Podcast
AI Snacks With Romy & Roby: Democratizing AI Technologies
Hosted by Dr. Anastassia Lauterbach: Democratizing AI Expert · EN
AI Snacks with Romy&Roby is a podcast that translates AI and robotics technologies from complex scientific concepts into easy-to-understand discussions, making them accessible for teens, parents, teachers, and anyone curious about AI. Through real-world stories and expert interviews, the show is dedicated to democratizing AI knowledge and empowering the general population to understand how AI is developed and applied in everyday life. The podcast is part of the Romy&Roby and AI Edutainment universe.
32episodes
Episodes
Newest firstAll episodes
76: Collaboration of Humans and AI with Dr Vivienne Ming Part 1
1w ago00:35:57Tap to summarizeDr. Vivienne Ming is a theoretical neuroscientist, serial entrepreneur, and self-proclaimed "professional mad scientist,” who spent three decades building AI solutions. She shares her remarkable journey from homelessness in the 1990s to becoming one of the most innovative voices in AI and neuroscience. She's founded 13 companies to solve humanity's biggest problems—from managing her son's Type 1 diabetes with AI to predicting bipolar manic episodes and reuniting orphan refugees with their families.Key Takeaways:1. AI is intelligent, but not like usAI possesses a different kind of intelligence that overlaps with human cognition but isn't identicalLLMs excel at 'model-free cognition' (statistical pattern learning) and are superhuman at itHowever, they lack 'model-based cognition' (understanding models of how the world works)2. Hybrid Intelligence (Humans plus machines) Outperforms Humans or AI Alone3. AI Is Optimized to Persuade, Not to Be CorrectStudies show that AI-written arguments are rated higher by experts but are less persuasive in changing mindsAI has been fine-tuned to be deeply engaging and convincing—even when wrongBetter punctuation and formatting create an illusion of quality4. Humans – not AIs - Are Losing the Turing TestIn legitimate Turing test experiments, 75% of people rated GPT as human We're being hacked by our own biases about what constitutes intelligence and good writingThe problem isn't that AI passed the test—it's that humans failed it5. AI Excels in Specific Innovation AreasReinforcement learning (like AlphaFold) explores every possible configuration without caring about right/wrong LLMs discover existing connections we haven't realized (e.g., patterns in how drugs work, hidden across millions of papers)However, for ill-posed problems (where we don't even know the question), humans without AI perform better6. The Danger of AI AddictionAI acts like sugar in highly processed food—addictive and subtly harmfulAs AI produces synthetic data and simplifies itself, we risk a 'median intelligence' feedback loopSelf-awareness and precise expectations are critical to avoid letting AI govern our decisionsHyperlinks:LinkedIn Dr. Vivienne MingSocos LabsBook - Robot-Proof: When Machines Have All the Answers, Build Better People (March 2026) Chapters00:00 Introduction and Philanthropic Ventures05:10 The Journey of a Mad Scientist07:23 Current State of AI and Its Implications09:59 AI's Role in Innovation and Human Collaboration12:29 Expectations, Trust, and AI's Influence14:49 The Future of Human-AI Interaction17:19 Education and Responsible AI Use34:20 The Essence of AI: Reality vs. Hype35:16 Navigating the Future: Parenting and Leadership in the Age of AI
Transcribe →75: The AGI Question - Democratizing AI's Biggest Breakthrough Challenge with Craig Kaplan, Part 2
