TruthArchive.ai - Related Video Feed

In this episode of The Koerner Office, Chris Koerner shares insights from conversations with about 30 entrepreneurs who are actively weaving AI into their businesses. The discussions reveal prevalent use cases: ChatGPT as a general thought partner, Claude as a coding ally, Perplexity for real-time research, and Gemini for Google Workspace workflows. The takeaway is that while good ideas abound, execution remains challenging; the path to success rests on building solutions that meaningfully help people increase revenue. A key thread is AI’s current appeal as an unbiased tool. The host notes AI’s tendency to avoid click-driven incentives, contrasting it with Google search or sponsored studies. This perceived impartiality is described as a superpower, though he acknowledges potential shifts in incentives over time. The conversation uses examples about diet studies to illustrate how hidden sponsorship can distort information that people rely on. The episode dives into nuanced portraits of several tools and their best-fit domains. Claude shines in coding, Perplexity excels at deep, real-time research, Grok is praised for speed and responsiveness, and Gemini stands out in Google-workflow integration. The speakers discuss context windows, file-based context, and how tools like Cursor may outperform others in specific tasks, while others offer broader accessibility. The dialogue highlights a push-pull framework: usability versus functionality, suggesting the unicorn is a product that harmonizes both. Beyond tool ratings, the hosts brainstorm practical applications and experiments. They debate mediation using AI, revenue-focused automation for mediators, and using data-driven billboard experiments to test marketing impact. The discussion culminates in a meta-observation: the strongest AI-enabled products directly tie to revenue generation, especially in sales, lead management, and customer support. The episode closes with plans to translate these insights into learnings for listeners and future experiments.

The episode centers on the rapid, hands-on potential of Google’s AI tools and the idea of building AI-powered apps with minimal code. The hosts explore how AI Studio and the Gemini ecosystem let users prototype and deploy AI-powered applications in minutes, stressing the accessibility of “vibe coding” where a single prompt can yield a working app. The conversation emphasizes that the barrier to building AI products has collapsed, making experimentation feasible for individuals and small teams, and it highlights how modern AI capabilities enable practical, real-world outcomes rather than abstract demos. The speakers acknowledge both the excitement and the caution required, noting that the best opportunities often come from solving specific, known problems within a person’s domain, such as a hairdresser crafting a tailored AI haircut experience or a travel workflow that orchestrates complex logistics rather than merely booking a flight. The dialogue delves into strategic advice for aspiring builders: start with problems you understand, embrace the idea that big success can come from many small, iterative prompts, and recognize the value of niche specialization that can scale via packaging multiple tools for a targeted audience. They discuss the “thousand papers” of possibilities created by a single platform and warn against overreaching—start with a focused, viable product, test, iterate, and expand as user needs emerge. They also examine how to market AI apps in a world of abundant experimentation, suggesting social-first outreach or bundled solutions for specific personas, as opposed to chasing universal “everything apps.” The podcast touches on broader implications for the tech landscape, including how AI is reshaping content creation, video and image analysis, and voice or browser agents. The speakers reflect on the pace of innovation, emphasizing that tools like Gemini enable true, end-to-end pipelines—analyzing video, extracting insights, and generating customizable reports in real time. They contemplate a future with “infinite content remixing” and discuss how large platforms, search, and AI modes will influence mainstream adoption. Throughout, the conversation stresses the importance of agency, resilience, and problem-solving over mere familiarity with technologies, arguing that the current moment makes it possible to build and ship more cheaply and quickly than ever before, while cautioning about the risks of hype and misaligned use cases. The episode includes a direct nod to a well-known book, Range, to illustrate the value of broad, cross-domain thinking over narrow expertise. It closes with a call to action for listeners to try AI Studio and engage with the developers, emphasizing that the most important takeaway is to begin experimenting now, even if the first attempts are imperfect.

