big Bang of deregulation to unleash their potential. Then, without missing a beat, he declares that Reform is the “party of small business” despite consistently portraying local economies’ immigrant-run enterprises as hotbeds of tax evasion. In Farage’s dreams, high streets swarming with scheming Turkish barbers must become lined with smiling ham-fisted grocers whistling the national anthem.

It’s doubtful that these visions of a resurrected financial services sector are realistic. The current contraction is so deep and wide that radical deregulation can only do so much. British financial businesses have declined by 11% since their 2020 peak and a survey of 650 companies spanning the service sector — which includes finance, IT, communications and property but excludes retail — found that it had the biggest fall in orders in almost three years. Deregulation won’t lure back the British companies that continue listing in America’s glimmeringly deep capital markets.

Cost-of-living has consistently been the public’s chief concern. Yet, Farage’s mantras of deregulation are clearly to appease the ultra-rich city donors he relies on — three-quarters of Reform’s donations since 2019 are linked to offshore financial entities — rather than his lower and middle-income white core. It’s much easier to sell them the dangers of unfettered immigration than trickle-down economics. His sermons on deregulation aren’t what they want to hear.

Valorising SME’s (small/medium enterprises) connects much better to these voters’ concerns. Inconveniently, while under 15% of UK residents are foreign born, 39% of Britain’s fastest-growing start-ups have at least one foreign founder. No wonder Farage avoided mentioning immigration in his speech. Besides, he’s equivocating about how exactly he’ll revitalise SME’s out of their current slump with one in three expecting to cut operations, sell up, or shut down. He urges Reeves to lower the minimum wage for younger works or to raise the threshold for employer National Insurance contributions, yet skirted away from committing to either.

British migration is at pre-Brexit levels and the effect of this on immigration-intensive industries like construction is already apparent. So far, Farage has been able to ride high on anti-immigration rhetoric. But as the economic necessity of immigration becomes harder to ignore, whether he can reckon with this while distinguishing himself from Labour’s macroeconomic orthodoxy will be a defining test for his campaign.

Originally published with New Tide Network as their monthly article, here.

As a Tibetan journalist and activist, I often feel disappointed by the way Western media covers issues related to communities from the margins. Through working in student journalism and interning at the Financial Times, I’ve observed how such coverage tends to exaggerate similarities and minimise differences. Recently, I wrote about how the international discourse romanticises Nepal’s so-called “Gen Z Revolution” and misconstrues India’s democratic backsliding, but another topic warrants mentioning: Western media coverage of Beijing’s plans to build a controversial mega-embassy in London.

Despite widespread coverage of this issue, I’ve noticed that very few stories include voices from those affected most by the proposal, such as community members with family ties to China and activists focusing on human rights violations in the region. When such protestors’ concerns are mentioned, they’re mentioned obliquely and—more importantly—homogeneously. Different demographics clearly face different threats from transnational repression at the hands of the Chinese Communist Party (CCP), yet Western coverage often smothers the differences between them.

A Uyghur protestor with family in Xinjiang, where there is an ongoing crackdown on the ethnic minority group, faces a very different set of threats compared to a Taiwanese student protestor without any family in mainland China. For the former, the mega-embassy poses a threat as a potential vehicle of coercion targeting not just overseas Uyghurs, but also their family members in Xinjiang. For the latter, the embassy may serve as a means for the party to counteract dissent on university campuses and suffocate academic freedom. Yet these nuances are often lost in mainstream coverage of this topic, which not only flattens these communities’ concerns but also feeds into the CCP’s racial philosophy of zhonghua minzu (the notion of a Chinese nationality transcending ethnicity) that submerges minorities within a collective Chinese family and history.

By not elaborating upon different demographics’ specific concerns and experiences, the nuanced and myriad techniques of the CCP’s surveillance and transnational repression campaigns are largely left unexamined in many recent reports. Such omissions also exacerbate the discordance I see around the British public’s perceptions of China, which often consists of unsubstantiated conspiracies and simplistic narratives about “China’s plan for world domination”. These narratives are not only often racist—reinforcing discrimination against people of Chinese descent abroad—but also discourage critical examination of the CCP’s specific policies and strategies.

When coverage portrays opponents of the CCP’s human rights abuses as a homogenous mass with identical struggles, only the voices of those affected can illuminate where these struggles overlap and diverge. More broadly, there can be no accurate coverage of China that doesn’t include our perspectives — and such coverage is desperately needed as we move forward into the new year.

A shortened and edited version of this piece was published in The Fence’s Capital Letter.
Enjoy some musings about a very niche industry.

As I dodged the dominatrix for the dozenth time, her paddle swishing past my tray of champagne as it nearly spilled over an intertwined couple in lace masks, I realised that this definitely wasn’t normal catering work. Neither was it normal modelling work. It sat in a murky intersection of the two. A subterranean world that exclusively hires models for the most exclusive events.

At first, I thought it was a scam. An account with only a few hundred followers commenting under my Instagram post of a modelling shoot and then DM’ing me to ask if I wanted to work at a completely unspecified event ‘for a good rate’. I was sceptical but cash-strapped; always an uneven match. I’m happy I took the risk.

Since, I’ve worked almost one hundred events with various agencies in this shadow industry. Some are serious, like industry summits surrounded by priceless artwork in oil tycoons’ Knightsbridge mansions. Some are glamorous, like fashion launches where London’s A-list celebrities mingle under chandeliers and behind hulking Russian security guards, plucking champagne flutes from mannequins like myself. Some are downright lurid, such as the lingerie event I mentioned earlier.

So, here are some do’s and don’ts if you should find yourself at such a swanky event in the near future.

If you see a waiter with a tray of full glasses, for god’s sake don’t put your dirty glass on it. Go and leave it at the bar or find someone who’s specifically taking away dirties. Nothing’s more offputting than lipstick-smudged glasses next to clean ones. They’ll make guests forsake the whole tray. And trying to manoeuvre in a packed room with a tray of glasses is risky enough without the unsolicited addition of a dirty throwing the whole thing out of balance. One hapless elderly gentleman standing next to me learned this the hard way. At first I thought he was shaking with rage but it was his teeth chattering from the icy champagne down the back of his neck.

For your own sake and everybody else’s, lighten up. One of the paradoxes of these events is that fancier events – particularly the fashion ones – have more uptight guests. I’ll offer a group some champagne. Everybody’s eyes swivel around each other’s; all is silent; the stand-off has begun. If sufficient time elapses without anybody taking a glass, nobody wants to be the first. If somebody takes one, nobody wants to be the last and suddenly everybody takes one. If you want a drink, have one! You’ll feel much less anxious if you do and hopefully stop clinging to your iPhone like a lifeline.