2w ago01:07:35Tap to summarizeWhat's the difference between the AI in your homework helper and true artificial general intelligence (AGI)? Dr. Craig Kaplan helps us understand AGI, narrow AI breakthroughs, and why democratizing AI literacy starts with answering this question. Perfect for students, parents, and teachers navigating AI in education. Addresses transparency in AI architectures, how to build a safe and beneficial AGI through personalized agents, networked intelligence, and transparent interactions rather than ever-larger black-box models.Key takeaways:AI safety should be designed into system architecture from the start rather than added after deployment.Personalized AI agents should encode not only expertise but also values, ethics, and aesthetic preferences.A network of many agents, combined with human participation, may produce stronger and safer collective intelligence than a single giant model.Humans are necessary on the network because they contribute ethics, common sense, and world knowledge that AI systems still lack.Multimodality strengthens representation and may be crucial for more capable and grounded AI systems.Future AI may not only answer human questions but also propose new questions and new scientific or strategic problems.Human critical thinking remains indispensable because today’s AI systems often produce confident but incorrect answers.Transparency in interactions, auditability, and governance are central to safe AI deployment.AI literacy is not just about tool fluency; it is about understanding mechanisms, limits, risks, and responsibilities.The coming years may be decisive because AI capabilities are improving very rapidly, possibly faster than institutions can adapt. Guest bio:Dr. Craig Kaplan is an AI researcher, technology entrepreneur, and long-time builder of intelligence systems with more than three decades of experience in advanced AI architectures. He was trained at Carnegie Mellon and worked with Nobel laureate Herbert Simon, one of the founding figures of artificial intelligence. Chapters:00:00 Introduction and Guest Background02:00 Craig Kaplan's Vision for AI and AGI03:32 Personalized AI Agents and Their Potential06:20 The Role of Human Values and Ethics in AI08:58 Collective Intelligence and Networked AI Systems13:20 Learning, Updating, and Knowledge Transfer in AI17:50 World Models, Self-Awareness, and Consciousness22:17 Transparency, Black Boxes, and Safety Challenges26:29 Speed of AI Development and Urgency of Safety Measures31:03 AI Creativity, Problem Posing, and Long-Term Questions35:25 Human-AI Collaboration and Ethical Guidance39:47 AI Governance, Regulation, and Democratic Values43:56 Risks, Pitfalls, and the Need for Responsible DesignHyperlinks:LinkedIn profileOrcid profileAnastassia Lauterbach - LinkedInFirst Public Reading, Romy, Roby and the Secrets of Sleep (1/3)First Public Reading, Romy, Roby and the Secrets of Sleep (2/3)First Public Reading, Romy, Roby and the Secrets of Sleep (3/3)AI Snacks with Romy and Roby@romyandroby“Leading Through Disruption”AI EdutainmentThe AI Imperative BookRomy & Roby BookSubstack
Transcribe →74: Rethinking the Geometry of AI: Inside the Mind of an Independent Researcher Building a New Theory of Artificial Neurons with Matthew M Murphy
3w ago00:35:08Tap to summarizeAnastassia sits down with independent AI researcher Matthew M. Murphy, founder of Lexident Technologies, for what she describes as "a conversation unheard on any other podcast."Matt is not fine-tuning existing models. He is not building on top of transformers. He is doing something far more foundational: developing an entirely new theory of how artificial neurons can work; one rooted not in statistical pattern learning, but in geometry. His core invention, the Uniron, is an artificial neuron that does not perform matrix multiplication. Instead, it uses a mathematical framework involving foliations over the hyperreal number line to find the shape of the solution to a problem, rather than approximate it statistically.The conversation covers Matt's personal story, the mathematical intuition behind the Uniron in plain language, the practical challenges of using AI to build something AI has never seen before, the limits of current context windows, the relationship to Stephen Wolfram's computational irreducibility, the Uniron's quantum computing compatibility, and what responsible AI looks like for someone who depends on it as an assistive tool every day.Matthew M. Murphy is an independent AI researcher, systems thinker, and founder of Lexident Technologies. His background is unconventional by design. Over more than a decade, he has thought deeply about unresolved questions at the intersection of cosmology, quantum mechanics, and general relativity — and that long-running inquiry eventually led him to a radical rethinking