The episode opens with a rapid-fire tour of today’s tech and business headlines, starting with a viral Campbell Soup internal recording in which a company executive allegedly disparages the product and its customers. The hosts frame the incident as a PR crisis that reveals deeper questions about hiring, corporate culture, and product strategy, while weighing how senior leadership should respond publicly and internally when a scandal erupts. The conversation then shifts to Nvidia versus OpenAI in the AI arms race, with Michael Burry’s critique of Nvidia’s depreciation and earnings practices drawing pushback from Nvidia and shifting attention to how AI hardware costs, scaling, and accounting policy shape market expectations. The panel uses the moment to discuss how large language models (Gemini, ChatGPT, Perplexity) compete for speed, context, and real‑world utility, with Tom outlining how “who powers your agent” matters as much as which model is fastest. A live comparison of Gemini 3 against ChatGPT, including user experiences and source‑quality considerations, underscores a larger trend: AI usefulness is defined by integration into everyday workflows and trusted data sources, not just headline performance metrics. The show pivots to policy and finance, highlighting the AI Super PAC campaign to push uniform federal AI regulation and what that implies for consumers, startups, and incumbents. The hosts debate whether centralized federal rules would help or hinder innovation, and they connect this to broader debates about liability for AI errors, the underwriting of such risks by insurers, and the difficulty of equitably pricing coverage for rapid AI deployment across industries. The conversation then broadens to macro trends: insurers warning they may not cover AI mistakes as automation scales, and housing and inflation dynamics that influence insurance costs, construction inputs, and affordability. Brandon and Tom trace how building costs, labor shortages, and supply chains feed into higher premiums and how policy levers—ranging from energy policy to “behind the meter” infrastructure—could ease consumer burdens. On Florida’s property‑tax debate, DeSantis’s proposals to eliminate or reduce homestead tax are weighed against potential consequences for homeowners risk and state revenues, with panelists offering nuanced takes about who would benefit and how it could shift regional investment and housing markets. The second half of the episode shifts to education and employment, highlighting Bloomberg and Cleveland Fed data showing college grads facing rising unemployment in a digitizing economy, and the ongoing debate about the value of degrees versus trades in a tech‑driven market. The hosts explore how to prepare for a future where AI handles more routine tasks, stressing the need for problem‑solving, leadership, and real‑world skills. The Thanksgiving close provides a personal capstone: a reminder to practice gratitude, reflect on plans for 2026, and invest in self‑improvement, with a call to attend the Business Planning Workshop and to stay curious about how policy, technology, and markets interact.

OpenAI's new browser, ChatGPT Atlas, integrates advanced AI models, particularly ChatGPT, directly into the core browsing experience, moving beyond traditional browser add-ons. Developed by browser veterans Ben Goodger and Darin Fisher, Atlas aims to transform web interaction by allowing users to command the internet using natural language. This innovation is timely due to the rapid progression of AI capabilities, enabling compelling user experiences that were previously impossible. Atlas features an "agent mode" where ChatGPT can take actions on the web on the user's behalf, such as synthesizing data into charts, reviewing documents, or managing cloud services. This agent operates in its own workspace with segmented tabs, offering a controlled environment where users can observe or halt its actions, addressing concerns about AI autonomy. The browser also boasts enhanced memory features, allowing it to recall past browsing activities and personalize future interactions, like remembering preferred airlines for flight searches. The design philosophy behind Atlas emphasizes simplicity and accessibility, aiming to make complex computing tasks more approachable for non-experts. It features a unified "one box" input for both navigation and AI queries, streamlining the user experience. The "Ask ChatGPT sidebar" provides instant assistance, summarizing pages, answering questions, or initiating agent tasks without leaving the current site. This fosters serendipitous discovery and helps users navigate the web more effectively, breaking free from content "rabbit holes." Technically, Atlas is built on Chromium (referred to as "Owl") but with a unique architecture that separates the browser's core rendering from the Atlas application, enhancing stability and performance. This allows for features like "scrolling tabs" that efficiently manage thousands of open tabs without clutter or performance degradation. The team also leverages AI tools like Codex for accelerated product development, even enabling non-engineers to contribute code. OpenAI views Atlas as a long-term investment, with plans for multi-platform expansion (Windows, mobile) and continuous feature development, aiming to make AI beneficial and accessible to all humanity by delegating "toil" to intelligent agents.

