In this episode of the Professor Insight Podcast, we examine one of the most striking new studies in AI security, titled Poisoning Attacks on LLMs Require a Near-Constant Number of Poison Samples. Conducted by researchers from the UK AI Security Institute, Anthropic, the Alan Turing Institute, and the University of Oxford, this study challenges a long-standing assumption about how large language models can be compromised. The finding is as unsettling as it is important: a handful of poisoned samples can corrupt a model trained on billions of tokens.

Listeners will hear how the research team ran some of the largest pretraining poisoning experiments ever attempted, using models ranging from 600 million to 13 billion parameters. The experiments revealed that as few as 250 manipulated documents could reliably implant hidden “backdoors,” regardless of model size or dataset scale. The episode explains how these backdoors work, why they persist even through fine-tuning, and what it means for AI safety practices that rely on filtering or data scaling to defend against attack.

This episode matters because it highlights a quiet but critical shift in how we must think about AI security. If the number of poisoned examples required for an attack remains constant as models grow, then scaling up will not make systems safer. Instead, the risks expand with the data itself. For anyone working in AI development, governance, or policy, this conversation offers a grounded look at how small vulnerabilities can have large consequences, and what steps the research community is beginning to take to close that gap.

Hallucinations are a daily reality in the AI and LLM tools many of us use. In this episode of the Professor Insight Podcast, we explore new research from OpenAI and Georgia Tech titled “Why Language Models Hallucinate.” The findings shed light on why large language models often produce confident but false statements, and why this problem persists even in the most advanced systems.

Listeners will discover how hallucinations begin during pretraining, why they survive post-training, and how current benchmarks actually encourage models to guess instead of admit uncertainty. We’ll walk through real examples, the statistical roots of the issue, and the socio-technical traps created by the way we evaluate AI today. The episode also highlights the bold proposal from researchers: to redesign scoring systems so that honesty is rewarded, not punished.

This conversation matters because hallucinations aren’t just harmless quirks. They can shape trust, decision-making, and even safety in classrooms, businesses, and healthcare systems. By unpacking the causes and potential fixes, this episode offers listeners a clearer understanding of how we might steer AI toward becoming not just more capable, but more trustworthy.

Artificial intelligence is no longer just a support tool in business, it is becoming a core driver of growth and efficiency. In this episode, we explore Google’s new report, The ROI of AI 2025: How Agents Are Unlocking the Next Wave of AI-Driven Business Value. The findings reveal a major shift from asking whether to use AI to focusing on how to scale it effectively. With companies now moving into the agentic era, AI agents are stepping up to perform real work and deliver measurable impact.

Listeners will hear surprising statistics and insights from the report. For example, 88 percent of early adopters are already seeing ROI from generative AI, and more than half of executives using AI have put agents into production. The report highlights where the biggest returns are happening, from productivity gains and customer experience improvements to marketing and security. You will also hear how companies are deploying agents across different industries, what role executive sponsorship plays in success, and why data privacy and system integration remain top concerns.

This episode matters because it shows the practical reality of AI adoption, not just the theory. Businesses that move quickly are pulling ahead, and the lessons from early adopters provide a clear picture of what it takes to see real value. Whether you are in leadership, strategy, or operations, understanding how AI agents are being used today can help you make better decisions about where to invest and how to prepare for the next stage of digital transformation.

When you receive unexpected news from a company, who delivers it may matter more than the news itself. In this episode, we explore a fascinating study from the Journal of Marketing titled Bad News? Send an AI. Good News? Send a Human. The research, led by Aaron Garvey, TaeWoo Kim, and Adam Duhachek, reveals surprising insights about how consumers react to offers depending on whether they come from a human representative or an AI agent.

Listeners will hear how the empirical package of five studies tested situations ranging from concert ticket pricing to ride-sharing fees, showing that people are more willing to accept worse-than-expected offers when delivered by AI, and respond more positively to better-than-expected offers when delivered by humans. We also discuss how making AI more humanlike changes these reactions, why perceived intentions play such a powerful role, and what this means for marketing practices across industries.

This conversation matters because it sheds light on a subtle but important shift in customer relationships. As businesses adopt AI in more consumer-facing roles, understanding when to rely on AI and when to emphasize human connection can affect trust, satisfaction, and long-term engagement. The episode offers a thoughtful look at how technology and psychology intersect, giving you practical insights into communication, strategy, and ethics in an increasingly AI-driven marketplace.

Artificial intelligence is often celebrated for its breakthroughs in science, business, and daily life, but what about its hidden environmental cost? In this episode, we take a closer look at a new research paper from Google titled Measuring the Environmental Impact of Delivering AI at Google Scale. While training large models has long been seen as the main driver of energy use, the surge in everyday AI adoption means the real focus now is on inference, the act of generating responses to billions of user prompts worldwide.