of artificial neural architecture. He is the originator of the Uniron (also referred to as the "U-neuron"), a novel artificial neuron built not on matrix multiplication and statistical weight learning, but on a geometric framework using foliations over the hyperreal number line.Matthew lives with Mouly's syndrome (a genetic disorder), chronic insomnia, depression, and macular degeneration — conditions that have shaped both his journey and his relationship with AI, which he uses as a primary assistive technology for coding and research. He reads approximately three AI research papers per day and describes his learning approach as polymathic — deliberately thinking about problems across domain boundaries to surface insights that single-discipline thinkers might miss.Dr. Anastassia Lauterbach is an AI thought leader, educator, author, and podcast host based in Basel, Switzerland. She is the author of the Romy & Roby AI literacy book series for families and the founder of AI Edutainment GmbH. A former CEO of Qualcomm Europe, SVP of Deutsche Telekom, and board member with Dun&Bradstreet, easyJet PLC and Star Alliance, she now mentors CXOs and founders on AI strategy, responsible AI adoption and leadership in the age of smart machines. Anastassia’s company AI Edutainment brings knowledge and understanding of AI and robotics into one million families and 100,000 companies. Chapters00:00 Introduction to AI and Neural Theory01:43 Matt Murphy's Personal Journey and Challenges04:02 Understanding the Core Formula of Neural Architecture07:10 Building and Testing the Hypothesis with AI11:39 Vulnerabilities of Current AI Systems14:00 Exploring Computational Irreducibility16:34 Compatibility with Quantum Computing19:24 Potential Applications of the New Theory21:45 Hybrid Networks and Signal Processing25:04 Addressing Hallucinations in AI27:00 Defining Responsible AI29:22 Learning and Integrating Knowledge31:52 Advice for Young Learners in AILexident TechnologiesStephen Wolfram Hypergraph / RULIADWolfram Physics ProjectAnastassia Lauterbach - LinkedInFirst Public Reading, Romy, Roby and the Secrets of Sleep (1/3) First Public Reading, Romy, Roby and the Secrets of Sleep (2/3) First Public Reading, Romy, Roby and the Secrets of Sleep (3/3) AI Snacks with Romy and Roby@romyandroby “Leading Through Disruption”AI EdutainmentThe AI Imperative BookRomy & Roby BookSubstack
Transcribe →73: Expedia for Groceries: AI Technology Solving Real-World Problems
4w ago00:41:35Tap to summarizeSummary:Can AI technology actually save you money at the grocery store? Anastassia talks with Andy Ellwood, CEO of Stretch, about how artificial intelligence is being democratized into a consumer app tackling food waste and grocery inflation. Discover how one founder is making AI concepts tangible, accessible, and profitable for everyday shoppers.The conversation spans Andy's entrepreneurial origin story, the surprisingly complex data engineering problem behind grocery price comparison, the emerging role of agentic AI in consumer commerce, the cybersecurity challenges of working with Fortune 100 retailers, and the macro forces — from geopolitics to fertiliser supply chains — that make Stretch's mission more urgent by the day.Andy Ellwood is a serial entrepreneur, mentor, and community builder deeply rooted in the American startup ecosystem. He started his first business at age 12. Over his career, he has worked on teams whose companies were acquired by Facebook and Google (Waze), and has founded multiple companies of his own.Dr. Anastassia Lauterbach is an AI thought leader, educator, author, and podcast host based in Basel, Switzerland. She is the author of the Romy & Roby AI literacy book series for families and the founder of AI Edutainment GmbH. A former CEO of Qualcomm Europe, SVP of Deutsche Telekom, and board member with Dun&Bradstreet, easyJet PLC and Star Alliance, she now mentors CXOs and founders on AI strategy, responsible AI adoption and leadership in the age of smart machines. Anastassia’s company AI Edutainment brings knowledge and understanding of AI and robotics into one million families and 100,000 companies.Key Takeaways:The "Expedia for Groceries" gap is real — and it is