Don’t overcompensate on the drinks, though. You’ll just embarrass yourself and your company. At a recent event I worked, a couple went from blissfully slow-dancing together to a furiously hissed argument behind a statue. They stormed off in opposite directions, leaving a table of emptied cocktails behind them. Know when to call it a night.

The mannequin metaphor earlier is apt. Our identical uniforms — there’s sometimes even hair and make-up artists in the changing rooms — elicits some very entitled behaviour from guests. Being rude looks insecure, not classy. We’re hired for our looks but also for our charm (it’s a low bar). If you treat us respectfully, you’ll get the latter out of us and feel more at ease. I don’t take it personally and am quite pleased when an initially tricky guest realises their attitude after I keep smiling at them. But let’s get there without the preamble.

Eat your fill, don’t be shy. The wastefulness of these events is astounding. I’ve watched helplessly as tens of thousands of pounds’ worth of champagne has vanished down sinks. Trays of canapes prepared by Michelin star chefs are emptied into bin bags carried past homeless people on their way to the skip. It’s depressing, but extravagance is the game. Coy abstinence is just self-deprivation. No matter how much you eat, the organisers will continue ordering the same inordinate quantities.

Behind the opulence and authentic camaraderie between staff, there’s sadness here. It merits only a very small violin in the grand scheme of things, but the precarity is real. Almost all staff are juggling some combination of acting, modelling, voiceovers, and singing and dancing as they scrabble to make ends meet. Ageing beaus fight against the twilight of their modelling careers; in this industry, mid-thirties is getting old. The same faces vanish between event seasons, reappearing more haggard each time. This is everyone’s side hustle until it becomes their only one. The plush carpets swallow my footfalls and sometimes, as I walk the labyrinthine corridors beneath these events, I feel like a ghost in the flickering lights.

“Judge a man not by how he treats his equals but by how he treats his inferiors”. I don’t know who said that, but they must’ve been a model waiter. Take their advice at these events and everybody will have a better time for it.

One year ago, it might have looked like an innocuous advertisement. Just Anduril Industries, a Californian defence contractor founded by a virtual-reality wunderkind, promoting its British operations.

The copy (“defending British interests”) seemed unremarkable enough. So did the Union Jack fluttering in the background.

But flags and slogans tend to have meanings and those meanings tend to change with time. In 2025, as Britain’s far-right resurges and Silicon Valley’s militarists find fertile ground in Westminster, Anduril’s imagery looks less like patriotism than something wrapped up in profit and provocation .

In the past year the idea of “defending Britain” has been hijacked by a new nationalist movement. Not since the marches of the National Front in the 1970s, have the Union Jack and St George’s Cross been so weaponised by far-right groups to channel resentment against migrants and Muslims.

Under the banner of Unite the Kingdom, the 12th September rally saw Britain’s largest far-right gathering ever. And there, amid racist chants and brawls, the flags of England and the Union flew from every lamp-post.

Days earlier, Operation Raise the Colours, a grassroots campaign to hoist national flags across public buildings, had become a social-media cause célèbre, debated on Good Morning Britain and endorsed by none other than US vice-president JD Vance, who urged Britons to “push back against the crazies” opposing it.

Such rhetoric has seeped into mainstream politics. Conservative leader Kemi Badenoch declares that Britain must “defend the values, language, institutions and culture that made our country great.” Populist agitator Tommy Robinson speaks of “defending the values that built Britain.” When Elon Musk joined the Unite the Kingdom rally by video link, warning that “violence is coming” and that Britain must “fight back or die”, the overlap between Silicon Valley’s libertarian bombast and Britain’s emboldened far right became impossible to ignore.

Anduril’s patriotic advertisement, launched in the same news cycle, fitted perfectly into that cultural mood music. And Action on Armed Violence’s position is that we believe they knew it would.

Anduril, founded by Palmer Luckey, the billionaire gamer-turned-arms-dealer who boasts of “not losing sleep making tools of violence to preserve freedom”, has become one of Britain’s most energetic foreign lobbyists.

Since January 2023, AOAV has established that Anduril has held at least nineteen meetings with government officials, eleven being with the Ministry of Defence; in the two years before that, it had met only twice. The company has hired at least eleven ex-MoD or armed-forces figures, among them the former operations director of Army helicopter units, and embedded itself within defence-industry associations such as ADS UK and TechUK.

In 2021, it signed a £3.8 million contract to develop an AI “base-protection” system for the MoD, and has since won a £30 million deal to supply attack drones to Ukraine.

The company’s expansion is part of a wider American incursion into British procurement. Palantir Technologies, the data-mining firm co-founded by Peter Thiel, has become Britain’s most controversial tech supplier. In September 2025 it pledged £1.5 billion of investment to make London its European headquarters and to co-develop artificial-intelligence systems for the MoD. Defence Secretary John Healey hailed it as a leap forward; critics called it dependence by another name. The promise of jobs, roughly 350 in total, did little to quiet any scepticism. The issue of ceding sovereignty – something so often raised by nativists and far-right – was rare.

Former Defence Secretary Ben Wallace bucked the trend.

“Court jesters in the UK gov,” he posted on X/Twitter on 12 September, “will lay before Trump gifts of defence contracts with US tech companies Anduril and Palantir—all at the expense of British companies. True defence collaboration is sharing and growing manufacturing jobs. It isn’t some fake London office with a few PR people and ad campaigns abusing our Union Jack.”

Palantir’s own record justifies the unease. The firm already runs Britain’s NHS Federated Data Platform, a £330 million contract linking hospital and patient data. Civil-liberties groups warn that the arrangement grants a surveillance company deep access to sensitive medical information. In the United States Palantir’s software helped Immigration and Customs Enforcement track and deport migrants under Donald Trump. In Britain, the same platform now underpins the Home Office’s post-Brexit border data system.

Both firms have bought an intimacy by hiring those who once wielded power. Palantir hired Leo Docherty, a former Conservative defence minister who was in office when an £80 million Palantir contract was signed. He now advises the firm on “AI trends and geopolitical risk.” Former MI6 chief Sir John Sawers and Labour peer Lord Watson have joined its UK board.

Anduril’s appointments follow the same pattern. The revolving door spins ever faster: a recent study by Transparency International found that more than four-fifths of ex-defence ministers and senior officials move into related industry roles.

The ideological overlap is equally striking. Luckey, a Trump donor, markets his technology as an extension of Western freedom. Thiel, an early backer of Anduril and the driving force behind Palantir, has described democracy as incompatible with liberty and support for the NHS as a form of “Stockholm syndrome.” He helped pour millions into US Republican causes.

