Weekly Reading List: 13/12/2025

Things I Read This Week That Made Me Think

What does it mean to have ideas about something you didn’t build?

This is the question that keeps coming back to me this week. Sam Altman built a company around the transformer architecture he didn’t invent, using scaling laws discovered by researchers who’ve since left for competitors. Diddy built an empire on a myth of self-made genius while systematically exploiting everyone around him. Charli XCX watched her album get diluted into wrong hex codes and wrong interpretations the further it traveled from her hands. In rural China, a chicken wears an ankle bracelet generating data for systems it will never comprehend, creating value it will never capture.

The question isn’t just who gets credit. It’s who gets to narrate, and what disappears when they do.

1. “Sam Altman Has No Idea What He Is Doing”: Neural Foundry
Link

The title is provocative but the argument is precise: Altman’s actual contribution to OpenAI is fundraising and network-building, not technical insight. The transformer architecture came from Google. The scaling laws came from researchers like Kaplan and McCandlish. Ilya Sutskever saw what transformers could become. The people who built the thing keep leaving for Anthropic or Google or xAI, and those labs are now matching or beating OpenAI on benchmarks.

What’s left when the scientists who understood your technology decide they’ll have more integrity elsewhere? A reputation for being “not consistently candid” with your board, former employees under NDAs that threatened their equity, and a valuation held together by the mystique of being first.

What I got out of it: The distinction between “scaling” as an insight versus “scaling” as the phase that begins after someone else has done the conceptual heavy lifting. Altman didn’t pick up the transformer architecture and champion it internally, he arrived once it was in motion, then retrofitted the curve into a personal philosophy. There’s something to be said about watching someone narrate a revolution they didn’t start. The confidence never wavers. The vocabulary adapts. “Scaling” becomes a worldview rather than an engineering phase. And the people who actually understood the thing, who saw what it could become before it became anything, keep leaving for places where they won’t have to watch their work get repackaged as someone else’s vision.

2. “The Reckoning We Still Refuse”: Gerrick Kennedy, Future Tense
Link

This is ostensibly about Netflix’s Diddy documentary. It’s actually about complicity: the institutional kind, the cultural kind, the kind you don’t notice until someone names it. The piece traces how Puff’s violence was always public knowledge: he beat Steve Stoute so badly he broke his jaw, paid $500K, and was partying in the Hamptons with J-Lo five months later.

The documentary shows the predator but can’t seem to examine the ecosystem. No experts on domestic violence. No interrogation of how hip-hop’s foundational misogyny provided cover. No exploration of how Black celebrity culture looked away, how white consumers maintained moral distance, how media traded access for flattering profiles. And it’s produced by 50 Cent, who has his own documented history of domestic violence charges, revenge porn, and mocking Terry Crews for coming forward about sexual assault.

What I got out of it: The question one of the talking heads asks himself: “Does that make me part of the Sean Combs cult?” That question, Kennedy argues, is the one the documentary can’t face. The system that made Diddy possible is the same system that made the documentary. And maybe that’s always the limit of institutional reckoning. The institution can show you the monster, can even profit from showing you the monster, but it cannot show you how the monster was fed. Because the feeding involved too many people who are still in the room. The camera points at the predator so it doesn’t have to point at the audience.

3. “Training, Reasoning, Coordination: The Three Levers of AI Progress”: Marcel Salathé
Link

A mental model for understanding where AI gains are coming from. Three overlapping S-curves: training (more data, bigger models), reasoning (more compute at inference), and coordination (multiple agents working together). Each lever delivers great returns until it doesn’t, then marginal gains shift to the next one.

GPT-4.5 was the signal that training scaling is hitting real constraints. 10x the compute of GPT-4, and the gains were... modest. OpenAI’s own system card initially called it “not a frontier model.” Now the action is in reasoning: o1’s log-linear relationship between test-time compute and accuracy. Coordination is still speculative, but the historical precedent is interesting. Human brains haven’t changed in 10,000 years, yet here we are splitting atoms. The difference isn’t individual intelligence, it’s that those brains learned to coordinate.