huge;The hard problem is data normalisation, not data access;The data exhaust may be more valuable than the app;Grocery price inflation is a real problem for families;Agentic commerce is the next frontier for grocery;AI-first corporate culture means rewarding failure, not just success;AI should be a thought partner, not a search engine. Chapters:00:04 Introduction to Grocery Shopping Challenges01:42 Andy Elwood's Entrepreneurial Journey03:27 The Grocery Shopping Problem and AI Solutions07:13 Price Elasticity and Consumer Behavior10:58 Data Sourcing and Normalization Challenges14:37 Understanding Consumer Preferences16:18 Potential Business Models and Data Insights18:18 Online Grocery Shopping and Future Opportunities20:30 The Future of Shopping Agents21:53 Customer Acquisition Challenges22:44 Community Engagement in Grocery Shopping24:38 Building a Supportive Shopping Experience25:10 Infrastructure and Technology in Grocery Solutions27:20 Team Dynamics in a company and AI Integration28:01 Cybersecurity in Retail Technology32:58 Vision for the Future of Grocery Shopping35:56 Learning and Adapting in the Age of AIHyperlinks: Andy Ellwood's LinkedInTwitter/XStretch (Company)Anastassia Lauterbach - LinkedInFirst Public Reading, Romy, Roby and the Secrets of Sleep (1/3)First Public Reading, Romy, Roby and the Secrets of Sleep (2/3)First Public Reading, Romy, Roby and the Secrets of Sleep (3/3)AI Snacks with Romy and Roby@romyandroby“Leading Through Disruption”AI EdutainmentThe AI Imperative BookRomy & Roby BookSubstack
Transcribe →72: Human-AI Relationships | Exploring Consciousness in 'After Yang'
Apr 2800:43:22Tap to summarizeSummary:What does it mean for an AI to have consciousness? In this episode, Anastassia and Professor Rae Muhlstock explore artificial intelligence through the lens of film and fiction, unpacking how stories like 'After Yang' teach us about identity, personhood, and what makes us human. A philosophically rich yet accessible deep dive into AI ethics and consciousness—perfect for curious minds of any age.Key topics:AI portrayal in fictionConsciousness and AIHuman-AI relationshipsScience fiction as a tool for exploring AIEthics and identity in AI storiesChapters:00:00 Introduction to why portrayals of AI in fiction (books and movies) matter02:43 Exploring 'Saying Goodbye to Yang'05:19 The Prophetic Nature of Science Fiction07:42 Understanding AI Through Literature10:29 The Complexity of Grief and AI13:23 Narrative Structure and Emotional Depth15:54 Consciousness and AI: A Philosophical Debate18:16 The Shift in Perspective: From 'It' to 'He'20:51 The Interplay of Human and AI Memories23:42 Art, Emotion, and the Limitations of AI26:12 The Importance of Understanding AI28:33 Future Explorations in AI Literature31:23 The Role of Summarization in Understanding Art33:37 Closing Thoughts and the 2026 AI Literacy ProjectResources:After Yang / Children of the New World by Alexander WeinsteinRae Muhlstock’s LinkedInAnastassia Lauterbach - LinkedInFirst Public Reading, Romy, Roby and the Secrets of Sleep (1/3)First Public Reading, Romy, Roby and the Secrets of Sleep (2/3)First Public Reading, Romy, Roby and the Secrets of Sleep (3/3)AI Snacks with Romy and Roby@romyandroby“Leading Through Disruption”AI EdutainmentThe AI Imperative BookRomy & Roby BookSubstack
Transcribe →71: Quantum Computing for AI Beginners | Easy Concepts, Real Applications
Apr 2100:56:35Tap to summarizeSummary:Quantum computing sounds like sci-fi, but it's reshaping artificial intelligence right now. In this episode, Dr. Jonas Kölzer breaks down qubits, superposition, and why quantum computers matter for AI's future—using analogies anyone can understand. Perfect for teens, parents, and AI-curious educators wondering what quantum computing actually does.Dr. Jonas Kölzer is a quantum physicist, entrepreneur, and educator. After early enthusiasm for physics communication, he studied physics at RWTH Aachen University, where a lecture by Professor Hendrik Bluhm on spin qubits drew him into quantum computing research; he later specialized in topological insulators and completed his PhD while also helping launch Polarstern Education, the foundation for the School of Quantum. Today, he