His associate Robert Shillman, the billionaire founder of Cognex, a supplier to Lockheed Martin, is a former board member of Friends of the IDF and the benefactor behind the so called “Shillman fellowship” in 2017 to rightwing Canadian website Rebel Media – now called Rebel News – that gave Tommy Robinson about £85,000 over a year. The money that equips drones also bankrolls demagogues.

The traffic runs both ways. When protesters tried to block Thiel’s lecture at Cambridge Union last year, Nigel Farage denounced the “angry mob” trapping “my friend” inside. A few months later he flew to Washington to testify before the US House Judiciary Committee, railing against Britain’s “authoritarian censorship” under the EU Digital Services Act. His appearance was arranged by the Alliance Defending Freedom, a conservative Christian lobby funded by the same network that backs Thiel and Vance.

For America’s defence-tech tycoons, this convergence of cultural and military power is no accident. It is a strategy.

Britain’s establishment, desperate to project global relevance after Empire and after Brexit, offers a receptive market. A fusion of innovation rhetoric and nationalist imagery lets US firms sell both weapons and ideology under a single banner. When Luckey’s company wraps a Union Jack around its drones, it is not simply advertising. It is aligning the dark undertones of nativism with global profits unshackled from any nation.

The implications extend beyond contracts. Silicon Valley’s entry into Britain’s security apparatus entwines the country with America’s culture wars. In the name of defending freedom, it also imports the late capitalism politics of fear and grievance. Anduril’s automated sentry towers now stand watch on the Channel coast; Palantir’s algorithms manages our NHS’ hospital beds and our UK’s borders. Both endanger blurring the line between civilian and military data, between protection and surveillance.

Civil society has begun to push back.

The British Medical Association called Palantir’s NHS deal “deeply worrying.” Amnesty International demanded its cancellation. Action on Armed Violence warned that war-zone algorithms were being normalised inside civilian bureaucracy.

Yet resistance remains muted. The narrative of “innovation” has political momentum; questioning it risks being cast as unpatriotic. The result is a quiet re-engineering of British defence under American direction. Thiel’s consulting firm (co-founded with the ever-connected Peter Mandelson) once advised Palantir while lobbying for contracts. Mandelson’s subsequent resignation from a diplomatic post after his connection to the convicted paedophile Jeffrey Epstein was exposed barely dented the firm’s progress.

What began as corporate outreach has become something deeper: a meeting of minds between Britain’s nationalist revival and America’s militarised capitalism.

The Union Jack, surely a symbol of civic and patriotic pride, now adorns both far-right nationalistic rallies and the marketing of drone companies. The story of Anduril and Palantir is not only about procurement but about ideology. It witnesses the merging of patriotic language, commercial interest and cultural resentment and turns it into a single, exportable brand.

Britain’s generals may welcome the technological edge. But as Silicon Valley’s engineers and ideologues settle into Whitehall, the question for policymakers is whether they are buying hardware. Or are they importing an entire world-view?

Puja outside a temple in Bhaktapur, Nepal.

A version of this piece was published by The Wire.

Amid the coverage of the recent Nepali political overthrow — starting with mass September 8th protests in Kathmandu and resulting in the resignation of the PM and inauguration of a former Supreme Court chief justice until elections in March — the overall tone is optimistic and congratulatory.

Commentators present the young protesters as a political force there to stay. Indeed, multiple mainstream outlets have dubbed the unrest as the “Gen Z Protests”. In a recent piece Karishma Vaswani, Bloomberg’s opinion columnist covering Asia, calls them “Nepal’s new power brokers”. The narrative seems to be that future Nepali administrations won’t enjoy the same endemic unaccountability and corruption faced with a newly emboldened swathe of the electorate; around half of Nepal’s population is aged 0-40.

But international coverage’s narrow political focus excludes two potent actors from discussion. Throughout the revolving door of Nepal’s three main political parties, the army and the police have held enormous influence since it became a republic in 2008. They have been critical in brutally suppressing dissent against its 14 successive governments. Unless the next political administration lessens their influence, Nepal’s democratisation will remain severely constrained.

Living between the cities of Kathmandu, Boudha, and Lalitpur during last year’s summer while I studied Tibetan grammar, I saw how ubiquitous Nepal’s police is. From patrolling busy shopping centres, to operating checkpoints at city borders, to directing traffic at main junctions, it has various municipal roles alongside its policing ones. Similarly, the army builds and maintains rural critical infrastructure like bridges, roads, and healthcare in rural regions, and has primary responsibility for natural disaster relief.

The brutality and impunity of the army and police is commensurate with their ubiquity. The latest protests were met with live ammunition; so far, 72 civilians have been murdered and at least 2,000 injured. Nepali police have long used disproportionate force to suppress protests, ranging from water cannons against civilians to batons against teacher unions. This trend dates back to the 1996-2006 civil war between Maoist insurgents and Nepal’s monarchy, characterised by widespread war crimes and enforced disappearances. In 2014, Nepal formed two commissions to investigate them; they have registered over 60,000 cases and failed to resolve even one. To date, there has not been one prosecution for torture since the crime was recognised in Nepali law in 2018.

In return for repressing on behalf of the state, the army and police freely embezzle public funding. For example, the Centre of Investigative Journalism Nepal tracked Nepal’s 2020 COVID-19 Fund and found that the Ministry of Defence directed almost half of government expenditure — more than spent by the Ministry of Health — to Nepal’s army which claims it used the money procuring masks and sanitisers. The army is increasingly commercially active through endeavours like revitalising defunct textiles factories and is lobbying to overturn a law which limits its investment into business enterprises. So far, this money-for-might contract is poised to blossom.

The army implementing a curfew — since lifted — and negotiating with the youth is motivated by containment, not support. Elections are scheduled for March. With parliament dissolved and former Chief Justice Sushila Karki inaugurated the interim Nepali PM on September 12th, it has plenty of time to build its influence.

Karki’s history of ardently combating corruption makes her a promising candidate for election. In 2017, she was suspended as Chief Justice after the two main parties in the ruling coalition launched impeachment proceedings against her in response to the Supreme Court overturning their appointment of Chief of Police. Her suspension was seen as politically motivated, with the UN High Commissioner for Human Rights at the time describing her removal as “[giving] rise to serious concerns about the Government's commitment to transitional justice and the rule of law".

