August 2026

Gavin Baker, an investor focused on AI infrastructure, argues that July’s selloff in AI and semiconductor stocks reflected fears about financing and model-layer disruption rather than deterioration in compute demand. He says rising GPU rental rates, accelerating hyperscaler operating cash flow and growing private-lab and open-source workloads support a different reading: capacity contracted at older prices can reset higher and fund more of the buildout internally. The thesis fails, he says, if GPU prices remain depressed, demand weakens, debt becomes essential or regulation blocks new data-center power.

TBPN host John Coogan argues that AI is already changing the terms of work and competition, even where its broader economic consequences remain unsettled. He treats the backlash to Hank Green’s research use as a question of trust, OpenAI’s reported mathematical results as evidence that formally verifiable work may be especially exposed, and AI-assisted solo companies as easier to start but harder to defend. Meta, by contrast, is betting that ownership of models and infrastructure will matter more than renting the capabilities that smaller firms use.

Battery researcher Chibueze Amanchukwu argues that lithium-ion’s roughly 99% cost decline has made batteries core infrastructure for electric vehicles, grid-scale renewable integration and industrial electrification, but that no single chemistry resolves every constraint. In his account, cathode choices such as nickel-based NMC and NCA or lower-cost, more thermally stable LFP determine much of a cell’s cost, performance and fire risk. The next generation, he says, will be defined less by a wholesale replacement for lithium-ion than by chemistries designed around specific limits in materials, safety, energy density and scale.

Sarah Kapnick, a former NOAA chief scientist now advising clients at a bank, argues that companies and cities must stop treating historical weather data as a reliable guide to climate risk. As extreme rainfall and flooding outgrow infrastructure built for past conditions, she says resilience requires updated probability assumptions, added water-storage capacity, community-level planning and financing arranged before disaster forces improvised decisions.

Climate scientist Kate Marvel argues that climate models are conditional accounts of human choices, not predictions of an inevitable future: continued greenhouse-gas emissions produce a hotter, more dangerous world, while stopping carbon dioxide emissions can stabilize warming. Drawing on her work and her book *Human Nature*, Marvel also rejects the idea that scientific rigor requires emotional detachment, arguing that facts and feelings both belong in communicating what is at stake on Earth—the “only good planet.”

Elisabeth Moyer argues that the U.S. power system’s central challenge is no longer simply replacing coal with cheaper wind and solar, but building the transmission needed to connect new generation with rapidly growing demand. Electricity use is rising after decades of stagnation, while variable renewable output, limited storage and slow permitting leave grid operators relying more often on costly plants. In her account, decarbonization depends on electrifying more of the economy—and on whether the country can expand a grid built for a different era.

Climate economist Amir Jina argues that the damage from warming cannot be read from global temperature averages alone: health, crops, labor and energy use respond non-linearly, with harms rising sharply once local thresholds are crossed. His evidence suggests that already hot, lower-income populations face the largest losses because they are closer to dangerous heat and have less access to cooling, resilient infrastructure and other forms of adaptation. While adaptation can reduce those effects, Jina says it remains incomplete and unevenly available.

Extreme heat and wildfire smoke are turning routine sports decisions into public-health and equity questions, argued Lindsey Burghardt, Sam Schmitz, Jennifer Vanos and Graham Zimmerman at Aspen Ideas: Climate. They said schools, leagues and event organizers should assess conditions at the field—not rely on citywide forecasts—then give trained adults clear authority to alter or stop play, with shade, cooling and hydration built into participation. The speakers’ broader case was that safe outdoor activity cannot depend on a family’s ability to buy private alternatives.

Base Power CEO Zach Dell argues that home batteries should be treated not as consumer backup products but as grid infrastructure: company-owned assets that can provide households with outage protection while helping utilities manage periods of peak demand. After raising a $1 billion Series D at a $13 billion post-money valuation, Base is building and operating a distributed battery fleet that it says can raise use of existing grid infrastructure and add flexible capacity without turning the company into a utility.

Bloomberg’s Mark Gurman says Apple’s upgrade program is designed to recast a $1,000-plus hardware purchase as a manageable monthly payment, encouraging customers to replace devices more often. The company can then benefit not only from recurring payments but also from reselling returned devices and reusing their parts, he argues. But the strategy depends on inventory: Gurman says shortages of the latest MacBook Air could frustrate customers ready to trade in.

