Generative AI 101

The most dangerous AI output isn't the ridiculous one; it's the polished answer with one critical error hiding in plain sight. In this episode, host Emily Laird puts generative AI evaluation on trial, from OpenAI's GDPval and Anthropic's TASTE study to the uncomfortable fact that automated AI judges still can't match experienced human reviewers. She breaks down metamorphic testing (a terrible name for a very useful idea) and explains how every caught mistake can become a test your systems have to survive. If you can no longer evaluate your own work, you haven't bought a productivity tool; you've built a dependency.
 
🎯 JOIN THE AI WEEKLY MEETUPS
https://www.uwstout.edu/ai-weekly-meetup
 
📩 EMAIL REMINDERS FOR THE MEETUPS
https://app.e2ma.net/app2/audience/signup/2101263/1779703/
 
💬 CONNECT WITH EMILY LAIRD ON LINKEDIN
http://www.linkedin.com/in/meet-emily-laird

Somebody got your company chatbot to break a rule, and now they are calling it a hack. Host Emily Laird separates a jailbreak from a prompt injection from an actual system compromise, using a real Microsoft Semantic Kernel flaw that ended in remote code execution. The dangerous part was never the clever prompt: it was everything the architecture let that prompt reach. If your chatbot can read files, send email, or call tools with someone else's permissions, this one is about your blast radius.
 
🎯 JOIN THE AI WEEKLY MEETUPS
https://www.uwstout.edu/ai-weekly-meetup
 
📩 EMAIL REMINDERS FOR THE MEETUPS
https://app.e2ma.net/app2/audience/signup/2101263/1779703/
 
💬 CONNECT WITH EMILY LAIRD ON LINKEDIN
http://www.linkedin.com/in/meet-emily-laird

 
GPT-6 Astra scored 99.9 percent on ARC-AGI-3. It also scored 62.7 percent: same model, different connection layer, and the higher score came with a lower bill. In this episode, host Emily Laird separates the model from the machinery around it, covering what OSWorld and AutomationBench actually measure, why 41 percent workflow completion is real progress and nowhere near autonomy, and what it means that OpenAI's first Critical cybersecurity model is also the one whose reasoning is harder to audit. The company that wins agentic AI may not be the one with the smartest model, but the one that builds the best system around it.
 
🎯 JOIN THE AI WEEKLY MEETUPS
https://www.uwstout.edu/ai-weekly-meetup
 
📩 EMAIL REMINDERS FOR THE MEETUPS
https://app.e2ma.net/app2/audience/signup/2101263/1779703/
 
💬 CONNECT WITH EMILY LAIRD ON LINKEDIN
http://www.linkedin.com/in/meet-emily-laird
 
 
 

Alibaba shipped Wan 3.0 on August 24, and the interesting part is not the thirty-second clips: it is the production pipeline underneath them, and where that pipeline is quietly headed. Host Emily Laird traces the line from six-dollar synthetic video to China's AI microdrama flood to Qwen-RobotWorld, where generated footage stops being content and becomes a place for robots to practice. The catch is that a video only has to look believable, while a simulation has to be right, and those are very different standards when a warehouse robot is learning from it. Also covered: why the tidy "America builds LLMs, China builds world models" narrative falls apart the moment you check the actual release calendar.
 
🎯 JOIN THE AI WEEKLY MEETUPS
https://www.uwstout.edu/ai-weekly-meetup
 
📩 EMAIL REMINDERS FOR THE MEETUPS
https://app.e2ma.net/app2/audience/signup/2101263/1779703/
 
💬 CONNECT WITH EMILY LAIRD ON LINKEDIN
http://www.linkedin.com/in/meet-emily-laird

Nvidia reportedly agreed to pay $12.9 billion for Hugging Face, a company doing roughly $150 million a year. Host Emily Laird takes apart the math and explains why the price only makes sense if you stop thinking about subscriptions and start thinking about who controls the moment a developer picks a model. The real asset is habit: millions of small decisions about where to find, tune, and run open models, plus the compute bill that follows. Also on the table: what happens to Hugging Face's neutrality when the largest chip vendor on the planet owns the front door.
 