However, Karki declared she will hand over power to the next administration. This decision is a boon for the army and police; their greatest threat is now out of the running. The likeliest winner of March’s elections is the Kathmandu Mayor Balendra Shah, who enjoys significant support among Nepali youth. A rapper with a strong social media presence who has clashed with federal government and pushed for urban reform in Kathmandu, his anti-establishment branding and party-independence chimes with the youth’s revolutionary fervour. He was swift to tweet his support for the protests and endorse Karki.

Youth support does not determine suitability. Shah’s mayorship has been characterised by a populist, gung-ho style and cruel and counterproductive crackdowns on marginalised groups. Symbolising this, soon after being elected he deployed police to evict squatters from informal settlements around Kathmandu; the police miscalculated the squatters’ strength and retreated with a dozen officers injured, including the Police Chief. He has also deployed police violence to remove street vendors from Kathmandu’s sidewalks, despite official statistics finding that the informal economy employs 74% of workers and contributes to 39% of GDP.

Overthrowing Nepal’s corrupt and inept three-party political cabal is welcome. But comprehensive democratisation requires severing the money-for-might contract between the army, government, and police. Karki would have been the best candidate for this, but her decision to not stand for re-election opens the door to Shah’s likely election. His history of weaponising police violence as Kathmandu Mayor suggest that he’ll continue to do so as PM. International discussion of Nepal’s political overthrow has largely overlooked the influence of Nepal’s army and police. As long as they do, celebration is inaccurate and premature.

Image generated using Canva AI

The last corridor light had long switched off. To anyone looking through the government building’s glass walls, the computer screen was suspended in front of Thupten Nangpa like a mirror palely illuminating his tired face. He rubbed his eyes and tried looking outside. But all he could see was a tunnel of his own reflections, identical men at desks stretching into infinity. A band of light fell across the General Secretary’s portrait on his desk, eyes glittering out at him from the darkness.

Thupten was the only cadre working this late. Next week marked the 90^th anniversary of the 2008 Sino-Tibetan Uprising. For this momentous occasion, he’d been tasked with writing a commemoration of the Uprising to be included in the forthcoming re-release of history textbooks in the Tibetan Autonomous Region (TAR).

It was a task he was uniquely suited to. He was among the last generation of ethnic Tibetans raised speaking their mother tongue before the Party decreed that all Tibetan children attend Chinese-medium boarding schools. His Mandarin was perfect too. As perfect as a Tibetan’s could ever be. He couldn’t remember much before boarding school, but his teachers all assured him how lucky he was to have been rescued from his primitive pastoral life.

His only clear recollection was a face from a photograph: dark eyebrows knotting over darker eyes, a vivid red birthmark curving from the corner of one eye, down through bristly cheek to coarse lips. Learning it was his grandfather’s face didn’t make it scare him less. Even as other memories faded, his grandfather’s face remained etched as deeply into his mind as the lines etched on its leathery skin. He knew nothing else about the man, only that he’d died when Thupten was very young. His mother never mentioned how.

After one last attempt to penetrate the darkness outside, Thupten turned his attention back to his task. While lengthy, writing the main body was easy enough. The Beijing Chief Bureau for Pedagogical Correctness had already sent him a list of government-approved statistics about the Uprising and instructions for their use. With each re-release of TAR history textbooks and their updated statistics, previous textbooks were recalled from schools and previous statistics deleted from the Lhasa Bureau’s databases. He understood the importance of eliminating any “historical ambiguity” the Chief Bureau warned against. The Party’s achievements must be preserved in their full glory.

Thupten knew the government’s explanation of events perfectly, but the statistics were strangely slippery. Despite seeming so vivid while he worked with them, each set of statistics was shunted to the back of his memory by the arrival of new ones and soon forgotten. It was as if he were a camera aperture across which cycled endless reels of government statistics, every snapshot replaced by another as soon as it was captured.

By this late hour, he’d already faithfully reported how Tibetan terrorists unhappy with the post-Liberation harmony between Tibetans and Han had launched an insurgency in Lhasa and murdered hundreds of Tibetans and Han. He’d darkly insinuated the suspected involvement of capitalist monasteries with clandestine support from Western imperialist regimes. He’d swelled with pride as he praised the patriotic Tibetans who waved Chinese flags in protest and trampled the images of the splittist Dalai Lama emblazed on the terrorists’ pamphlets. He’d dedicated a whole section to saluting the noble Chinese military for indiscriminately defending both Tibetan and Han civilians from the terrorists.

As he read the government’s reports of the atrocities the terrorists committed, his feelings darkened into anger. Anger tempered only by shame at being Tibetan like them. Shame in turn tempered by burning pride that no, he was born Tibetan but chose to be Chinese. His Han wife and son’s faces threaded through the images of the victims swimming before him. It could have been them. Innocent people, just trying to reclaim a land rightfully theirs.

All his commemoration now needed was reproductions of the original photographs of the Uprising, along with their annotations. Since the Party adopted its policy of Cultural and Historical Digital Preservation after extremists vandalised several museum exhibitions during the late-2040’s, all major historical artefacts including photographs had been taken into its custody for safekeeping and replaced with government-issued copies. Photographs were reproduced in black-and-white, again to eliminate “historical ambiguity”.

Thupten was honoured to be one of very few cadres trusted with access to reproductions. He leaned further into the computer, its screen’s reflection in his glasses eclipsing his eyes, and scanned his fingerprint to access the reproductions. His eyes widened as he saw they were in colour, and widened further as he saw they were missing the watermark indicating they were government-issued. A suspicion swiftly blossomed in his mind that he couldn’t suppress. These must be unedited copies of original photographs.

Checking the file’s date confirmed it: 17/06/2049, the very year that Cultural and Historical Digital Preservation began.

Hunching over the computer, he looked closer at the first picture. Thupten instantly recognised its general outlines from the pixelated black-and-white reproductions shown in his history lessons about the Uprising. Several men in combat fatigues assaulting an unarmed man on his back, his head snapped towards the camera by a big black boot arcing back for another kick. Then, the annotations had labelled it as depicting Tibetan terrorists assaulting an unarmed Han shopkeeper.

But the higher definition and presence of colour told another story. He knew those combat fatigues – as would anyone who’d witnessed a military parade in the General Secretary’s honour. Flicking through the next few images confirmed this. But he swiftly returned to the first one. On the man’s upturned face, distorted by the kick’s impact yet unmistakable, was a vivid birthmark curving from bristled cheek to coarse, bloody lips.

The night lengthened, its shadows sifting across Thupten’s immobile figure. His tears sparkled in the screen’s glow as they fell. After some time he straightened and gazed at the Chinese flag flying at his desk. The portrait next to it was in shadow. He gazed at the flag for a long time. Eventually, he started typing. He knew what he had to do.