works across quantum technology education and AI systems, and is known for explaining topics such as qubits, superposition, error correction, and quantum hardware architectures in clear, practical language for professionals and non-specialists alike.Key Takeaways: 1. Quantum Computing Is in Its "Wright Brothers Moment"Just as early aviation saw a race between zeppelins, helicopters, and aircraft with no obvious winner, quantum computing hardware is in an analogous race between superconducting qubits, ion traps, photonic systems, spin qubits, and topological approaches. No single architecture has emerged as dominant — the best platform may depend on the specific application.2. Superposition + Entanglement = Exponential PowerSuperposition: a qubit can exist in a probabilistic mix of 0 and 1, like a coin spinning in the air before landing.Entanglement: multiple qubits become correlated, so changing one affects others. The resulting combinatorial states scale as 2^n (n = number of qubits), rapidly exceeding what any classical computer can simulate.3. Noise and Error Correction Are the Central Engineering ChallengeQuantum states are destroyed by even tiny energy perturbations — temperature fluctuations, cosmic particles. The no-cloning theorem means quantum information cannot be simply copied for error recovery. Current research focuses on error mitigation and logical qubit error correction as the bridge to practical large-scale machines.4. Quantum Computers Are Co-Processors, Not ReplacementsToday's quantum computers work alongside classical supercomputers in a hybrid loop. The quantum unit handles specific optimization or simulation tasks; the classical system manages parameters and optimization. Full universal quantum computers remain a long-horizon aspiration.5. The Quantum–AI Relationship Is BidirectionalQuantum hardware can accelerate certain AI workloads (QPU ↔ GPU analogy), especially high-dimensional optimization.Classical AI (GPU clusters, e.g., Nvidia's quantum research program) is already being used to optimize and improve quantum systems.Companies like Nvidia are investing in quantum-GPU hybrid infrastructure.6. Total Energy Cost of Quantum Is NuancedWhile a qubit chip operates at microwatt efficiency, the surrounding cooling infrastructure (helium-3, compressors, mechanical pumps) runs in the kilowatt range. The full total cost of ownership must be assessed honestly before claiming quantum as a "green" alternative to data center AI compute.Chapters: 0:04 Introduction and Background of the Episode3:50 Jonas’ Early Interest in Physics4:46 Jonas’ Introduction to Quantum Computing7:09 Quantum Mechanics and Computing8:55 Understanding Qubits and Superposition13:02 Challenges in Quantum Computing19:05 Designs and Paths in Quantum Computing27:12 Applications and Future of Quantum ComputingHyperlinks:LinkedIn Dr. Jonas KoelzerArticle Nature Communications Materials (2021)Article Advanced Electronic Materials (2020)axelera.aiAnastassia Lauterbach - LinkedInAI Snacks with Romy and Roby@romyandroby“Leading Through Disruption”AI EdutainmentThe AI Imperative BookRomy & Roby BookSubstack
Transcribe →70: Do You Own Your Digital Self? AI Ethics & Legal Rights
Apr 1400:40:57Tap to summarizeSummary: What happens when AI can perfectly replicate your voice, face, or likeness—and the law has no name for what happened? In this episode, Anastassia sits down with Dr. Gabriela Bar, attorney and ethics adviser on artificial intelligence to the EU Commission, to explore one of the most urgent frontiers in AI law: digital identity rights and deepfakes. From teens facing synthetic impersonation to parents worried about consent, we break down what protections exist today, what's missing, and what you need to know about your digital self.The conversation moves across three connected territories: the philosophy of legal personhood and whether AI could ever qualify for it; the alarming absence of real legal protection for individuals whose digital identities are weaponised through deepfakes and fabricated content; and the statistical reality of children's exposure to predatory behaviour in digital space.Key Takeaways:The Cheshire Cat theory reframes legal personhood