The Moonshots roundtable centers on Gemini 3 and what its breakthrough means for everyday life, work, and the global economy. The panel emphasizes that Gemini 3 marks a step function change: not just faster or smarter, but capable of multimodal reasoning, autonomous action, and dynamic user interfaces that weave images and interactive widgets into responses. The guests explain that the real impact comes from a shift toward AI that can plan, execute, and optimize across complex tasks, lowering barriers to software development and enabling humans to work with machines as collaborators rather than mere inputs. They frame Gemini 3 as a potential turning point where people can build software or even entire businesses by talking to an AI, dramatically accelerating problem solving in math, science, engineering, medicine, and beyond. A central discussion item is the “Vending Benchmark” and other practical tests that translate lofty AI capabilities into real-world economic engines. Gemini 3 reportedly delivers superior profitability in simulated AI-driven businesses, outperforming rivals on long‑term planning, multi-step reasoning, and email-like interaction with other agents. The panel argues this foreshadows broader shifts: AI-enabled automation could spawn new companies with few or zero human employees, reframe employment, and create an AI-enabled economy where decisions and operations run with minimal human toil. The conversation also grapples with risk, safety, and governance as capabilities scale. They discuss layered defenses against AI-assisted biosafety threats, the need for co‑scaling safety measures with AI power, and the challenges of open-source models in security contexts. OpenAI’s GPT‑5.1 and Google’s Gemini trio surface as competitive accelerants, each pushing new business models for enterprise and consumer use. The hosts acknowledge the social and regulatory questions tied to abundance: how to ensure affordability, access, and benefit distribution while avoiding runaway wealth concentration. Looking ahead, the group muses about the broader implications for education, healthcare, housing, and transportation. They envision a world where AI-driven tools dramatically reduce costs and unlock universal access to essential services. The dialogue closes with a pragmatic optimism: as intelligence per cost falls by orders of magnitude, humanity should steer these gains toward solving grand challenges, while maintaining vigilance about safety, ethics, and equitable distribution. ], topics Gemini 3, AI benchmarks, autonomous agents, AI-enabled software development, vending benchmark, OpenAI GPT-5.1, Prometheus project, biosafety and alignment, regulatory and economic implications, education and healthcare transformation, universal abundance otherTopics Moonshots podcast format, Silicon Valley AI race, AI in daily life, safety and governance, impact on employment, future of work, AI-powered manufacturing, AR/AI interfaces, scalable AI safety booksMentioned Rainbow's End

Is AI slowing down? This episode with Nathan Labenz and Erik Torenberg wrestles with that question by separating immediate usefulness from long term progress. They discuss Cal Newport's skepticism about near term risk while arguing the pace of capabilities is still healthy, with GPT-5 offering meaningful gains over GPT-4 in areas like extended reasoning and context handling, even if simple QA comparisons may obscure the difference. They emphasize that progress today comes not only from bigger models but from better post training, tool use, and smarter prompting. Beyond language, the conversation covers non-language modalities: image, biology, robotics, and scientific problem solving. The Google Gemini example and the IMO gold problems illustrate that modern AIs can reason, hypothesize, and even suggest breakthroughs in fields like virology and antibiotics. An MIT study on new antibiotics shows how AI-driven discovery can yield novel mechanisms of action. They discuss the value of extended reasoning, multi-step prompts, and structured workflows that let a single model perform tasks previously reserved for teams of researchers. On jobs and productivity, the Meter study is debated: engineers may feel faster but actually move slower, and the real world impact depends on how people and companies adopt AI tools. The speakers discuss customer service, software development, and high volume tasks where agents can resolve tickets or generate code with far less cost than human labor. They also warn about reward hacking, misalignment, and the unpredictable behavior that can emerge as task length doubles, underscoring the need for safety, governance, and monitoring. Looking ahead, the conversation touches open-source versus frontier models, US-China dynamics, and whether AI progress will be spurred by competition or collaboration. Labenz argues that progress will continue, that a positive vision matters, and that education and creative work, like writing or biology papers, can benefit from AI as a learning partner. They advocate for broad participation, from philosophers to fiction writers, to shape a future where technology expands abundance rather than concentrates risk.