“Thupten” means “one with the capacity for attaining truth”. “Nangpa” means “illumination”. This story uses the 2008 Uprising to explore CCP historical propaganda and its rewriting of China’s invasion and occupation of Tibet.

Thupten realises his grandfather was murdered by Chinese forces in the 2008 Uprising, and that his Han identity is a fabrication. With this knowledge, he reveals the truth in his commemoration.

Image sourced from BYD

BYD, the Chinese electric vehicle (EV) and battery manufacturer, has had a brilliant run. Reaching over one million EV and hybrid vehicle sales in Q1 2025[1], overtaking Tesla as the world’s best-selling EV manufacturer[2], and doubling its first-quarter profit from a year earlier[3], its momentum seems inexhaustible as it seeks to further its success in overseas EV exports. Here, it has also made leaps — it outsold Tesla in Europe for the first time in May[4].

But look closely under the bonnet and complications emerge. BYD is still heavily reliant on its domestic market — in the first five months of 2025, China accounted for 80% of total sales[5]. This domestic market is cooling. A survey by the China Automotive Dealer Association in May found that BYD dealers held an average inventory of 3.21 months — the highest among all Chinese EV manufacturers — compared to the industry average of 1.38 months.

Correspondingly, BYD has slowed its output. Data from the China Association of Automobile Manufacturers showed that BYD’s output growth slowed to 0.2% year-on-year in May, with average April and May output 29% lower than in Q4 of 2024. BYD has also cancelled night shifts and reduced production by at least a third at some factories, and its large dealership in Shandong has gone out of business[6].

But despite slowing output and viciously cutting prices across 22 models by up to 34%[7] in May, domestic demand seems exhausted. While June saw BYD’s highest amount of sales this year — up 11.98% year-on-year — this represented a minuscule increase of 0.03% from May, the third consecutive month of growth below 1%. Within this, plug-in hybrid electric vehicle sales actually decreased 1.05% after the price cuts[8]. BYD’s market penetration in China is now such that price cuts yield negligible marginal returns in terms of demand while invoking ominous warnings from regulatory bodies threatening “corrective action”[9]. Investors’ recognise this: BYD’s Hong Kong-listed share price dropped by more than 9% immediately after news of the price cuts[10].

Facing inert domestic demand, it is imperative for BYD to expand its exports. They have grown rapidly, with June’s exports representing a 233.58% year-on-year increase and up around 30% since January[11]. Europe is at the heart of BYD’s international expansion, with European sales including Britain tripling in Q1 2025[12].

But BYD’s increasing European sales must also be contextualised against a very low baseline. Although BYD has driven the increased demand for hybrid vehicles which are exempt from the higher tariffs that battery-powered EV’s face[13] — the BYD Seal U was the third-bestselling hybrid vehicle in the first half of 2025[14] — Chinese market share of European EV’s has doubled to just 5.1%[15]. BYD’s ambition to onshore production into Europe has also been stymied, with an EU Commission probe into Chinese subsidies behind its €4bn Hungarian plant prompting BYD to shift production to Turkey[16].

This regulatory obstruction represents strange mixed messages from the EU. Antipathy to distorting Chinese subsidies is understandable, but the probe represents the bloc’s mistrust of deepening economic ties between Hungary and China — Hungary accounted for a quarter of Chinese investment into Europe since 2022[17] — and jealousy that China is not providing technology transfer of its EV know-how.
The Hungarian plant alone is projected to create 10,000 jobs[18] and Europe can hardly expect BYD to share its coveted technology, nor even capitalise on it given China’s stranglehold over critical minerals for EV batteries. Given the compelling logic of deepening supply-chain intertwinement with China in light of Donald Trump’s aggressive and erratic trade protectionism, one wonders when enough will be enough for Brussels.

Nevertheless, this probe indicates that Brussels dislikes Chinese subsidies enough to unravel near-complete production plans; an attitude that looms over BYD’s engagement with the bloc.

BYD has penetrated the Chinese market so deeply that radical price and output cuts seem powerless to revive BYD’s flatlining domestic demand. With domestic demand accounting for 80% of sales, pivoting to Europe is imperative. However, BYD’s rapid growth in Europe is impressive but still a drop in the ocean — its European market share is 1%[17]. Amid regulatory obstacles and a wider slowdown in European automobile demand[18], BYD is still a long way from transcending its domestic market and capturing Europe’s.

[1] https://www.bloomberg.com/news/articles/2025-04-08/byd-flags-bumper-first-quarter-profit-as-sales-top-1-million

[2] https://www.ft.com/content/716c9b0b-d8cd-491a-a91b-d70c1e540797

[3] https://www.reuters.com/business/autos-transportation/byds-first-quarter-profit-doubles-2025-04-25/

[4] https://www.cnbc.com/2025/05/23/byd-beats-tesla-in-european-ev-sales-despite-higher-tariffs-report.html

[5] https://en.ilsole24ore.com/art/byd-slows-down-china-production-but-accelerates-europe-AH7efoOB

[6] https://www.reuters.com/business/autos-transportation/byd-slows-production-delays-capacity-expansion-china-factories-sources-say-2025-06-25/

[7] https://electrek.co/2025/06/13/byd-says-ev-price-war-not-sustainable-yet-still-slashes-prices/

[8] https://cnevpost.com/2025/07/01/byd-sells-382585-nevs-jun-2025/

[9] https://www.ft.com/content/74958651-ec86-4a07-a743-502445f54553

[10] https://www.cnbc.com/2025/05/26/chinas-byd-sees-shares-plunge-8percent-as-ev-maker-cuts-prices.html#:~:text=The%20plunge%20of%20more%20than,stiffer%20competition%20in%20the%20sector.

[11] https://cnevpost.com/2025/07/01/byd-sells-382585-nevs-jun-2025/

[12] https://www.reuters.com/business/autos-transportation/china-ev-giant-byd-reboots-europe-operations-after-strategic-stumbles-sources-2025-04-23/

[13] https://www.reuters.com/business/autos-transportation/chinese-ev-makers-sell-more-plugin-hybrids-eu-avoid-tariffs-research-firm-says-2025-05-01/#:~:text=BYD%20pays%20a%2027%25%20tariff%20on%20BEVs%20it%20sells%20in,PHEV%2C%20it%20pays%203%2C999%20euros.