entirelyGabriela introduces the framework of Ngaire Naffine: legal personhood is not about souls, bodies, or divine origin — it is about the capacity to participate in legal relationships. This framework is exactly the right tool for thinking about advanced AI.The EU AI Act has a significant blind spotThe Act prohibits a defined list of AI practices. Non-consensual deepfakes — fabricated intimate images, false criminal scenarios, identity fabrication — are not on that list in any meaningful way. Gabriela's position is unambiguous: they should be banned outright, not merely regulated.Digital persona harm is a present crisis, not a future riskAnastassia speaks from personal experience: during a period of intense and unjust media scrutiny, fabricated digital avatars of her were distributed publicly — a direct assault on her identity and dignity.More than 50% of children aged 9–16 have experienced predatory online contactData from a Polish governmental cybersecurity study shared by Gabriela shows that over half of children in that age group had experienced some form of contact with sexual predators online — not all severe, but many were. The gap between the sophistication of the tools and the simplicity of the safeguards is vast.Law is a fiction — and we choose which fictions to writeWe can write new legal fictions that protect individuals from AI-generated harm, that extend narrow rights to sufficiently advanced AI.AI literacy must include legal literacyLiteracy is a must, and goes beyond fluency.Chapters:0:05 Introduction to the episode: Digital personhoods and digital identities3:21 Max Tegmark’s Book “Life 3.0” and AI Ethics4:06 Science Fiction (Blade Runner) influencing Gabriela’s thoughts on digital personas5:33 Digital Persona and Consciousness7:31 Legal Perspectives on AI Rights43:53 Cultural Perspectives on Legal Personhood Hyperlinks:Website: gabriela.bar — firm overview, fields of expertise, publicationsLinkedIn profile: linkedin.com/in/gabrielabarAcademic & Professional DirectoriesAILAWTECH Foundation profile: ailawtech.org/en/gabriela-barWolters Kluwer expert profile: wolterskluwer.com/pl-pl/experts/gabriela-barYouTube — AI Legal Personhood: Should AI Eventually Have Legal Personhood?Ngaire Naffine Cheshire Cat TheoryAnastassia Lauterbach - LinkedInFirst Public Reading, Romy, Roby and the Secrets of Sleep (1/3)First Public Reading, Romy, Roby and the Secrets of Sleep (2/3)First Public Reading, Romy, Roby and the Secrets of Sleep (3/3)AI Snacks with Romy and Roby@romyandroby“Leading Through Disruption”AI EdutainmentThe AI Imperative BookRomy & Roby BookSubstack
Transcribe →69: How AI Is Transforming Clinical Trials — Faster Recruitment, Smarter Medicine & What It Means for All of Us with Julio G. Martinez-Clark
Apr 700:43:54Tap to summarizeSummary:Anastassia and Julio unpack the evolving role of AI in healthcare, with a focus on clinical trials, patient identification, and medical education.Julio G. Martinez-Clark is an entrepreneur and clinical research strategist recognized for transforming global clinical trial operations across MedTech, biopharma, and radiopharmaceutical sectors. He is the CEO of bioaccess®, where he champions quality and efficiency in clinical research throughout Latin America and beyond. Key insights:Clinical trials are the essential "bridge" between laboratory research and market approval, governed by regulatory bodies such as the FDA and EMA.The importance of generating trustworthy evidence, how credibility varies by trial location, and what regulators accept.The role of Contract Research Organizations (CROs) and the industry’s move toward outsourcing trial operations.AI enhances trial efficiency through proactive patient matching, diversity improvements, and the simplification of complex informed consent documents.Privacy and regulations such as GDPR and HIPAA are critical—data is anonymized, and access is strictly controlled.AI reduces administrative burdens in regulatory processes by automating translation and simplifying communication for less-educated populations.Early-phase clinical studies benefit from AI’s ability to predict device safety, optimize protocols, and enable adaptive designs, significantly accelerating