May 2026
In a Hoover Institution presentation of an NBER paper, Krzysztof Karbownik argues that student composition should be treated as a workplace condition for teachers, not just a background feature of schools. Using Swedish register data on secondary teachers, the study finds that teachers in schools serving more disadvantaged students have higher rates of doctor-certified sick leave, especially for psychiatric and stress-related diagnoses, and that the pattern persists when comparing teachers to themselves as student cohorts change. Karbownik’s mechanism is classroom interaction—conflict, threats, lack of respect and poor climate—rather than workload or school management alone.
Michael I. Jordan argues that modern AI is being framed around the wrong object: an isolated intelligent machine rather than the collective economic systems in which machine-learning components actually operate. In this conversation, the Berkeley statistician and computer scientist says AGI is mostly a PR term, and that the field’s harder problems lie in inference, uncertainty, incentives, markets, and mechanism design. His case is not that recent models are unimpressive, but that prediction and fluent language are only pieces of systems that must be engineered around human institutions.
John Coogan and Jordi Hays use Google I/O to argue that Alphabet is being repriced less as a search incumbent threatened by AI than as a full-stack AI company, though they say Google still has to prove it can turn models such as Gemini Omni and Flash into useful products without cluttering every surface. The Diet TBPN episode also treats distribution as the common pressure point behind several unrelated fights: whether smartphones help explain the timing of global fertility decline, why a small Spotify icon change provoked backlash, and whether podcasts or childcare are eroding the market for serious nonfiction.
Stanford’s CS336 lecture on evaluation, led by Percy Liang with sections from Tatsunori Hashimoto, argues that model evaluation is not a single scoreboard but a choice about what behavior is being measured and for what purpose. The lecture treats perplexity, exam benchmarks, chat preferences, agent tasks, reasoning puzzles, safety tests and realistic professional evaluations as different instruments with different failure modes. Its central claim is procedural: before reading or designing a benchmark, define the object being evaluated, the use case it serves and the trade-offs among difficulty, realism and validity.
Stanford CS336’s data lecture, taught by Tatsunori Hashimoto, argues that training data is both the most consequential and least transparent part of modern language models. Hashimoto says models are not trained on “the internet” in any simple sense, but on static corpora shaped by crawlers, access limits, licensing, copyright risk, filtering, deduplication and conversion choices. The lecture’s central claim is that data construction is a legal and operational pipeline, not a passive input, and that those choices materially distinguish otherwise similar models.
Stanford adjunct lecturers Shervine Amidi and Afshine Amidi present text-to-image model training as a problem of allocating scarce learning signal across the full model lifecycle, not simply choosing a diffusion or flow-matching loss. In Lecture 6 of Stanford’s CME296 course, they argue that practical training depends on emphasizing hard timesteps, adjusting for resolution, using data curricula and representation alignment, then applying post-training, personalization, and distillation methods to improve control and reduce inference cost.
Stanford’s CS336 scaling-laws lecture, taught by Tatsunori Hashimoto, argues that modern language-model scaling is less about accepting a single Chinchilla-style rule than about controlling which training choices drift with size. Hashimoto presents scaling laws as useful empirical tools for choosing model/data tradeoffs, learning rates, batch sizes, sparsity, optimizers, and architectures, but repeatedly cautions that their transfer depends on the regime that produced them. Techniques such as µP and WSD schedules can reduce some uncertainty, he says, while data mixtures, optimizer details, weight decay, architecture changes, and post-training can still break clean extrapolations.
Google’s problem at I/O, Jordi Hays argued, was no longer proving that its AI models are impressive, but making Gemini useful rather than redundant across products investors now increasingly view as part of a full-stack AI business. The TBPN discussion extended that framing across the rest of the show: AI’s value, the hosts and guests argued, depends less on model spectacle than on distribution, workflow integration, economics and adoption by institutions. That distinction ran from Google’s risk of crowding users with Gemini entry points to SendCutSend’s physical capacity constraints, Commure’s push to automate healthcare administration, and METR’s effort to turn frontier-model risk into something auditable.
IBM quantum systems chief Oliver Dial argues that quantum computing has moved from an open-ended promise to a testable engineering timeline. In a podcast interview with Craig Smith, Dial says IBM reached quantum utility in 2023, is targeting quantum advantage in 2026 through public benchmarks, and now sees a credible path to useful error-corrected machines by 2029 after a lower-overhead error-correction code changed the scaling math. His claim is narrower than saying quantum computers are broadly useful today: present systems remain noisy, quantum AI is still toy-scale, and advantage claims will depend on verification against classical methods.
Catherine Ostler tells Andrew Roberts that the Dreyfus affair was not an isolated miscarriage of justice but the eruption of a France already “soaked in anti-Semitism.” Using the Cahen d’Anvers sisters, painted by Renoir as children and later scattered by conversion, war and deportation, she links the military frame-up of Alfred Dreyfus to the social world that first admitted wealthy Jewish families and then turned on them. The paintings, in her account, survive as evidence of both Belle Époque assimilation and the limits of that acceptance.
Bloomberg Technology’s Caroline Hyde and Ed Ludlow frame Google’s new venture with Blackstone as an attempt to turn Google’s TPU capacity into an AI cloud business outside Google Cloud. Bloomberg Intelligence’s Mandeep Singh argues the structure could help Google meet external demand for its chips by shifting more of the data-center burden to Blackstone, creating a TPU-based rival to Nvidia-centered neocloud providers.