[14] https://cnevpost.com/2025/07/23/chinese-car-brands-double-share-europe-h1-2025/

[15] https://www.jato.com/resources/media-and-press-releases/chinese-car-brands-continue-their-ascent-outselling-mercedes-in-june-and-ford-in-h1

[16] https://www.digitimes.com/news/a20250724PD216/byd-europe-hungary-turkey-plant-automakers.html

[17[ https://www.ft.com/content/ca5f5ce6-46ad-4a34-80b1-b0f7d4b17d58

[18] https://www.ft.com/content/0ef28741-6194-4ee6-8f23-945415de7458

[19] https://fortune.com/2025/07/29/byd-china-electric-cars-europe-hungary-manufacturing/#:~:text=BYD%20sold%20nearly%2055%2C000%20cars,analysts%E2%80%94to%20gain%20market%20share.

[20] https://www.reuters.com/business/autos-transportation/european-car-sales-slump-june-automakers-woes-deepen-2025-07-24/#:~:text=Sales%20in%20the%20European%20Union,%25%20and%208.7%25%2C%20respectively.

Image sourced from Pixabay.

The other day, I finished reading Vile Bodies by Evelyn Waugh. Thanks to Kyiga, my little sis, for lending me it in her quest to corral me into reading some “classics” instead of what she sees as my regressive orbits around Irvine Welsh and cyberpunk. I’ll convince her to read Neuromancer one day — or to watch the upcoming television series with me…

Vile Bodies, published in 1930 and set in the late 1920’s, captures the hedonism and uncertainty of the Roaring Twenties as it follows a cast of young aristocrats and nouveau riche — named the Bright Young Things — as they try to party their way into a post-war identity and future. Despite the characters’ desperate attempts at sophistication, forcing down unholy combinations of “sick-making” kippers and cocktails, these cannot distract from the memory of war looming behind them and threat of war looming before them. Interwoven between Waugh’s deft and amusing satirisation thus lies the novel’s darker moods, with characters frequently slipping from gaiety into existential worry such as in one of my favourite quotes: “Soon someone would say the fatal words, ‘Well, I think it's time for me to go to bed.’”

Befitting the characters’ immaturity, these historical insecurities are seldom, if ever, directly expressed by Waugh or themselves. Their social insecurity, however, is laid bare in excruciating and laughable detail, most significantly through a very 1920’s phenomenon: gossip columns. In the novel’s very first “proper” party, two gossip columnists turn to one another and say, “Hullo… Isn’t this a repulsive party. What are you going to say about it?”. After some witticisms are traded, they begin scrutinising the guests. One asks the other, “Who’s that awful-looking woman?”. Their interlocutor reassures them that “She’s no-one”. After some prodding, they admit that “As a matter of fact, that’s my mother” but “I’d sooner you didn’t [put her in]. She’s been divorced twice since then”. Their interlocutor is sympathetic: “My dear, of course not, I quite understand”. Such are the high stakes of the game that even family ties are swiftly renounced.

Emerging alongside relaxation of libel laws in the early 19th century, these gossip columns could make or break an aspiring socialite’s climb. In them, their identities would be thinly veiled — if at all — and their future standing determined by the avaricious court of public opinion. So, when the gossip columnists are on the prowl, “most of [the Bright Young Things], as a matter of fact, wanted dreadfully to be photographed and the others were frozen with unaffected terror that they might be taken unawares”.

In Waugh’s time, the allure and prestige of being selected as worthy enough to be reported on by a gossip column largely derived from their exclusivity — only a select few made it into them. But I wonder what Waugh would make of how the rationale and business model of gossip columns has been entirely inverted by social media. True, celebrity magazines are still popular but only concern uber-celebs. Now, ordinary people do their own promo for free. With tagging and hashtags etc., we draw and display our social environments, painstakingly choosing what we include and exclude. Our images are completely self-curated. A time-travelling gossip columnist from Waugh’s day would be out of a job.

Social anxiety in our age is just as potent as it was in Waugh’s. The difference is that it’s been outsourced to us through social media. Rather than relying on gossip columnists to portray them as the bohemian, urban chic, these apps make today’s Bright Young Things do the work themselves. This outsourcing has killed the sense of mystery and aloofness that’s surely intrinsic to social prestige. Readers of gossip columns would glimpse socialites only fleetingly through the columnist’s eyes as their coattails or hems swished around a banister. Imagine the same socialite alone at home obsessing over a caption or filter for their post. The allure swiftly dissipates.

Read Vile Bodies here:
https://archive.org/details/in.ernet.dli.2015.182542

Image created with MicrosoftDesigner

Over the past couple of weeks, I’ve had lots of articles about UK pension reform pass my desk. The writers make some compelling arguments for the economic benefits of dynamising the UK’s bloated yet fragmented pensions system by corralling managers to invest in private assets. But there’s little appetite for it among the public or British investors living in this current economic stagnation. Even in the current uplift in British stocks, investors and companies are selling off. Indeed, Reeves has implemented these changes while international pension funds are rethinking their exposure to private assets. Until Brits’ pervasive risk-aversion fades, Reeves’ pension reforms will sound hollow to deaf ears, no matter their merits.

The UK has the third-largest pension market in the world by total assets, worth £3tn. But it also has one of the least dynamic. 38% of assets are bonds while just 30% of overall funds are allocated to equities — the fourth-lowest proportion of any pensions market in the world. This year, the UK recorded the weakest growth in its pension assets among the seven biggest pension markets in the last decade. Britain saw a -0.4% decline while the others saw an average growth of 3.4%. It’s also the only country among 22 studied to have seen negative pensions growth this year.

Given this, Reeves has understandably lasered in on pension reform as the economic lightning bolt she so badly needs. June 5th’s Pensions Bill would allow the Financial Conduct Authority and the Pensions Regulator to force workplaces with direct contribution (DC) schemes to invest in specific equity including private debt, venture capital, and property. Reeves also signed a pact with the 17 biggest pension fund managers to invest at least 10% of their default funds in private markets by the end of the decade, with half of these funds ringfenced for the UK. As a kick up the arse to get them into action, the bill also mandates that multi-employer DC schemes have assets of at least £25bn by 2035.

The logic of revitalising Britain’s flatlining stock market since 2008, stemming the flight of British start-ups publicly listing in the US, and encouraging pension investment into Labour’s expensive infrastructure projects reads nicely. But it stands in stark contrast to investor and public sentiment. Driven by the scramble to diversify from American markets, British stocks are set to have their best half-year run since 2021 and overseas inflows into British equity have hit a three-year high. Foreign direct investment into Britain remains the second-highest in the EU despite bloc-wide declines. In spite of this, British investors have been net sellers all year. If British investors are currently selling in this unusual uplift, can we expect them to buy when times are leaner?