In this interview, Anthony Pompliano speaks with the Synthesis team, including Josh Dahn, Chrisman Frank, and Ana Lorena Fabrega, about their innovative approach to education. Josh shares his background, detailing how he co-founded Ad Astra School with Elon Musk to create a better educational experience for children. Synthesis emerged from this initiative, focusing on collaborative problem-solving through engaging games that challenge students to think critically and work as teams. Chrisman emphasizes the need for innovative educational methods, noting the limitations of traditional schooling. He describes the unique experiences at Synthesis, where students learn to navigate complex problems and develop essential skills for the future. Ana, a former elementary school teacher, highlights the shortcomings of conventional education, expressing her excitement about Synthesis's ability to foster creativity and meaningful learning. The program promises to transform children into world-class problem solvers by engaging them in simulations that mimic real-world challenges. Parents are drawn to Synthesis due to their dissatisfaction with traditional education, seeking a more relevant and engaging learning experience for their children. The team believes that by teaching kids to think critically and collaboratively, they are preparing them for future success, regardless of the specific paths they choose. The conversation underscores the importance of rethinking education to better equip students for the complexities of the modern world.

Google's AI turnaround is real: Gemini just hit the number one app in the app store, and the internal energy at Google has changed, says Robby Stein, VP of Google Search. The company maintains that its core mission—making information universally accessible—remains, but the AI moment has created a tipping point where models can genuinely deliver for consumers. The shift is not about replacing search but about multiplying its reach through AI overviews, AI mode, and multimodal tools like Lens, all designed to deliver faster, more accurate answers while weaving live data into results. There's three big components to what we can think about AI search: AI overviews at the top, which provide quick answers; multimodal and Lens for visual search; and AI mode, which binds it all into a single conversational experience. AI mode uses all of Google's information, including 50 billion products in the shopping graph updated two billion times per hour, 250 million places in Maps, and the entire context of the web, so you can ask anything and follow up. It can be accessed at google.com/ai and is integrated into core experiences so you can ask follow-ups directly or take a photo and go deeper in AI mode. Stein emphasizes three big features of AI search: AI overviews at the top, which provide quick answers; Lens for visual queries; and AI mode, which binds it all into a single conversational experience. He notes that Google’s data backbone—shopping graph, Maps, finance, and web signals—allows the AI to understand context and surface authoritative sources. The interface aims for a consistent, simple experience; you can start in core search and have follow-ups, then dive deeper in AI mode or Lens as needed. The goal is to make the transition between AI and traditional search seamless rather than a toggle. Looking ahead, AI is expanding into inspiration and multimodal creativity, with live AI search and 'AI corner' experiments such as visual inspiration boards and Nano Banana-like tools. The team emphasizes testing with labs and trusted testers, then scaling to IO launches and global rollout. Public examples include live conversational search and ongoing integration across products, all aimed at giving users effortless access to knowledge with reliable sources.