Parallel founder and former Twitter CEO Parag Agrawal argues that AI agents are breaking the web’s existing content economics by using publisher and creator material to perform valuable work without tying compensation to that value. In a Bloomberg Technology interview, Agrawal said Parallel’s new Index marketplace is meant to pay publishers, data providers, and independent creators according to their content’s measured contribution to an agent’s completed task, rather than through ads, subscriptions, citations, or flat licensing deals.
University of Cambridge structural biologist Ben Luisi argues that antimicrobial resistance is a permanent race against bacterial adaptation, not a problem that can be solved with one new drug. In a Google DeepMind source, Luisi and colleagues Martin Welch and Marta Wojnowska say tools including AlphaFold, Gemini and Co-Scientist are changing the pace and scope of that race by compressing structural analysis from years to minutes, widening hypothesis generation and surfacing biological patterns researchers might otherwise miss.
Google DeepMind presents WeatherNext, its AI-based global weather forecasting model, as having helped forecasters predict Hurricane Melissa’s Category 5 intensification and landfall in Jamaica three days in advance. Ferran Alet says the model provided a more accurate early signal than previous systems, while National Hurricane Center officials Michael Brennan and Robbie Berg say its confidence supported more aggressive warnings before the storm arrived. Jamaica’s Evan Thompson argues that the added notice gave authorities time to move people out of danger.
JPMorgan global chief information officer Lori Beer told Bloomberg that the bank is already seeing 10% to 30% productivity gains from early AI tools in its technology organization, with agentic systems likely to expand the opportunity. She framed AI less as a headcount-reduction program than as a way to increase capacity for product and engineering work, while warning that the same tools raise cybersecurity risks and require tighter controls, flexible vendor choices, and leadership capable of managing through uncertainty.
Dr. Daudi Jjingo of Makerere University argues that AI-enabled biology can move Ugandan breast-cancer research earlier and closer to where the disease burden is being seen. In a Google DeepMind source, he describes using tools including AlphaFold and AlphaGenome to narrow 15,000 possible sites in a highly expressed breast-cancer protein to 15 candidates for lab validation, a step he says could eventually support vaccine development. The source presents the immediate change not as a finished vaccine, but as local capacity: work Jjingo says once required better-resourced settings abroad can now be done with a laptop and server access.
Bilge Yücel of deepset argues that AI sovereignty is an engineering constraint that has to be designed into a system, not a legal or procurement requirement applied after deployment. She frames sovereign AI around control of data, models, infrastructure, and operations, and shows how retrofits expose hidden dependencies: jurisdiction-crossing data flows, model APIs embedded in application logic, managed services that masked operational work, and systems that cannot be traced or audited.
Cathy Lanier, the NFL’s chief security officer and former Washington, D.C., police chief, recounts filing a sexual harassment complaint as a young Metropolitan Police Department sergeant after a lieutenant repeatedly put his hands on her and retaliated when she objected. In a Tim Ferriss interview, Lanier argues that the formal process sustained her complaint but failed to protect her: confidentiality broke, the case died on a missed deadline, and she was warned the complaint could cap her career.
Ara Khan of Cline argues that agent maturity is less about adding autonomy than about knowing what not to add. In a talk structured around four levels of agent building — from frameworks to state machines, Kanban-managed workflows and cloud deployment — Khan says frontier models increasingly reward simpler prompts, deliberate architecture and visible human control. His central warning is that every extra instruction, abstraction or automation layer can make an agent worse.
Andrew Huberman uses the crude meme “retardmaxxing” to make a narrower argument about self-improvement: introspection becomes harmful when it turns into rumination and replaces action. In a discussion with Chris Williamson, Matt McCusker and Tom Segura, Huberman links that idea to Marc Andreessen and Dana White’s critiques of excessive emotional processing, while acknowledging that the posture can sound dismissive of mental health. The conversation extends the same concern to public judgment of high achievers, where Huberman argues that resentment and secondhand reputational claims can also become substitutes for doing something concrete.
Kenyan duo Akoth Jumadi and Mr. Lu’s TED Countdown performance in Nairobi presents the fusion TED labels East African roots, cosmic trap and celestial R&B through structure rather than explanation. The piece sets Mr. Lu’s compact, image-driven verse about Kashiko against Jumadi’s recurring vocal calls and refrains, making contrast and repetition carry the argument of the performance while the climate-action frame remains outside the lyric itself.
At a16z’s American Dynamism Summit, Erin Price-Wright’s conversation with Michael Duffey and Dino Mavrookas recasts the “trillion-dollar” defense question as a production problem. Mavrookas argues that autonomy and software-first design can make new maritime platforms cheaper, simpler, and faster to build, while Duffey says the Pentagon must change acquisition incentives so industry invests in capacity rather than waiting for government-funded expansion. Their shared case is that defense cannot scale without a broader industrial base built around producibility, commercial demand, private capital, and faster procurement.
Zepto co-founder and CEO Aadit Palicha argues that the company is not mainly a quick-commerce app but a grocery infrastructure business built around dark stores, supply-chain control and the promise of 10-minute delivery. In a Startup School India conversation with Jared Friedman, Palicha traces Zepto’s path from a COVID-era WhatsApp grocery group in Mumbai to a platform handling millions of daily deliveries, saying the decisive moves came from staying close to dissatisfied customers and working backward from speed, quality, selection and price.
Spotify’s Shivam Verma argues that LLM-era personalization requires translating both users and catalog items into forms a model can process alongside language. In his account, Spotify combines long-term user embeddings, Semantic IDs that turn tracks and episodes into token sequences, and soft tokens that project a listener’s profile into an LLM’s embedding space. The aim is a generative recommender that can produce catalog-native recommendations without full fine-tuning, while still relying on traditional ranking layers for production use.