Listeners will discover how Google developed a comprehensive method for tracking not only energy use but also emissions and water consumption across its AI infrastructure. The episode explores surprising findings such as how a single Gemini text prompt consumes just 0.24 watt-hours of energy and only 0.26 milliliters of water, far lower than many previous public estimates. We also highlight the difference between narrow measurements and Google’s full-system approach, as well as the efficiency gains that led to a 33 times reduction in energy and a 44 times reduction in emissions within a year.

This conversation matters because AI is no longer a niche technology, it is a daily tool for millions of people. Understanding its environmental footprint helps industry leaders, policymakers, and users see both the progress being made and the challenges that remain. By unpacking the details of Google’s study, we reveal why transparent measurement is essential if AI is to scale responsibly while minimizing impact on energy systems, climate, and water resources.

Can AI truly understand how people think, or is it just guessing based on patterns? In this episode of the Professor Insight Podcast, we explore a compelling new study that challenges the growing belief that large language models can stand in for real human participants. Titled Large Language Models Do Not Simulate Human Psychology, the paper examines how models like GPT-4 and CENTAUR handle moral decision-making scenarios and whether their responses align with actual human judgment. The findings reveal important limits that anyone relying on AI-generated insights should take seriously.

You’ll hear how researchers tested these models against real human participants by subtly changing the wording of moral scenarios and measuring the shifts in responses. While people reacted strongly to semantic differences, the models barely moved. We break down what this tells us about how LLMs process meaning, where their generalizations fall short, and why semantic nuance is still a uniquely human strength. You’ll also learn what this means for the growing use of synthetic data in research and business, and why treating AI responses as a proxy for human behavior may be more misleading than helpful.

This episode matters because it brings clarity to a topic that is gaining traction in marketing, research, and product development: using AI to simulate customer behavior. While the appeal of synthetic data is understandable, this study reminds us that human nuance cannot be fully predicted by token patterns. For leaders making data-driven decisions, understanding the limits of AI-generated insights is essential for maintaining relevance, integrity, and real-world effectiveness.

What if you could peek behind the curtain and see exactly how people are using AI at work right now? In this episode, we dive into an exclusive, unpublished study from Microsoft that does just that. Titled "Working with AI: Measuring the Occupational Implications of Generative AI," the paper analyzes 200,000 real, anonymized conversations between users and Microsoft Bing Copilot to uncover how AI is reshaping the workforce. This isn’t theoretical. This is actual, on-the-ground usage, rich with data, surprising insights, and implications for nearly every job you can imagine.

We explore what people are really asking AI to help with and what the AI is actually doing in response. From writing and research to coaching and advising, the results may surprise you, especially the fact that in 40 percent of cases, what users wanted and what AI did were completely different tasks. The study maps these interactions to job roles using the O*NET occupational database, producing an “AI applicability score” that highlights which professions are most and least exposed to AI capabilities today. Spoiler: knowledge workers, communicators, and information professionals should pay close attention.

Whether you’re a business leader, knowledge worker, or educator, this episode offers a grounded look at how generative AI is actually being used across different types of work. The findings show that AI’s current strengths lie in supporting tasks like writing, information gathering, and communication, while its direct performance is most visible in roles involving teaching, advising, or coaching. Physical and manual occupations remain less affected, for now, but even those show signs of interaction. By focusing on real-world data rather than predictions, the episode provides a more nuanced view of how AI is fitting into the workplace today.

In this episode, we explore the results of a major new global study from the University of Melbourne and KPMG titled Trust, Attitudes and Use of Artificial Intelligence: A Global Study 2025. Drawing on the views of more than 48,000 people across 47 countries, this research offers one of the most detailed snapshots to date of how AI is perceived, trusted, and used around the world. It examines differences between advanced and emerging economies, workplace adoption, student use in education, and the growing call for stronger governance.

The conversation unpacks why emerging economies are leading the way in AI trust and uptake, and why advanced economies are showing more scepticism. It highlights the gap between public confidence in AI’s technical ability and concerns about its safety and ethical use. You will hear about patterns in workplace behaviour, from productivity gains to policy breaches, and how students are using AI both to enhance learning and, in some cases, to bypass it. The episode also discusses the widespread demand for stronger AI regulation, especially to counter misinformation.

This discussion matters because it captures the reality of AI adoption beyond the headlines, showing both its opportunities and its risks. The findings reveal where trust is being built and where it is eroding, and why literacy, governance, and clear regulation are critical as adoption accelerates. Whether working in a business, leading a team, or studying in a university, understanding these trends can help in making informed decisions about how to engage with AI responsibly and effectively.