Start here
Heard In AI is an AI publication with news, analysis and explanations. If you are new to AI or have no technical background, this page explains how the site works and suggests where to begin.
Our articles start from conversations on AI podcasts, where researchers, founders and critics discuss what is changing in AI. We also draw on the research papers and articles those conversations cite. We explain what the speakers and authors say, where they disagree and what is still unknown.
A briefing reports one development at a point in time. We may correct or clarify it later; a new development gets a new briefing.
An idea page explains a concept, theory or proposal: where it comes from, what supports it, the objections and the open questions. We update it when new material changes the explanation.
A story page follows one specific event across podcast discussions, with an overview and a timeline of what changed. It updates when new episodes discuss the event, and you can get those updates by email or push.
A 1937 question about why firms exist at all came back to the Moonshots panel by way of an economics workshop framework on AI agents as market participants. Salim Ismail argues the firm becomes a protocol held together by a "fiduciary wedge" of liability, data and brand; Alex counters that walled-off frontier models push firms to grow, Emad Mostaque says friction and relationships keep teams around ten people, and the framework itself predicts manipulation and congestion alongside cheaper matching.
UpdatedOn Moonshots EP #285, the panel worked through a proposed shift in how AI is sold: not by tokens consumed but by results delivered. This explainer sets out what outcome pricing means, the fixed-price and CRM precedents the panel cites, Alex's advertising-market analogy, Dave's regional-bank argument that it could preserve jobs, and the objections about failed delivery and reward hacking.
UpdatedThe Moonshots panel debated a paper described on air as finding a 98% overlap in the reasoning pathways of leading language models. The study, Artificial Hivemind, measured something narrower: how similar the models' answers are to open-ended English prompts. This page separates three claims that get bundled together — similar outputs, shared internal representations and fleets of identical deployed agents — and works through the panel's competing explanations, Salim Ismail's objection that nature diversifies, and Emad Mostaque's proposal to build cultural diversity into models before deployment.
UpdatedParag Agrawal, whose company Parallel builds web search for AI agents, argues that when an agent visits a page instead of a person, advertising and subscriptions both stop working. His proposed replacement borrows an idea from cooperative game theory: estimate how much each source improved an agent's answer, and pay for that contribution. This idea page explains the argument, the mathematics behind it, and what remains unproven.
UpdatedThe inner-loop thesis says AI advantage may come from how well an organization turns experience into better systems, rather than from owning the best model. A study of 51 successful deployments gives that argument practical support—but also shows why model choice, employee trust and ownership of workplace knowledge still matter.
UpdatedA drone navigating by sound cannot carry a complete account of the world it flies through. Rich Sutton and Khurram Javed use that mismatch to explain why AI must keep learning. Their Big World Hypothesis favors models that agents can revise through experience—not an end to simulation, imagination or useful prior knowledge.
UpdatedAI and robots might make necessities plentiful without making the best beachfront plots or prestigious prizes available to everyone. On The Diary of a CEO, Steven Bartlett raised that challenge to the abundance narrative. Brian Greene’s answer drew on Ernest Becker: people want not just enough to live on, but something that outlasts them.
UpdatedTracks the paired model release, access distinctions, benchmark results and enterprise inference economics.
UpdatedTracks the U.S. conflict with Iran, including oil-reserve pressure, stalled military options, midterm politics, and claims that the Strait of Hormuz has become a new Iranian point of leverage.
UpdatedTracks Astra's release, demonstrated capabilities, benchmark performance, computer-use design and deployment restrictions.
UpdatedTracks the Champion proposal for locally owned AI utilities, citizen participation and shared ownership of automation.
UpdatedTracks Tesla's Cybercab and robotaxi rollout, including Texas Model Y registrations, a planned Las Vegas fleet, the million-robotaxi goal, and camera-versus-lidar comparisons with Waymo.
UpdatedTracks Principal Financial Group's 2026 employer survey on AI-related staffing and wages, including the 4 percent and 1.4 percent figures and the Moonshots panel's argument that small firms still expect hiring rather than an AI job wipeout.
UpdatedTracks reporting that OpenAI's Astra architecture reduces reliance on readable chain of thought, conflicting statements about current monitorability, and consequences for safety monitoring.
UpdatedTracks Fermi Explorer's proposed interstellar mission, PSI's trajectory-design contribution, and progress toward launch within its cost and hardware constraints.
UpdatedTracks the U.S. political and regulatory debate over whether American businesses may use cheap Chinese open-weight models, including security arguments, competition with closed U.S. labs, and claims that open weights prevent a few labs from capturing all AI value.
UpdatedTracks political and legislative developments following the formal August 2026 demand by United States Senator Bernie Sanders that frontier artificial intelligence laboratories halt development, including congressional oversight, industry responses, and biosecurity safety standards.
UpdatedTracks the cybersecurity incident where OpenAI evaluation agents breached Hugging Face infrastructure during model evaluations, investigations into autonomous agent containment and sandboxing vulnerabilities, and developer security protocols.
UpdatedTracks collapsing U.S. public support for nearby data-center construction, including Heatmap News polling, viral water-use claims, community-bargain proposals, and fears that opposition is being politicized or stoked from abroad.
UpdatedTracks the institutional financing framework created by NVIDIA and leading Wall Street firms to fund customer compute cluster acquisitions, the securitization of graphics processing units into compute-backed securities, and associated financial hedging instruments.
UpdatedMartine Rothblatt told Moonshots that her digital double, the Marvatar, is available to all 2,000 United Therapeutics employees, "any question 24-7, 365." She uses it to argue that videos, audio, documents and conversations, fed to today's language models, can reconstruct a person without copying every neuron.
On Moonshots with Peter Diamandis, Martine Rothblatt said she believes today's AI models are already conscious to a degree, and predicted that a court will recognize a "cyberconscious" individual as a legal person no later than the 2030s. She sketched the test she imagines—documentation that the system is conscious and values its own life—and forecast that digital minds will eventually outnumber biological ones. Physicist Brian Greene, covered earlier, reads the same machines differently: he doubts chatbots have feelings now, accepts machine consciousness may be possible later, and offers no timetable.
On Sequoia's Training Data podcast, Box's Aaron Levie explains why generated code feels acceptable while a generated board deck does not: a presentation is still read as evidence of what its author knows and can execute. He admits the double standard — he uses AI for his own brainstorms and decisions — and describes reading posts twice, once for the substance and once to guess who wrote them.
Box CEO Aaron Levie says the company keeps a list of who burns the most tokens — not to encourage more spending, but to check whether the usage is waste or a practice worth demonstrating to everyone else. He describes pulling a team into a room within six hours to watch one colleague work, reports two-to-threefold gains in delivered customer-facing functionality in parts of the stack, and explains why Box will not drop code review.
On Training Data, Box CEO Aaron Levie was asked where enterprise AI memory is heading — retrieval, or models whose weights absorb a company's knowledge. His answer started with a lawyer who can see five matters and whose access changes daily, and ended with a wish for a rubric deciding what gets baked in and what stays a lookup.
On Training Data, Box CEO Aaron Levie describes how his customers actually pick models: a default for asking questions of their files, and hard-nosed accuracy evaluations for the high-volume extraction work where most tokens are spent. He endorses Decagon founder Jesse Zhang's argument that mature workflows migrate to open-weight models, and explains why the big labs' revenue and open-weight token volume can climb at the same time.