What I got out of it: The Ilya Sutskever framing: 2012-2020 was “the age of research,” 2020-2025 was “the age of scaling,” and what comes next is “back to the age of research again, just with big computers.” What strikes me is how this reframes the current moment not as acceleration but as transition. The easy gains are behind us. The next gains require thinking differently, not just spending more. And if the coordination lever is real, and if the future is multiple agents working together rather than single models getting bigger, then we’re not building minds anymore. We’re building societies. With all the emergent dysfunction that implies.

4. “Trustless Chickens” — Thejaswini M A, Token Dispatch
Link

In Guizhou province, free-range chickens wear ankle trackers. The data flows into a blockchain. Middle-class families in distant cities pay 10x the normal price, scan a QR code, and watch their dinner’s step count.

The technology exists because Chinese consumers have been repeatedly traumatized by food safety scandals. You’re not paying for premium protein, you’re paying to worry less about what’s on the plate. Blockchain here isn’t solving a technology problem. It’s prosthetic trust, bolted on after the original trust was hollowed out.

Farmer Jiang doesn’t own the blockchain, the hardware, the software, or the customer relationship. One year Lianmo ordered 6,000 chickens. The next year, zero. No warning.

What I got out of it: “The ledger records everything it was designed to see and almost nothing about why anyone thought this was the best we could do.” This is the sentence that’s been stuck in my head all week. We keep building systems that optimize for legibility, for making things trackable, auditable, verifiable, without asking what gets lost when legibility becomes the goal. The chicken’s step count is visible. The farmer’s precarity is not. The consumer’s anxiety is soothed. The conditions that created the anxiety remain untouched. Blockchain here isn’t a technology story. It’s a story about what happens when trust has to be manufactured because it can no longer be grown.

5. “The China Playbook Indian Founders Haven’t Read”: The India Notes
Link

An interview with Alysha Lobo, who cold-emails Chinese CEOs and shows up at their headquarters without appointments. Once got inside DJI because she stopped to pet a stranger’s Golden Retriever during a typhoon. The whole thing is a masterclass in “you can just do things.”

The substance: Chinese hardware competitors share labs. Rivals can literally walk into the same facility and tinker with each other’s robots. The government acquired Germany’s Kuka robotics and then used it as a live classroom for domestic players like Estun and SIASUN, turning a foreign acquisition into competitive acceleration rather than a monopoly. WeChat has replaced the App Store. Every major retail store has a dedicated section where influencers come and do TikToks. The Shenzhen-Guangzhou-Hong Kong cluster is now the #1 science and technology cluster in the world, ahead of Tokyo-Yokohama, ahead of San Francisco.

What I got out of it: Chinese chopsticks are round at the eating end and square at the holding end. The symbolism: the part facing outward means be kinder to others; the part pointing toward you means be harder on yourself. I keep thinking about this as a design philosophy, that the interface you present to the world and the interface you use to grip reality might need to be different shapes. And maybe that’s what the Chinese hardware ecosystem understood before anyone else: that competition and cooperation aren’t opposites. You can share labs with your rivals because the hard part isn’t protecting your secrets. The hard part is building the capacity to have secrets worth protecting.

6. “The Meesho Must Go On”: Tigerfeathers
Link

A 15,000-word deep dive on India’s largest e-commerce platform by order volume (bigger than Amazon India, bigger than Flipkart) that most English-speaking metro Indians have never heard of. The trick: Meesho exists to serve buyers and sellers traditional e-commerce ignored.

The founding story is a masterclass in listening. They started with hyperlocal fashion, pivoted to Shopify-for-merchants, noticed that WhatsApp boutiques run by housewives were more engaged than the shopkeepers they’d built for, pivoted to serve them, then noticed their end customers had matured enough to buy directly, pivoted again. Each time following the user, not defending the business model. One of their core values is literally “Listen or Die.”

