Those viral headlines claiming AI diagnoses better than doctors? They're based on fundamentally flawed research—and believing them could be dangerous to your health.

In this episode, I break down the alarming trend of studies from prestigious journals like Science and JAMA Network Open claiming large language models outperform physicians in diagnostic reasoning. The problem? These studies exist entirely outside the realm of actual clinical medicine.

Here's what they actually tested: They gave ChatGPT-style AI and physicians the same written notes about patients neither ever met—then compared who could guess the diagnosis better from text alone. That's not medicine. That's a reading comprehension test.

Real medicine is irreducibly multimodal. When you sit with your doctor, they're simultaneously processing your facial expressions, your tone of voice, your body language, your lab values, your imaging, and that clinical intuition built from years of experience. They notice when you claim "high pain tolerance" but wince at a light touch. They read between the lines of what you said—and what you didn't say.

None of this is accessible to a text-based language model.We also explore why "being right most of the time" isn't good enough when lives are at stake. LLMs are probabilistic tools—they predict statistically likely answers. But medicine requires considering this specific patient, not how previous patients typically presented.

Think AI facial recognition is reliable? Tell that to the Nevada man detained for 12 hours after AI flagged him as a "100% match" for someone else—despite having three forms of ID in his wallet. Or the grandmother jailed for over six months for a crime committed 1,200 miles away because generative AI flagged her as a suspect.

If we can't trust AI to reliably identify faces, why would we trust probabilistic language models with medical diagnoses?

This episode covers what every patient and medical professional needs to understand: the difference between probabilistic and deterministic AI systems, how to recognize badly trained models, why "external validation" claims deserve skepticism, and why the most dangerous outcome isn't the AI itself—it's the false narrative convincing the public that doctors are obsolete.

AI has a role in medicine—but it's task-specific tools for imaging analysis, waveform recognition, and structured data patterns. Not chatbots making diagnostic guesses about your life.

The studies got clicks. The conclusions were irresponsible. And someone needs to say it out loud.