Dwarkesh Patel argues that if frontier AI revenue grows far faster than compute capacity, the gap will have to emerge in higher margins, higher compute prices or a shift of hardware toward inference. He expects physical supply constraints—from fabrication capacity to wafer allocation—to limit compute growth even as more capable models raise the value of each unit. That dynamic, he says, would favor labs able to extract more useful work from scarce capacity and could deepen concentration in AI infrastructure.

Stanford’s Aakanksha Chowdhery and Azalia Mirhoseini argue that progress in AI is shifting beyond scaling model training toward using more computation at inference time, selecting among candidate outputs, and feeding verified successes back into training. In the opening lecture of CS329A, they trace how post-training made large language models usable assistants and how reasoning models and agents extend that capability into multi-step work. Their central constraint is verification: self-improvement is most tractable where systems can reliably tell whether an answer or action is correct.

Stanford’s Azalia Mirhoseini argues that improving multi-step AI agents requires more than extending a chain of thought: agents must plan, act, use feedback and revise their course. In this CS329A lecture, she examines three approaches to that problem—LATS, which searches alternative action paths at inference; SPRINT, which identifies work that can run in parallel; and SWiRL, which trains next-action choices from offline tool-use trajectories. Each shifts the trade-off among accuracy, latency, inference cost and the reliability of feedback.

Stanford’s Aakanksha Chowdhery argues that longer agent task horizons should not be mistaken for dependable autonomy. Her CS329A lecture distinguishes among benchmarks that measure how long agents can complete tasks at given reliability levels, whether their outputs match professional work, and whether they can synthesize research with verifiable evidence. Across those measures, the central limitation is persistent: agents improve on well-specified tasks but remain unreliable when success requires recovery, organizational context, or judgment about what work needs doing.

Stanford’s Azalia Mirhoseini argues that repeated sampling improves reasoning-model performance only when a system can reliably identify correct candidates—and that verification cannot recover answers the generator never produces. In this CS329A lecture, she examines outcome-based and process-based reward models, rollout-generated labels and verifier ensembles as approaches to the gap between producing a correct solution and selecting it. The central systems question is how to spend a fixed inference budget among generating more candidates, using stronger or more numerous verifiers, and verifying each candidate more deeply.

Stanford adjunct professor Aakanksha Chowdhery argues that train-time scaling can make reasoning models more reliable by feeding verified outputs back into training, but only when a base model can already produce some useful solutions and sampled answers provide meaningful reward variation. In CS329A, she uses STaR, DeepSeekMath’s group-relative policy optimization and DAPO to show how self-generated rationales, relative rewards and RL stability techniques improve performance on verifiable tasks such as mathematics and code. The unresolved question, she says, is whether these methods expand a model’s problem-solving ability or chiefly make solutions it could already reach more likely to appear.

Stanford computer scientist Azalia Mirhoseini argues that test-time compute can turn an existing language model into a more capable system by generating, revising and evaluating many candidate answers after a query arrives, rather than relying on a single response or additional training. The approach works best where outputs can be reliably verified—through formal checkers, unit tests or comparable signals—and where the value of better results justifies added cost and latency. But she cautions that extra inference cannot substitute for stronger pre-training on the hardest problems, and that weak verification can leave correct answers buried among more plausible wrong ones.

ElevenLabs presents ElevenAgents as a platform for turning customer-service procedures into conversational agents that can complete defined tasks, from processing refunds to booking appointments, rather than merely answer questions. The company says teams can define workflows and system access in natural language, simulate them before release, then monitor production conversations through its Spotlight tool for operational problems and test proposed fixes. Configurable guardrails, private-cloud deployment and data-residency options are intended to constrain sensitive actions and support enterprise use.

Peter Santenello, an independent journalist who reports from communities often reduced to stereotypes, argues that America’s deepest division is often isolation rather than ideology. From Appalachia and Native nations to Amish settlements, company towns and drug-affected city neighborhoods, he contends that national labels obscure how differently people live, work and understand their circumstances. His case is not for one hidden “real America,” but for replacing categories with direct contact.

Kiron Skinner argues that the Trump Doctrine seeks to make U.S. power more durable by replacing aid conditioned on U.S. social and political preferences with trade and investment, pressing allies to bear more of their own defense, and abandoning Cold War assumptions about how to compete with China. She contends that China’s integration into global commerce makes a democratic coalition model inadequate, while U.S. institutions remain poorly equipped for long-term strategy. Skinner acknowledges that the doctrine’s soft-power framework and several of its policies remain unfinished, and says its durability will depend on whether the political coalition behind it outlasts Trump.