They moved to zero commissions when every marketplace relies on commissions as primary revenue. They built discovery-based feeds when every competitor built search. 73% of all orders come via algorithmic recommendations, not search, because their customers are used to browsing physical bazaars without preconceived buying intent.

What I got out of it: The Viktor Frankl epigraph that opens the piece “Those who have a ‘why’ to live can bear with almost any ‘how’” as a description of product-market fit. Meesho had its why from Day 1: serve the people traditional commerce ignores. The how shapeshifted from pivot to pivot- hyperlocal fashion, Shopify clone, WhatsApp resellers, direct B2C- but the why never changed. Most startups die because they fall in love with their how. They build a technology, a business model, a growth loop, and when the market shifts they can’t let go. Meesho kept letting go. They killed working products to chase what their users actually needed. “Listen or Die” as a core value sounds like corporate pablum until you realize they’ve actually done it four times.

7. “The Death of Cool”: Charli XCX
Link

Charli on what happens when your work escapes your control. After brat, the narrative wasn’t hers anymore. Brands adopted a visual aesthetic that was clearly tapping into the album but got it slightly off each time. The more time passed, the more bastardized the representations became. These representations were replicated, reproduced, deemed as truthful.

The death of cool isn’t popularity. Julia Fox could have a McDonald’s meal and little Julia drones delivering Amazon packages and still be cool because she’d find a Warholian way to make it make sense. The death of cool is “the second you apply a ‘something for everyone’ approach to art in an attempt to deliberately appeal to more people.”

What I got out of it: One of the vinyl plants accidentally printed 10,000 copies of brat in the wrong hex code. That mistake, she says, felt inherently brat; a bastardized version born from accident. Maybe if you spin it with the right confidence, anything can be cool. Maybe cool can live forever. But I think the deeper point is about authorship and entropy. The moment you release something into the world, it begins to degrade semantically. Other people’s interpretations layer over yours. Brands borrow the aesthetic without understanding the ethos. The thing you made becomes a thing that references the thing you made. And at some point you have to decide: do you fight to preserve the original meaning, or do you let go and find the brat energy in the decay itself?

8. Dream11’s Existential Pivot: Money Control Link

India’s largest fantasy sports platform lost 95% of its revenue overnight when Parliament banned real-money gaming. The Promotion and Regulation of Online Gaming Act, 2025 made no distinction between skill games and chance games. Dream11, rummy apps, all of it, gone.

Here’s the remarkable part: they didn’t sue. They didn’t lay off anyone. CEO Harsh Jain split 1,200 employees into eight independent business units (FanCode, DreamSetGo, Dream Money, Dream Sports AI) each operating like a seed-stage startup with its own P&L. And the core Dream11 app is pivoting to... watch parties. A “second screen” where users can watch sports alongside influencers doing real-time commentary and banter. The thesis is loneliness: “People want to scream, shout or banter while watching a fixture.”

What I got out of it: “If you have a sports match where a decision went against you, a refereeing decision, and you lost the final, it doesn’t mean you change the team. It means you play another World Cup six months later and then you bring home the trophy.” There’s something almost spiritually instructive about how Dream11 handled this. No lawsuits, no layoffs, no public tantrums about regulatory overreach. Just: okay, this is the new game, how do we play it? The watch party pivot sounds almost absurd (from a billion-dollar fantasy sports platform to... group viewing?) but the underlying insight is serious. They’re not selling fantasy sports. They’re selling a way to not watch sports alone. The mechanic was always incidental to the loneliness it solved. When you know your why, you can survive the death of your how.

9. “Can Indian Movies Ever Recreate the Magic of a Housefull Friday Show?”: Avantika Shankar, Vogue India
Link

Hemant Chaturvedi was the cinematographer on Company, Maqbool, Rendezvous with Simi Garewal. He quit the film industry ten years ago. He hasn’t watched a movie since. Instead, he’s spent six years traveling to 20 states, photographing over 1,250 single-screen cinema halls, self-funded, self-produced, documenting what remains of the medium’s golden age.