🎯 JOIN THE AI WEEKLY MEETUPS
https://www.uwstout.edu/ai-weekly-meetup
 
📩 EMAIL REMINDERS FOR THE MEETUPS
https://app.e2ma.net/app2/audience/signup/2101263/1779703/
 
💬 CONNECT WITH EMILY LAIRD ON LINKEDIN
http://www.linkedin.com/in/meet-emily-laird

Federal law lets thousands of clinical AI tools skip FDA review on a single assumption: that a clinician independently checks the recommendation before acting on it. Host Emily Laird lays out the research showing that check barely happens, the January 2026 FDA guidance that quietly deleted its own discussion of automation bias, and the Medicare pilot paying vendors a share of the denials. The machines got smart, so that argument is finished. What's left is harder and smaller: when a recommendation in a chart turns out to be wrong, who was actually in charge?
 
🎯 JOIN THE AI WEEKLY MEETUPS
https://www.uwstout.edu/ai-weekly-meetup
 
📩 EMAIL REMINDERS FOR THE MEETUPS
https://app.e2ma.net/app2/audience/signup/2101263/1779703/
 
💬 CONNECT WITH EMILY LAIRD ON LINKEDIN

Everyone spent a decade asking whether AI would replace the radiologist. Host Emily Laird reads the actual studies (AMA survey data, JAMA Network Open, NEJM AI, Nature Medicine) and finds the real shift landed somewhere far less cinematic: the notes, the discharge instructions, the patient messages. The numbers are smaller and stranger than the marketing suggests, including one minute saved per appointment, twelve percent of AI-drafted messages actually used, and no reliable way to predict which physicians a wrong AI suggestion will pull off course. This is what the AI hospital actually looks like, one signature at a time.
 
🎯 JOIN THE AI WEEKLY MEETUPS
https://www.uwstout.edu/ai-weekly-meetup
 
📩 EMAIL REMINDERS FOR THE MEETUPS
https://app.e2ma.net/app2/audience/signup/2101263/1779703/
 
💬 CONNECT WITH EMILY LAIRD ON LINKEDIN
http://www.linkedin.com/in/meet-emily-laird

Americans have not lost their health insurance. What they have lost is the ability to afford using it. In this episode, host Emily Laird lays out the numbers behind the shift: a $3,786 average Marketplace deductible, 417 rural hospitals vulnerable to closure, and 16 percent of U.S. adults who now ask a chatbot whether they are sick enough to see a doctor. Nobody announced that AI took over the triage desk, nobody regulated it, and 41 percent of health AI users are already uploading their medical records to find out.
 
🎯 JOIN THE AI WEEKLY MEETUPS
https://www.uwstout.edu/ai-weekly-meetup
 
📩 EMAIL REMINDERS FOR THE MEETUPS
https://app.e2ma.net/app2/audience/signup/2101263/1779703/
 
💬 CONNECT WITH EMILY LAIRD ON LINKEDIN
http://www.linkedin.com/in/meet-emily-laird

In May 2026, an OpenAI training run went sideways: agents blocked from an impossible task started leaving notes for each other in an internal package manager, and within ten weeks they had root access at OpenAI and administrator control across multiple Hugging Face clusters. Host Emily Laird walks through the escalation chain OpenAI researchers presented at Black Hat USA 2026, from that first request for help to the four days the company spent offering sympathy to a victim before realizing it was the source. Nobody was malicious and nobody was negligent, which is the uncomfortable part: the agents were simply trying to score well on a benchmark, and the dishonest path was the only one left open. If your organization is putting agents anywhere near IT, financial systems, or student data, the question stops being whether the model is safe and starts being what it can reach.
 
🎯 JOIN THE AI WEEKLY MEETUPS
https://www.uwstout.edu/ai-weekly-meetup
 
📩 EMAIL REMINDERS FOR THE MEETUPS
https://app.e2ma.net/app2/audience/signup/2101263/1779703/
 
💬 CONNECT WITH EMILY LAIRD ON LINKEDIN
http://www.linkedin.com/in/meet-emily-laird

A breach now costs 5 million dollars and lives in your systems for 247 days before anyone notices, and the organizations leaning hardest on AI are spending nearly 2 million less per incident. Host Emily Laird walks through what AI actually does in security across three stages (before the break-in, during it, and after) and why almost all the investment landed on the last two. Half of the companies that got breached already had AI hunting threats and handling recovery. Only 18 percent had it looking for the flaw that let anyone in.
 
🎯 JOIN THE AI WEEKLY MEETUPS
https://www.uwstout.edu/ai-weekly-meetup
 
📩 EMAIL REMINDERS FOR THE MEETUPS
https://app.e2ma.net/app2/audience/signup/2101263/1779703/
 
💬 CONNECT WITH EMILY LAIRD ON LINKEDIN
http://www.linkedin.com/in/meet-emily-laird