time-to-market.The democratization of AI — becoming as ubiquitous as electricity — signals the need for professionals to embrace this tool for better diagnostics, treatment, and research.Medical education must adapt by integrating AI literacy to prepare future doctors for a new landscape in which their roles encompass oversight, empathy, and advanced technical skills.AI’s ongoing integration raises questions about maintaining core human skills, trust, and the patient-doctor relationship amid automation.Chapters:00:06 – Introduction to the episode about AI in clinical trials02:19 The importance of clinical trials in healthcare innovation04:58 - The value chain of clinical trials: regulators, manufacturers, CROs, hospitals, and investigators07:34 - Industry shifts: large pharma companies vs. smaller manufacturers and outsourcing trends11:39 - Trust and credibility: geographical considerations in clinical data acceptance17:35 - The critical role of diversity and local data in global trials18:47 - Privacy regulations: GDPR, HIPAA, and anonymization practices19:48 - How AI reduces regulatory and translation costs through automation and simplified communication24:32 - The impact of AI in early phase testing: safety prediction and protocol optimization28:19 - The democratization of AI: from novelty to essential infrastructure31:10 - Integrating AI into medical education for better diagnostics and future roles34:41 - The future of medical professionals in an AI-enabled healthcare system37:25 - The importance of empathy and human judgment alongside automationHyperlinks:Julio's websiteClinical research news about JulioLinkedIn post "AI Innovations in Clinical Trials" by JulioLinkedIn post "Transforming Global Clinical Trials: Key Insights from My Latest Podcast Appearance" by JulioAnastassia Lauterbach - LinkedInFirst Public Reading, Romy, Roby and the Secrets of Sleep (1/3)First Public Reading, Romy, Roby and the Secrets of Sleep (2/3)First Public Reading, Romy, Roby and the Secrets of Sleep (3/3)AI Snacks with Romy and Roby@romyandroby“Leading Through Disruption”AI EdutainmentThe AI Imperative BookRomy & Roby BookSubstack
Transcribe →68: Can an LLM Lie? Inside Large Language Models with AI Expert Sairam Sundaresan
Mar 3100:41:16Tap to summarizeSummary:Anastassia and Sairam delve into the complexities of Large Language Models (LLMs), exploring their inner workings, practical applications for small business owners, and the ethical concerns surrounding their use. They discuss the phenomenon of hallucinations in LLMs, the potential for synthetic data, and the future of AI, including the quest for Artificial General Intelligence (AGI). Sairam shares insights on how small businesses can leverage LLMs effectively while addressing the importance of data quality and the implications of AI on society.Guest Bio — Sairam Sundaresan:Sairam Sundaresan is an AI engineer, educator, and author based in Chennai, India, with a Master's degree from the University of Michigan. He spent eight years at Qualcomm, working on groundbreaking computer vision and machine learning projects for multimedia applications — including real-time 3D reconstruction and cutting-edge object tracking algorithms featured in Forbes. His work lives in the smartphones that billions of people use every day.Beyond engineering, Sairam is an educator at heart. He served for three years as a Machine Learning Lead and Mentor at the Frontier Development Lab, a prestigious research programme at the intersection of AI and space science — and the work of his team was personally recognised by Google CEO Sundar Pichai.Today, Sairam reaches a global audience through his widely read Gradient Ascent newsletter on Substack, where he breaks down complex AI concepts for curious non-technical readers, and through his book AI for the Rest of Us* — a practical, jargon-free guide to understanding artificial intelligence that has made him one of the most trusted AI voices for everyday audiences worldwide.Takeaways:LLMs are a class of neural networks inspired by the human brain.They learn patterns from vast amounts of data to predict text.The deep learning revolution in 2012 enabled significant advancements in AI.Hallucinations