Could a computer truly feel happiness, or is embodiment the irreplaceable spark of being human? Einstein’s happiest thought about weightlessness frames the opening question, as Blaise Agüera y Arcas argues that the brain is fundamentally computational: sensations are encoded as neural spikes, and a computation could, in principle, generate experiences even without a body. The talk moves from embodiment to whether AI, including transformers, can be a genuine experiential being rather than a solver of equations. They note VR can evoke real anxiety and delight, suggesting the boundary between human consciousness and machines may be more porous than we think. They also discuss lock-in, where entrenched symbioses with hardware shape what comes next. They turn to capabilities: can neural networks do physics like Einstein, and will AI threaten physicists’ jobs? The guests share experiences using large language models for math and physics, rearranging equations and exploring new angles. They contrast this with Apple’s cubit paper on reasoning; the appendix lists prompts, and Bratton and Agüera y Arcas discuss how prompts can produce general strategies, challenging a claimed limit. They stress the need for human baselines when evaluating AI reasoning and warn against equating language skill with true understanding. Beyond theory, the dialogue explores AI’s role in education, therapy, and lifelong learning. Ipsos data shows greater AI optimism in developing countries, while developed regions worry about disruption. They describe classrooms where prompts guide problem solving and data generation, arguing that teaching must adapt to AI’s capabilities. They discuss biology and life, comparing computation, life, and intelligence, and envision collaboration rather than competition between human and machine minds. The conversation also touches on poetry and art as collaborative practices in science, and the value of improvisation in human–AI partnerships. Philosophical questions anchor the talk: what is life, what is intelligence, and how do information, function, and purpose relate? Schrödinger’s What Is Life? is cited, and the speakers discuss computation as a substrate‑independent function, using terms like computronum and copyrum. They contemplate whether universal compute or universal access could democratize expertise, and they describe collaborations that blend science and art, improvisation, and noise as engines of creativity. The episode ends with a call to reflect on the future of intelligence as humans and machines increasingly collaborate.

Transforming education with AI and mastery-based learning forms the central thesis of this dialogue, sparked by a stark claim: the U.S. spends about a trillion dollars on K-12 education with disappointing returns. Liemandt traces a career arc—from Trilogy's AI roots to Alpha's education experiment—and insists the fix is simple in theory: go back to basics, engage learning science, and deploy AI tutors to accelerate mastery. He argues that with mastery-based progress and a two-hour daily learning block, most eighth graders could perform as today’s top decile. The promise is rapid learning paired with motivating, time-efficient content, not longer school days. Central to the plan is a reconceived learning architecture: mastery, not time-based progression; game-like motivation; and AI that adapts to each student. The education science underpinning this idea (including Bloom’s two-sigma result and the zone of proximal development) implies that a one-on-one tutor guiding a student to mastery can raise performance two standard deviations beyond typical outcomes. Liemandt emphasizes the practical mechanisms: fact fluency in math, avoidance of careless errors caused by working-memory overload, and remediation through targeted pre-requisites. GenAI is framed as a microscope enabling precise measurement of what a student knows and needs to learn next. Alpha's model extends beyond academics into life skills and athletics, with a structure that redefines the school day. Two hours of AI-guided lessons become the engine, while the remaining day features workshops, sports, and leadership development. The program uses extrinsic motivators, including cash incentives and time back, to boost engagement, arguing that motivation is the critical lever for all students. The school also advocates a spectrum of campuses—from high-end Alpha exemplars to lower-cost, scholarship-heavy affiliates—designed to scale while preserving the core mastery-based, personalized approach and a culture of high standards and high support. On implementation, Liemandt acknowledges deep frictions: parental demand, teacher prep and buy-in, and the enormous reengineering required to replace standard classrooms with AI-enabled, tutor-led learning. He insists teachers become guides and mentors rather than lecture-givers, assisted by AI tutors that handle mastery content. He cites his daughter\'s experience—gradual remediation of math gaps that unlocked sustained achievement—as a human-scale example of the model's potential. He envisions a future where on-device AI brings personalized mastery to a billion kids, supported by philanthropy, partnerships, and Stanford-led learning-science research.

Sal Khan discusses the potential of AI, particularly through tools like Conmigo, to transform education positively. He argues that while concerns about cheating exist, AI can provide personalized tutoring for every student and assist teachers, enhancing learning experiences. Citing Benjamin Bloom's Two Sigma study, he emphasizes that one-on-one tutoring can significantly improve student performance. Conmigo can engage students in various subjects, offer guidance, and facilitate deeper understanding of literature and writing. Khan believes that AI can address educational disparities and enhance human intelligence, urging active participation in shaping its use to ensure positive outcomes while implementing necessary safeguards.