The stories he’s collected: Raj Kapoor’s secret underground tunnel at Eves Talkies in Meerut so his family could come and go without being mobbed. A theatre in Burhanpur with a sign specifically forbidding bananas (the region is ripe with plantations). One cinema owner in Jalna who refuses to cancel shows despite poor ticket sales because the theatre is named after his grandmother. Projectionists who trained for years before they could run machines on their own. Artists who hand-painted film posters. The chai-wallahs and paan shops that sprouted around every hall.

Some owners have held onto their projectors from the 1920s like treasures. Others have discarded them at nine rupees a kilo.

What I got out of it: The piece started with a memory from Yun Hota Toh Kya Hota (2006), where Chaturvedi dedicated a single lens to each of the film’s four storylines to create distinct visual languages. Naseeruddin Shah told him afterward: “After so many years of being an actor, today I have understood what lenses can actually do.” Creativity born from limitation. Now the industry’s appetite for experimentation has shrunk, Anurag Kashyap is relocating to South India, and Chaturvedi photographs what remains: monuments to a medium that once demanded craft at every level. There’s something about the projectionists who trained for years, the artists who hand-painted posters, the ecosystem of chai-wallahs and paan shops, an entire economy of skill that existed because cinema was hard. And now cinema is easy, which means it doesn’t need those people anymore. The single screens are discarded at nine rupees a kilo not because they failed but because we solved the problems they existed to solve. Progress often looks like this: making things easy enough that the people who made them possible become unnecessary.

10. Sophie Kinsella (1969-2025) Link

Sophie Kinsella, who wrote the Shopaholic series as millions of women’s introduction to the pleasure and pain of financial recklessness rendered comic, died this week of glioblastoma. She was 55, two days shy of 56.

Her final book, published last year, was a novella about a novelist who wakes up in a hospital bed and learns she has a brain tumor. She wrote it after her own surgery. “I just thought people might be curious to know what it feels like to go through this,” she told Robin Roberts. “It’s funny in parts, it’s sad in parts, but I hope it’s full of optimism and love most of all.”

Her family’s statement: “She died peacefully, with her final days filled with her true loves: family and music and warmth and Christmas and joy.”

11. “Does Anyone Really Know You?”: Joshua Rothman, The New Yorker
Link

The feeling of unknownness steals over us at odd moments, sometimes, perversely, when we’re surrounded by people who know us well. Like Levin in Anna Karenina, who notices “a wall between my soul’s holy of holies and other people, even my wife” despite his happy marriage, we become suddenly conscious of an inner sanctum no one has breached.

Rothman works through the intuitive assumptions: that being known is about overlapping Venn diagrams of information, that the people who know a lot about us must therefore know us. But your parents have the naked-baby pictures and still might not see you as you are now. In The Truman Show, millions observe someone constantly from birth, and their circles and his are congruent, yet Truman wouldn’t be wrong to say no one really knows him. Viewership is passive. If passively acquiring knowledge counts as really knowing someone, then Google really knows you.

The shift: maybe “Does anybody really know you?” is the wrong question. Like Schrödinger’s cat, we may not settle into any particular way of being until someone studies us. Other people help us know ourselves, working with us to create a shared idea of who we are. The question isn’t whether we’re known, it’s whether we’ve arrived, in collaboration with people we care about, at a conception of ourselves that we recognize.

What I got out of it: Stanley Cavell on It Happened One Night: Peter can see Ellie right in front of him but can’t let her enter his dream. “To walk in the direction of one’s dream is necessarily to risk the dream.” If they’re to really know one another, they have to merge dreams and reality. This is like putting together night and day. It’s scary. But what haunts me is the earlier observation: that if passively acquiring knowledge counts as really knowing someone, then Google really knows you. We live in an age of unprecedented information about each other. Our step counts are tracked, our purchases logged, our attention measured in milliseconds. And yet the feeling of being unknown persists, maybe even intensifies. Because being known isn’t about data. It’s about someone working to see you, choosing to see you, risking something in the seeing. The systems that know the most about us are often the ones that know us least.