Kiron Skinner argues that U.S. foreign policy has lacked an organizing doctrine since Cold War containment, even as the postwar order built under exceptional American power has frayed. She casts Donald Trump as the political figure who, in her account, made a strategic break possible by recognizing the rise of competing powers and constituencies alienated from the old order. Skinner’s case identifies the exhaustion of that framework more clearly than it defines the commitments of an emerging Trump doctrine.

Kiron Skinner argues that Donald Trump’s foreign policy was organized around sovereignty: states should secure their own borders, bear a greater share of their own defense, and deal with one another on reciprocal terms rather than through open-ended U.S. commitments. In her account, that logic drove Trump’s pressure on NATO, preference for flexible regional partnerships over new permanent alliances, and rejection of nation-building. It also recast China as a strategic competitor rather than a country expected to converge with the liberal international order through economic integration.

Felix Brooks-Church, co-founder of food-fortification enterprise Sanku, argues that “hidden hunger” can be addressed by adding essential nutrients to staple foods at the local mills that produce them. His model combines mill-compatible dosing machines, remote monitoring and local premix supply, but he says durable national scale ultimately depends on governments mandating fortification and enforcing standards through testing and inspection.

Hussein Aboubakr Mansour, an Egyptian-born scholar, argues that Palestine has become a self-reproducing system in which real Palestinian suffering is converted into political legitimacy, institutional standing, regional leverage and moral meaning for actors far beyond Israel and the Palestinians. Drawing on his upbringing in Egypt and his essay on the political economy of the symbol, he contends that Arab regimes, Islamist movements, rival states, Western institutions and NGOs can all be both users and captives of a conflict whose continued centrality serves distinct interests.

Two Minute Papers’ Karoly Zsolnai-Fehér argues that DeepSeek’s V4-Flash-0731 shows how much capability can be added through post-training rather than a larger base model. DeepSeek reports that the 384B-parameter model retains the prior Flash architecture and size but substantially improves on agentic and software benchmarks, including a 54.4 DeepSWE score versus 7.3 for Flash-Preview. Zsolnai-Fehér’s case is that improved planning, verification and error recovery—not new underlying capacity—account for the shift.

Liad Yosef and Ido Salomon argue that MCP Apps gives AI assistants a way to render service-specific interactive interfaces rather than reducing every result to text. Built from the MCP-UI project and being developed as an MCP extension, the approach lets a server return an app resource that a compatible host renders in a sandbox, while clicks and other actions return through the host for further orchestration. They present it as a portability and distribution model, though state management and interoperability remain active work.

Nathan Labenz argues that China’s weaker safeguards and disclosure practices do not validate the American claim that frontier AI safety requirements are futile because Beijing will not slow its own developers. Reporting from China, he finds a university-centered safety ecosystem increasingly engaged with Western work on deception, evaluation awareness, interpretability and hazardous capabilities, alongside a state able to delay or constrain domestic deployments. China remains behind the leading US labs, Labenz says, but its record complicates a simplistic “but China” case against American safety obligations.

ElevenLabs presents ByteDance’s Seedance 2.5 as an AI video model built to reduce the stitching and full-shot rerolls required in earlier workflows. Compared with Seedance 2.0, it generates continuous 4K clips of up to 30 seconds, accepts 50 reference assets rather than 12, and adds 3D-reference guidance and region-level editing. The company argues that these changes give creators more control over continuity, framing and corrections without having to regenerate an otherwise usable shot.

The discussion turns on whether some women would prefer to share a wealthy or high-status partner rather than have an exclusive relationship with a less desirable one. An unnamed speaker argues that monogamy protects the value of exclusivity while distributing partnership more broadly among men; Williamson questions whether that arrangement primarily serves men. The speaker’s narrower empirical case is that, even under looser sexual norms, most people still form largely exclusive pair bonds, usually serially rather than for life.

Researchers behind Hybrid Imitation Learning argue that a parkour controller can learn human-like movement from just 19 YouTube clips—about 30 seconds of footage—without becoming locked into replaying them. Their method trains a simulated character both to imitate motion in its obstacle context and to navigate new courses, producing stronger motion tracking than the tested baselines while retaining a reported 0.74 task-completion rate in noisy unseen scenes. The paper also shows the trade-off: a task-reward-only configuration completes more courses, and HIL can still fail or recover with unnatural poses.

Stanford neuroscientist Andrew Huberman says his view of psychedelics changed not because he became an advocate, but because clinically supported sessions gave him access to psychological material he had not been able to reach through therapy, journaling, or the drugs separately. He argues that MDMA and psilocybin served different roles in that work—one making difficult emotions bearable, the other exposing patterns and absences—and that their value depended on trained guidance and follow-through in ordinary life.