Education could become a tutor for every learner, and Sal Khan presents a path there. The origin story starts with tutoring his 12-year-old cousin Nadia across distances while he worked at a Boston hedge fund, a seed that grew into Khan Academy fifteen years ago as a not-for-profit response to misaligned incentives in education. He notes how edtech was once overlooked by venture capital, and how Khan Academy demonstrated a real demand for scalable, tech-enabled learning. The conversation then traces the choice to stay nonprofit, despite market pressures, and how that stance led to more mission-centered impact even as early control questions arose. It also chronicles the Khanmigo project, sparked by a 2022 OpenAI outreach, and the decision to pursue AI with safeguards: an assistant built on Khan Academy content, moderated for under-18 interactions, and designed to make processes transparent. The team framed risk—hallucinations, bias, cheating—as features to be mitigated rather than barriers to adoption, integrating Socratic tutoring with state-of-the-art technology. Sal describes Khanmigo’s practical uses, from answering questions and giving guided explanations to providing a feedback loop that emulates a personal tutor. He shares a demo of a chat about Einstein and E=mc^2, where the AI clarifies concepts while the human teacher stays involved. He envisions the AI as a teaching assistant that can draft lesson plans, rubrics, and assignments, then report back to teachers with full transparency about student work. The Newark, New Jersey example illustrates equity gains as Khanmigo helps students who cannot afford tutoring, and he cites Con World School with Arizona State University, where high school students spend roughly an hour to an hour and a half per day in Socratic dialogue plus collaboration on boards and clubs. He emphasizes that AI can reduce teachers’ administrative load—planning, grading, progress reports—without replacing human guidance—and that memory, continuity across years, and family involvement could be improved. Globally, he argues the U.S. should lead with experimentation and growth mindset while learning from others, and that AI co-pilots could transform both teaching and learning, expanding access to world-class education and reimagining the role of teachers as facilitators in a more productive, humane system.

A private school in Austin is reshaping the traditional classroom by letting AI run the core academics for two hours each day, while students pursue passion projects the rest of the time. The idea grew from MacKenzie Price’s experience at Stanford and a simple moment with her daughter—school felt boring to a bright, curious child. Price says the goal is to meet every student exactly where they’re at, with the right level and pace of learning, and to leverage artificial intelligence to unlock momentum that previously stalled in a standard classroom. Alpha School’s model centers on two hours of AI-guided core subjects, after which human guides shift to mentorship and life-skills development. There are no traditional lectures during the morning; the AI behind the scenes personalizes pace and content, while afternoons are filled with workshops on leadership, teamwork, communication, grit, entrepreneurship, financial literacy, storytelling, and public speaking. The school emphasizes time as a resource and uses a structured Pomodoro-like routine to move students through math, reading, language, and science, with the goal of turning earned time into opportunity. Assessment and personalization are built into the system. The MAPS assessment guides adaptive lessons, and AI analyzes progress, reading explanations versus guessing, then creates targeted plans. The platform uses no chatbots for instruction; instead it feeds results into tailored curricula and confidence anchors that connect mastery across subjects. Price cites Taylor Swift-inspired songs for AP US history and Teacht Tales as reading customization. Projects include a life-skills-based food truck and a 168-hour weekly planning exercise; GT School extends these ideas to gifted students with faster learning and advanced projects. Expansion plans include 12 new campuses this fall, 25 next year, and about 15 more in 2026, aided by deals repurposing existing buildings. Price notes she published a book earlier this year titled Super Agency exploring AI's impact on learning. Costs are a factor: Alpha pays teachers at least $100,000 annually and spends roughly $10,000 per student per year on the AI platform, while educational savings accounts could reduce out-of-pocket costs to around $5,000. She sees AI as elevating human intelligence, not replacing it.