12. “100 Years of Art Deco in Mumbai” — BBC
Link

Art Deco exploded at a Paris exhibition in 1925, and within years it had traveled from Miami’s pastel hotel facades to the mansion apartments along Mumbai’s Marine Drive. The style, geometrical ziggurats, sweeping curves, sunburst designs, circular rooftops, symbolized an unequivocal break from the past, celebrating the dawn of a new post-war modern age.

Mumbai is now home to what may be the world’s largest documented collection of Art Deco buildings (some estimates put it second behind Miami). The style transcended architecture to shape home interiors, furniture, fonts, jewellery, and some of the world’s most iconic cinema halls — Radio City Music Hall in New York, Regal and Liberty and Eros in Mumbai.

What I got out of it: Atul Kumar, founder of the Art Deco Mumbai Trust: “It represented hope, optimism and speed, coinciding with the emergence of the motor car and also concrete as a building material, which when compared with stone, could be worked with in a tenth of the time and fifth of the cost.” The style wasn’t elaborate like Victorian Gothic, it came with a classicism and simplicity that has survived the test of time. Speed as aesthetic. Optimism poured into concrete. What strikes me is that Art Deco was born from constraints too, concrete instead of stone, the need to build fast, the post-war hunger for something new. And yet the constraints produced beauty that outlasted the urgency. A century later, the buildings still stand along Marine Drive, monuments to a moment when modernity felt like a gift rather than a threat. Maybe every golden age is just a brief window when the new technology and the old craftsmanship overlap, before the technology makes the craftsmanship unnecessary.

Also in the mix:

• a16z’s “Big Ideas 2026” (Part 2): The VCs are bullish on “forward-deployed motions” taking AI to the 99% of companies outside Silicon Valley, voice agents managing entire customer relationship cycles, and ChatGPT becoming the AI app store. The most interesting prediction: prompt-free applications arrive. “The chat interface was training wheels. Now AI becomes invisible scaffolding woven through every workflow.”

• DeepSeek’s Forbidden GPUs (Link): DeepSeek is reportedly training its next model on Nvidia Blackwell chips the U.S. explicitly banned from China. The real point: without Nvidia, there is no competitive AI model. The moat isn’t data or algorithms anymore, it’s who can get GPUs through the border.

• TCS buys Coastal Cloud for $700M: Indian IT services giant buying its way into the top 5 Salesforce advisory firms globally. The bigger story: Microsoft announced $17.5B for India, Amazon took its commitment to $35B by 2030.

• “Predict Delinquency With 90% Precision” (Link): A fintech piece about detecting financial stress 7-15 days before a payment is missed. The interesting move: framing collections as relationship management rather than punishment. The early warning window creates space for intervention without the borrower feeling surveilled.

The chicken walks in circles on a mountain slope, generating data for systems it cannot comprehend. Altman attaches himself to work of capable researchers, blesses their discoveries as part of a personal doctrine of “scaling,” sells that doctrine as philosophy. The artist watches her aesthetic diluted into wrong hex codes. The documentary can show the predator but can’t examine the ecosystem. The cinematographer photographs what remains of the single screens, monuments to a craft that once demanded skill at every level, now discarded at nine rupees a kilo. The Art Deco buildings still stand along Marine Drive, hope and optimism poured into concrete a century ago, when speed itself was an aesthetic.

What’s the thread? Maybe it’s that the people who narrate aren’t always the people who know. Maybe it’s Rothman’s question about being known: that viewership is passive, that if passively acquiring knowledge counts as really knowing someone, then Google really knows you. And the systems that record everything are often the ones that explain nothing.

What are you going to do with this?

Until next week.