Clearly, even when UK stocks have a rare comparative advantage, British investors still don’t believe in them. Interestingly, this wariness is even mirrored in the activities of major FTSE 100 players. Over the past couple of years, dozens of British companies have offloaded assets and operations: the National Grid sold its UK gas network to the Australian investment bank Macquarie (infamously a major stakeholder and dividend pirate of Thames Water…); Barclays is negotiating selling its UK payments business to a Canadian investment giant; the engineering conglomerate Smiths Group is breaking up; and, depressingly, Unilever is re-listing its ice cream business — which owns Ben & Jerry’s, Calippo, and Magnum — to Amsterdam. Not very summery.

This de-risking and preparing for leaner times is also reflected in public sentiment. A survey last year found that 69% of respondents prioritise protecting their pension savings over a higher return. 94% of members of defined benefit contribution schemes say it’s important that “politicians can’t interfere with my pension”. These opinions seem to be held in a vacuum of actually knowing what pensions are. While 53% of respondents to a different survey thought they were knowledgeable about pensions, only a third could name the different kinds. Fair enough — pensions aren’t the sexiest topic and I didn’t know much until articles about them came across my desk. But government will find it hard to sell pension reform to a widespread combination of risk-aversion and ignorance.

It’s also not at all a given that pension funds will be able to keep riding on American returns as they have until now. Many large ones are diversifying from the US, particularly into Asian markets. American risks include its creaking debt burden; the looming decline in productivity — particularly in the services industry — that mass deportations will cause and that will only become apparent in this winter’s or next year’s US jobs reports; and the very real possibility that Trump opening up Americans’ 401k pensions to illiquid and opaque private markets will lead funds like Apollo and KKR to offload risky assets, as private equity struggles with diminishing returns and fundraising. All of which illustrate how Reeves is seeking riskier pension investments while international players are increasingly wary of global and particularly American exposure.

Leveraging pension funds’ dormant capital could be a very useful economic tool for Britain, as it has been for many other nations. But with such tepid sentiment among British investors and the public — even during British stocks’ current rare uplift — I wonder whether Reeves’ reforms will be enough to rouse UK pension managers into meaningful action. Particularly given international funds’ realisation that easy American returns aren’t guaranteed. With our pallid economic outlook, Reeves hasn’t got much license to ask taxpayers to accept her leveraging the pensions that are more valuable to them now than ever. Currently, Brits simply don’t have enough sense of a national project to get excited.

Scared money don’t make money.

Image sourced from Pixabay, The Digital Artist

When it comes to artificial intelligence, researchers and media alike almost exclusively focus on whether AI is intelligent - but rarely question whether it’s artificial. This was the central theme occupying me throughout my time studying philosophy then neuroscience at uni, and I think it’s the most profound intersection of the two fields: to what extent do artificial learning systems reflect the structures and computations of the human brain? For almost a century, this question has been tussled over by two competing theories: connectionism and symbolism.

The debate

Connectionism was first posited in the 1940’s but really took off after 1987 when McClelland and Rumelhart published Parallel Distributed Processing, connectionism’s bold challenge to prevailing cognitive science. In connectionist models, mental representations are realised by certain patterns of activity across highly-interconnected networks of neurons or nodes. It’s meaningless to try to decompose mental representations and evaluate their structure, since it’s the activity patterns that counts. Learning in any domain is underpinned by essentially the same statistical algorithms.

In the intervening period between the 1940s and PDP, an alternative paradigm that we can loosely term symbolism was dominant, thanks to Chomsky’s and Turing’s visionary work throughout the 1950s. In symbolist models, mental representations consist of the application of innate mental rules to highly structured constituents. Decomposing mental representations and evaluating their structure is highly meaningful; it’s the rationale behind the entire enterprise of generative grammar. Learning in language is underpinned by radically - even intrinsically - different computation as, say, motor learning.

Symbolists have long argued that certain linguistic phenomena cannot be satisfactorily modelled by connectionist systems. Take syntactical overgeneralisation, whereby language learners extrapolate a rule beyond the scope of their experience with it (e.g. a toddler saying “feets” after hearing “hands”). This was a central symbolist argument for the existence of innate linguistic rules (in this example, governing pluralisation of common nouns), such as in Chomsky 1957. In their 1988 paper, the classic critique of connectionism, Fodor & Pylyshyn argue that connectionist architectures are inherently unable to model such rule-evidencing behaviours.

Some recent research, however, has found that connectionist LLMs can mimic syntactic overgeneralisation and even outperform expert humans in making judgements thereof. Other research has found them to mimic or outperform humans at other non-linguistic tasks that still require the kind of rule-manipulation symbolists concern themselves with. These successes do support the notion that connectionist architectures can model human cognition, at least in language.

More connected isn’t more resilient!

Let’s turn the question on its head, though. Just as interesting - and much less explored - is whether AI mimics when language goes wrong. A fascinating property of human language functions is that they can be very dissociable. Since 1861, it’s been clear that lesions to brain areas underpinning language production can significantly impair it while leaving language comprehension intact, and vice versa. Critically, this dissociability shows that the neural architecture of human language systems has a certain minimum degree of disconnection.

In fact, across the whole brain disconnection between areas is just as striking as connection. The total number of neurons making connections within a hemisphere is a staggering 6 x 10^9. But there are only around 10^8 axons (neurons’ projections that transmit electrical impulses) in all the brain’s major long-range tracts combined. Of all the cells making connections between different areas across the brain’s surface, only 2% have access to the long-range tracts even within one hemisphere.

I’m not a computer scientist, but here’s my understanding of the upshot of this striking difference between the brain and LLMs in which virtually all nodes are interconnected. LLMs have been susceptible to catastrophic forgetting whereby redistribution of network weightings while learning a new task erases or damages representations underpinning a previous task. Like, say, learning to play the guitar and then forgetting how to play the violin.

Simply adding more nodes or hidden layers (intervening computations between input and output) doesn’t always fix things. Tellingly, catastrophic forgetting is best mitigated when researchers implement in LLMs the kind of post-learning structural changes that actually occur in the brain. For example, during new learning some nodes’ weights are pulled back to their previous values proportionally to how important those nodes are for maintaining previous learning. For very important nodes, it’s then only under very specific conditions that their weights are changed by new learning. This is inspired by discoveries that the brain makes similar adjustments to changes in synaptic weighting that consolidate learning. (If you’re interested, these adjustments in both brains and LLMs are often applications of Bayesian statistics).