Codie Sanchez hosts a guest who has built one of the leading AI companies that takes our human mind and recreates it online. The host asks, 'If any video you see online can be AI generated, how do you know what to trust?' The guest insists that 'the most unique thing that you have is your mind' and describes his work around a 'digital mind'—a bidirectional, personalized clone of a person’s thinking and voice. He notes that AI voiceovers almost caused a post to be made from someone else’s video, illustrating the trust challenge in a world of AI‑generated content. He sketches the arc from pattern recognition to a hyper-connected future. He says, 'AI is just math. It’s pattern recognition,' and argues that the endgame is hyperintelligent AI at our fingertips: 'I think the end in mine is AI that is hyper intelligent, generating realistic videos, generating infinitely all night, improving itself.' With that premise, he frames two camps: the doomer who fears disruption and the person who sees opportunity. He urges listeners to start with the end in mind: plan for a world where AI is at work and focus on what stands out. He predicts the creator economy will rise as distribution becomes easier but differentiation grows harder, so the 8020 likely becomes 955, where the 5% reap the benefits of the 95. On practical adoption, the guest explains how ordinary people can apply AI now. AI evolved from telling a cat from a dog in 2014 to predicting emotions from tweets. He highlights education as a positive AI outcome: Bloom's two sigma shows that private tutors boost achievement by two standard deviations. Alpha School’s model uses individualized education with AI assistance and two hours of active learning daily, then curiosity-driven exploration. Education becomes an interactive, choose-your-own-adventure guided by AI toward personalized paths and continual practice. On the future of work, he lists the first AI‑driven jobs as software engineering, consulting, and any role not focused on relationships. He notes that the 8020 becomes 955 because the best can scale while branding matters. He envisions UBI as likely to prevent mass disruption, and emphasizes data ownership—'you own your data, we're not sharing with other people it can be deleted at any time.' He argues authenticity and clear founder intent will shape trust, keeping the long‑term outlook hopeful: communities, creativity, and meaningful connection endure even as AI handles routine tasks.

The OpenAI Podcast episode features Andrew Mayne interviewing Kevin Weil, head of OpenAI for Science, and Alex Lupsasca, a Vanderbilt physicist and OpenAI researcher, about how AI is accelerating scientific discovery and what may lie ahead. The guests frame a new era where frontier AI models are being deployed to assist scientists across disciplines, potentially compressing 25 years of work into five by enabling rapid iteration, broader exploration, and deeper literature synthesis. They describe the OpenAI for Science initiative as a push to put advanced models into the hands of the best scientists, accelerating progress in mathematics, physics, astronomy, biology, and more. A central idea is that progress often arrives in waves: once a capability emerges, development accelerates dramatically over months. They share vivid anecdotes, including GPT-5’s ability to help derive a physics sum by leveraging a mathematical identity—though with occasional errors that are easy to check—demonstrating both acceleration and the need for careful validation. The conversation covers several practical use cases: accelerating mathematical proofs, aiding with literature searches to discover related work across languages and fields, and helping researchers explore many avenues in parallel instead of one or two. They discuss how AI acts as a collaborative partner that can operate 24/7, helping scientists move between adjacencies and bridging gaps between highly specialized domains. The guests highlight the potential for AI to assist with experimental design and data interpretation, especially in complex areas like black hole physics, fusion, and drug discovery, while acknowledging that the frontier nature of hard problems means models can still be wrong and require iterative prompting and human judgment. They also preview a research paper outlining current capabilities of GPT-5 in science, including sections on literature search, acceleration, and new non-trivial mathematical results, with authors from OpenAI and academia. Looking forward, the speakers offer a cautious but optimistic five-year horizon: software engineering has already transformed, and science is poised for profound, iterative changes in theory, computation, and laboratory work. They emphasize that AI should complement, not replace, human scientists, expanding access to powerful tools to a broader worldwide community and potentially enabling breakthroughs across fields such as energy, cancer research, and fundamental physics. The goal is to democratize AI-enabled scientific discovery while continuing to push the edge of knowledge.