AI for the Normal Guy

Over AI for the Normal Guy

AI is already in everything. It's reshaping how we think, work, educate, heal, entertain, learn, govern, manage, connect and even love. But unlike the hype machines and doomsday prophets, Phil Ybarrolaza gives you a candid look at what's actually happening. The good, the bad, and everything in between. No sugar-coating the risks, no overselling the benefits - just an honest look at the AI revolution that's already underway like it or not.

Anthropic @ War, Accountability, No Brakes 3-8-2026

In this episode of AI for the Normal Guy, Phil, Shane, and Loren pull back the curtain on how military and intelligence agencies are already using Anthropic’s Claude and other large language models to pick real‑world targets in the new war with Iran—and how a “garbage in, carnage out” mindset may have helped lead to the bombing of a girls’ school next to an Iranian Revolutionary Guard naval base, killing over 150 children and wounding more than 100 people on day one of the strikes. We start with the basics: what does it actually mean when the Pentagon uses an “air‑gapped” military version of an AI model that looks a lot like the Claude you and I can log into? Is it really “the same model with secret training data,” or something far more opaque and unaccountable? Loren breaks down how these models are tuned on classified material and plugged into sensitive databases, while still inheriting all the statistical weirdness and hallucination risk of consumer AI. From there, we go straight into the hard question nobody in power wants to answer on‑mic: what’s an “acceptable” failure rate when an AI system is helping pick bombing targets? Shane compares how appliance companies quietly tolerate 2–3% product failures in your washing machine versus what happens when even a 1–2% targeting error means dead civilians—then asks whether someone in the chain simply decided “Claude’s got this” and rubber‑stamped a comma‑separated list of targets. Phil pushes on the contradiction at the heart of the current Anthropic–Pentagon fight: Anthropic wrote explicit bans on mass surveillance of U.S. citizens and fully autonomous weapons into its contract, and the government’s response was to threaten to label the company a “supply chain risk” and purge Claude from federal systems over six months—while simultaneously treating the tech as too valuable to give up for cutting‑edge operations. If an AI tool is “too dangerous not to control” but “too powerful to walk away from,” who actually gets to say no? Along the way, we also get into: * Why “move fast and break things” culture becomes terrifying when there are live munitions involved, not just buggy software. * How a single sloppy prompt, outdated satellite imagery, or mis‑labeled compound can turn into a classroom full of dead children—and why “human in the loop” might be a comforting phrase that hides how rushed the real review process is. * The uncomfortable math behind post‑9/11 war casualties, and whether layering AI on top of already‑bloody systems makes them more precise…or just more efficient at doing damage. * Anthropic’s “responsible player” branding versus the regulatory and market advantages of being the company that says no to the Pentagon—until the Pentagon pushes back. If you’ve ever wondered where the line really is between “smart weapons” and outsourced moral responsibility, this conversation will force you to pick a side—or at least admit that letting experimental models steer our foreign policy is not a neutral choice. Hit play to hear where we think the buck should stop when AI recommends a target and a human signs off anyway—and what happens next time someone in a conference room says, “I can do that on my laptop right now.”

8 mrt 2026 - 52 min