Now, the question becomes: under what conditions can LLMs achieve maximal performance with minimal connectivity? And, which network architectures best exhibit the functional dissociability and resilience to damage of the brain? Then, we can seek to implement the computational analogues of increasingly sophisticated neural structures — such as the corpus callosum (which connects the hemispheres), the arcuate fasciculus (the white-matter tract connecting multiple critical language areas), and the various long-range tracts across the cortex — and see how LLMs fare after we selectively damage them.

Why is this important?

Some of you may be asking, “what’s the point of building LLMs on human neural architecture if they can replicate or outperform human cognition anyway?”. It’s a fair question. As discussed, doing so can overcome some of the unique weaknesses of LLMs. But beyond this, experimenting with incorporating and then damaging neural architecture in LLMs lets researchers enhance their resilience and also observe unique behaviours that may only be elicited through breakdown.

The International Energy Agency predicts that demand from AI data centres will quadruple by 2030 and be the biggest driver of data centres’ requiring almost as much electricity as Japan. Even today, ChatGPT (an LLM) consumes 100,000 times more power than the average American household. Clearly, any success in making AI need less power is a major win for the climate.

Most of all, the human brain is the most beautiful and complicated object in the known universe. That’s also true of you, reader. Implementing human neural architecture in LLMs reminds us of the sophistication of human cognition and to be careful when thinking we’re progressing in understanding it.

To the system theorist, system failure is more enlightening than system success. Even if LLMs and AI can beat us on our best days, they’ll never display the graceful dysfunction of machines that are living.

I sense it before I see it, a cold shard embedded deep in the constant rumble of the hive. Growing colder as I wind down through the catacombs where no worker’s been in generations. Being this deep gives me the heebie-jeebies as it is, and the hairs on my forelegs are standing up. Bad signs, mannnn.

I turn the corner and Syme’s there with him. He makes Syme look tiny. He’s an evil-looking fucker, no doubt. I’ve only ever seen wasps from the look-out points on the surface and each time I bloody make sure they haven’t seen me. His eyes are red and huge, at least twice the size of mine and Syme’s yellow ones, and set at a sharper angle like blades. He completely blocks out the tunnel behind him, his bulk casting grotesque shadows on the walls.

— That’s my mate, the one I told you about, Syme introduces on my behalf. I don’t trust myself to speak, I just nod. His eyes are brighter than usual and he’s speaking fast. He’s been at that bloody pollen again… hold it together, Syme… The wasp turns its head to me and I force myself to look back into its eyes. Nothing in there. Syme turns to it and clicks rapidly. He never said he could fucking speak to it!!! The sounds come out twisted, the vibrations scraping against my antenna. Our mandibles were never meant for that. This is some bad juju mannnn.

The wasp fires back at him, and Syme looks troubled. I’m feeling the pheromones coming off it. It’s a sickly, rotting, mist, cloying my receptors. Bits are stuck to Syme’s hairs. He replies even faster and he doesn’t even realise his wings are twitching like crazy. I can’t understand any of his patter but from how he’s angling his head and torso he seems to be asking it a question. It doesn’t move, just looks at him and those fucking eyes are sucking him in. Syme looks away and clicks once in reply. The wasp lifts its head and I swear it looks satisfied.

Symes turns back to me. — We have to move the plans ahead, he says. They want to make their move before it’s fully dark, so they can still see while we can’t. — Take this, he says, passing me a webbing sac. I look in it and it’s the abdomens of two juvenile fireflies, still weakly flickering. They can’t have been taken more than a couple of hours ago. I flash back to all the times when I’ve been out foraging too late and their lights have guided me back to the hive. Oh, Syme, fucking hell…

He must’ve seen my face. — I know, it’s horrible mate. But I had to. They’d gone too far from their resting grounds, they were goners anyway. Better me then dissolving in a spider’s nest right now. I made it quick, I promise.
I could see two neat punctures from his stinger, straight in and out. I hate it but I see his logic. He looks at me, urging me to meet his gaze. His foreleg brushes against mine. Alright.

He continues. — I’ve left a danger-message in the outer-north-east intersection. A patrol will come across it in fifteen minutes. They’ll rush to the queen and the larvae immediately. It’ll be dark and you’re faster than them though, you get to the brood chamber and tell the guards they’re ordered to relocate them to the inner-west chambers. Then the wasps will take the soldiers by surprise in the brood chamber. Once they’ve surrendered, they’ll take the queen and her guards. Some of us will have to die, but it’s worth it. Revolution is never bloodless. When the queen and her court are under our control, we’ll remember them as martyrs.

All I can do is nod again. — Good man, Syme says. Then, softer: — You’re doing your father’s memory proud. The sound of his damaged wing beating against the south entrance rings through my mind. The thuds of the badger blundering towards the hive as my brother screams at the soldiers to let him in.

Syme turns to the wasp and clicks again. It rests its cold eyes on Syme, then on me. This time, under my hairs I can’t stop myself from shivering. Syme quickly clicks something and it takes one last look at me before it moves down the corridor, its huge stinger swaying. As the light shifts, I see something swinging between two of its legs, thick droplets falling to the tunnel floor. Something like a new pupae.
Nooooo, Syme, what have you doneeee…

Syme’s eyes are brighter than ever as he turns to me. — This is what we’ve been waiting for, this is when it all changes. I’ll stay here to direct the other wasps when they arrive. You go to the intersection, quickly! He briefly presses his antenna against mine and then I’m off, the rocking tunnels dizzying me as I pass.

I reach a junction and skid to a stop, scanning the layout. That way. I turn the right-hand corner and something sharp jabs my side, knocking me to the floor. As the pain flashes through me, I look around for the outcrop I ran into. Instead, six legs enter my blurry vision and hairy forelegs turn me over. Over the blood rushing through my antenna and out of my thorax, I hear Syme’s voice, distant like it’s coming through a waterfall.

— It made me do it, it didn’t think you’d go through with it. It thought you’d bottle it and tell the soldiers. He uselessly pummels me. His forelegs don’t hurt. They feel like rain against my hairs. — Why couldn’t you be braver? Why did I have to do it all? You don’t know what it’s like, taking what could have been your own pupae as the queen watches.

I try to speak but my punctured trachea can only whistle. No Syme, it’s not like that, I swear, I wasn’t going to… I believe in this too… please Syme pleaseeee…
He leans over me, his face filling my vision.
— I’m sorry, brother.
The last thing I see is his eyes. As darkness closes in, they seem red and slanted.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

Went to a talk about bees with Dad over the weekend. The lady giving it had a bee-box she was passing around (pictured). Clearly, it stuck with me. Plus I’ve been watching some Alien clips that couldn’t help matters. Enjoying writing the non-human fiction as of late. Bzz

.