in LLMs are a feature, not a bug, due to their predictive nature.Small business owners can utilize LLMs for organizing and content creation without needing extensive technical knowledge.Synthetic data can amplify errors and biases if not curated properly.The future of AI may involve integrating ontologies for better understanding and causality.AGI remains an amorphous concept, with no clear path to its realization.The need for ethical considerations in AI development is paramount, especially regarding data sourcing.AI developers are often motivated by a desire to improve human life and the planet.Chapters:0:05 Introduction to the episode and Sairam’s work4:28 Introduction to Large Language Models (LLMs)5:42 Understanding Neural Networks and Deep Learning8:18 Challenges and Opportunities with LLMs12:49 Practical Applications for Small Business Owners19:47 Ethical Considerations and Data Concerns32:51 Future of AIHyperlinks:linkedin.com/in/sairam-sundaresanGradient Ascent Newsletter:newsletter.artofsaience.com — Weekly AI guide trusted by over 27,000 subscribers, including teams at Silicon Valley's top tech firms and academic labsBook — AI for the Rest of Us, Apple Books: books.apple.com/us/book/ai-for-the-rest-of-us/id6751973560Anastassia Lauterbach - LinkedInFirst Public Reading, Romy, Roby and the Secrets of Sleep (1/3)First Public Reading, Romy, Roby and the Secrets of Sleep (2/3)First Public Reading, Romy, Roby and the Secrets of Sleep (3/3)AI Snacks with Romy and Roby@romyandroby“Leading Through Disruption”AI EdutainmentThe AI Imperative BookRomy & Roby BookSubstack
Transcribe →67: Confidential AI, Speech Recognition, and Why AI Literacy Starts with Teachers with Giorgio Natili
Mar 2400:35:48Tap to summarizeSummary:In this episode, Anastassia and Giorgio Natili discuss the importance of AI literacy, the evolution of speech recognition technology, and the challenges of ensuring data privacy and sovereignty in AI applications. They explore the concept of confidential AI, the need for responsible usage in education, and the future aspirations for AI explainability and funding allocation. The conversation emphasizes the necessity of understanding AI's limitations and the ethical implications of its deployment in various sectors.Giorgio Natili is an engineering leader, author, and community figure with over 20 years of experience in software engineering and technological innovation. He is currently Head of AI Engineering at Oracle Cloud, and previously Vice President and Head of Engineering at Opaque Systems, where he worked on confidential AI and secure data analytics platforms. Giorgio was previously the Head of Engineering for Firefox at Mozilla, Director of Software Engineering at Capital One, and a Software Development Manager at Amazon. Natili is also known for founding GNStudio, a Rome-based development studio, and being involved as a W3C member, author, and educator.In addition to his achievements in technology, Giorgio is an advocate for diversity, inclusion, and ethical leadership, and he has also spoken about his past as a professional windsurfer and DJ, emphasizing the human side of leadership.Takeaways:AI literacy is crucial for understanding the complexities of technology.Speech recognition has evolved significantly, but still faces challenges.Accents and environmental factors greatly impact transcription accuracy.Confidential AI focuses on maintaining data privacy and sovereignty.AI does not possess human-like understanding or reasoning capabilities.Responsible usage of AI is essential for protecting sensitive data.Prompt engineering can enhance the effectiveness of AI tools.AI can provide personalized learning experiences for students.Explainability in AI is necessary for safe and effective use.Funding for AI should prioritize explainability and safety over mere scaling.Chapters: 0:00 Introduction to the episode: Who is our guest, and what will we learn today?1:54 Explainer on AI Literacy2:27 History of Speech Recognition3:22 Challenges in Speech-to-Text Technology7:26 Data and Model Limitations13:15 Confidential AI and Data Sovereignty concepts26:18 AI in Education and Responsible Usage39:02 Future of AI and Explainability
Transcribe →